The boundless carbon cycle

• Take into account inland water processes for climate change mitigation: ponds, lakes, wetlands, streams, rivers, reservoirs.

Compartment Models with Memory

In this paper, serial theory of age-dependent compartment systems is treated. My approach seems more flexible, allowing general network structure and relaxing differentiability assumptions by basing on Riemann-Stieltjes integration.

Balancing Earth science careers in an unequal world

Unequal research experiences among Earth scientists from around the world are an obstacle to achieving sustainability. We assess challenges and propose ways to balance the careers of early- and mid-career researchers in the Global South with those in the Global North.

Relevance of methodological choices for accounting of land use change carbon fluxes

Abstract

• Accounting of carbon fluxes from land-use and land-cover change (LULCC) involves a choice among multiple options, which grossly affect estimates:
  • temporal evolution of C stocks
  • initial state of C stocks
  • temporal attribution of C fluxes
  • treatment of LULCC fluxes that occurred prior to the simulation period
• new Bookkeeping of Land Use Emissions model (BLUE)
• quantify LULCC fluxes and atribute them to land use activities and countries

Land-use change emissions based on high-resolution activity data substantially lower than previously estimated

Abstract

• LULCCs contribute one third to cumulative anthropogenic CO$_2$ emissions from 1850 to 2019
• great important, high uncertainty
• integrate new high-resolution LULCC dataset (HILDA+) into BLUE
• lower $E_{LUC}$ compared to LUH2-based estimates
• decreasing global $E_{LUC}$ trends instead of increasing
• higher spatial resolution covers pristine-remaining areas better

1. Introduction

• net CO2 flux from land-use and land-cover change ($E_{LUC}$) is key component of global C cycle
• highly uncertain for many reasons
• HILDA+: Historic Land Dynamics Assessment, 1/100 x 1/100 degrees resolution
• allows BLUE to compare $E_{LUC}$ coming from different LULCC forcings, try to identify sources of uncertainty, candidates:
  • initialization time
  • spatial resolution to investigate role of successive transition
• BLUE compuationally efficient, hence high resolution of HILDA+ can be used
• 0.001 degrees would be even better for field-scale resolution of 1 ha
• satellite data cannot be used directly because of mix of anthropogenic and environmental effects
• goal: highlight spatial and temporal uncertainties in $E_{LUC}$ related to
  • LULCC reconstructions
  • resolution of LULCC forcing
  • initialization year

2. Methodology

• includes C transfer to pools of different lifetimes
• BLUE simulations with 3 different inputs (HILDA+ at 0.25 and 0.01 degrees, LUH2 at 0.25 degrees)
• with HILDA+ 0.25 four different initialization years: 1900, 1920, 1940, 1960
• more initialization dates with model HYDE3.2 based on LUH2
• HILDA+ does not provide information and wood harvest and does not distinguish between primary and secondary land, hence preprocessing required

3. Land-use change emission based on HILDA+ and LUH2

• divergent $E_{LUC}$ trends
• gross fluxes smaller with HILDA+
• lower resolution, higher component fluxes, lower resolution leads to much more in-cell transitions
• inititialization year leads to small changes if initialized at least 60 yr earlier

4. Discussion

• alignment of results depends on regions
• disagreement of LULCC datasets since 2000 needs to be solved
• implementation of shifting cultivation needs to be revised
• spatial resolution has significant influence
• effect of “successive transitions” probably also important for DGVMs and other BKMs
• initializiation of 60 years prior to analyzed period seems sufficient

5. Conclusions

…

Tracking 21st century anthropogenic and natural carbon fluxes through model-data integration

Abstract

• assimilate observation-based time series of woody vegetation carbon densities into bookkeeping model (BKM)
• disentangle C fluxes from woody vegetation into anthropogenic and environmental contributions
• data assimilation vastly improves BKM
• global woody vegetation C sink is weaker and more suceptible to interannual variations and extrem events than estimated by other models
• model data integration important

Introduction

• the Global Carbon Project annually updates the Global Carbon Budget (GCB)
• net land-atmosphere exchange of CO2 anthroppgenic fluxes from land use and land-use induced land cover change activities (LULCC) $+$ fluxes due to environmental processes = $E_{LUC} + S_{LAND}$
• $E_{LUC}$ estimated with semi-empirical BKMs
  • $\pm65\,\%$ uncertainty (for one standard deviation)
    • different assumptions on amount of C contained per unit area (C density) contributes much to uncertainty
    • based on contemporary C stocks, leads to error simulation the past (bookkeeping error)
  • simulate emissions due to LULCC in absence of environmental fluxes by combining assumptions on amount of C contained in vegetation and soils and empirical decay functions
• $S_{LAND}$ estimated with process based dynamic global vegetation models (DGVMs)
  • $\pm19\,\%$
    • different parametrizations
    • whether and how vegetation and soil processes are captured (found from model intercomparison project TRENDY)
  • account additionally (to BKM) for environmental effects on different C pools, simulate biogeochemical processes such as photosynthesis
  • loss of additional sink capacity (LASC): larger forested areas unde rpre-industrial land cover allow more C accumulation under favorable conditions (elevated CO$_2$), DGVMs without LULCC do not correct for that
• satellite data can reduce uncertainties, but comes with limitations
  • only gross fluxes/sub-component fluxes
  • difficult to distringuish anthropogenic from environmental fluxes
  • restriction to committed fluxes: biomass loss considered flux, except when fate explicitly tracked (legacy fluxes)
• propose approach to overcome BKP and satellite limitations
  • decompose observed fluxes into $E_{LUC}$ and $S_{LAND}$
  • based on time series of global woody vegetation C densities (2000-20019)
  • time series assimilated into BKM BLUE (Bookkeeping of Land Use Emissions) model

Results

A model-data integration framework for separating anthropogenic and environmental carbon fluxes

• $C_{B,trans} = $net effects of anthropogenic and environmeltal changes based on time series mention earlier
• $C_{B,fixed} = $effects of anthropogenic changes based on Land Use Harmonization 2 (LUH2) dataset
• $S_{LAND,B} = \Delta C_{B, trans} - \Delta C_{B,fixed}$
  • should not be used to compute GCB, for instance litter, dead wood and soil dynamics are ignored; delayed fluxes happen here immediately (decomposition)
  • no bookkeeping error after 2000

Anthropogenic effects (LULCC) on global woody vegetation carbon

• transient run shows higher C stocks, probably because of effects of elevated CO$_2$
• leads also to higher emissions caused by clearing and harvesting
• such emsissions not outbalanced by higher C uptake
• major land-use transitions in BLUE: clearing, harvest, abandonment

Environmental effects on global woody vegetation carbon

• TRENDY simulations assume current CO$_2$ levels in the past and hence overestimate C biomass
• interannual variabilty mainly due to environmental effects
• $S_{LAND,B}$ shows smaller sink than TRENDY estimates
• interannual variabbility (IAV) computed as coefficient of variation
• strong positive correlation between annual mean air temperature anomaly and and annualt anomaly in forest biomass
• eastern North America (NAM): air temperature and radioation limiting effects of growth, western NAM: precipitation and soil moisture
• higher temperature also increase heterotrophic respiration, working agains ositive growth effect
• DGVMs underestimate negative effects of drought on sink strength, often heat and water stress on plant productivity not modeled
• forest degradation implicit in the data now, has strong effect

Discussion

• integration of observation into models crucial
• budget imbalance ($B_{\mathrm{IM}}$) is measure of uncertainty in estimated terms of GCB, describes difference between emissions and sinks on land, ocean, and atmosphere
• atmospheric growth rate can ($G_{ATM}$) be estimated with confidence
• natural C sinks on land and ocean uncertain
• $S_{LAND}$ as sink should increase due to C fertilization and hence $B_{\mathrm{IM}}$ should increase, but it does not, hence $S_{LAND}$ probably not accurate
• possible improvements:
  • include non-woody vegetation C
  • sub-annual time scales: capture intra-annual response to extreme events
  • different measurements should be consistent in space and time

Methods

…

My ideas

• instead of computing correlations on model output data, try causal inference computations

Carbon sequestration in soils and climate change mitigation - Definitions and pitfalls

Abstract

• Carbon sequestration is a potentially misleading buzzword
• “It’s mostly used wrongly”, which implies that they are going to propose the right way to do it here
• It’s not just about stocks
• other potential concepts: loss mitigation, negative emissions, climate change mitigation, C storage, C accrual

1 | Introduction

• IPCC (2021) + COP21 (Paris): nature-based solutions
• “SOC” comprises all organic matter in soils that is dead.”
• C sequestration (in soils) is achieved, when the net balance is positive, i.e. soil C stocks increase

• What is ultimately of importance in relation to climate change mitigation is the SOC stock change on annual, decennial or centennial timescales, and the spatial domains in which this change occurs.

• Assessing the climate change mitigation potential of additional SOC stocks requires accounting for leakage effects (Lugato et al., 2018). Leakage describes additional GHG emissions caused by climate change mitigation measures that either reduce the strength of a C sink, or turn these measures into sources of GHGs.

• What is leakage?
• The paper aims at clarifiying the terms:
  • C sequestration
  • SOC sequestration
  • climate change mitigation
  • negative emissions
  • SOC storage
  • SOC accrual

2 | Definition of C sequestration: net C uptake of CO$_2$ from the atmosphere

• Carbon sequestration is defined by the IPCC as the process of increasing the C content of a C pool other than the atmosphere (IPCC, 2001).

• C sequestration in soils: Process of transferring C from the atmosphere into the soil through plants or other organisms, which is retained as soil organic carbon resulting in a global C stock increase of the soil (based on IPCC, 2001; Olson et al., 2014)

• SOC loss mitigation: An anthropogenic intervention to reduce SOC losses compared to a business-as-usual scenario

• Negative emissions: Net removal of CO2 -equivalents of greenhouse gases from the atmosphere

• Climate change mitigation has been defined as “a human intervention to reduce emissions or enhance the sinks of greenhouse gases” (IPCC, 2021)

• SOC storage: The size of the SOC pool (i.e., SOC stock or SOC content)

• • SOC accrual:* An increase in SOC stock at a given unit of land, starting from an initial SOC stock or compared to a business-as- usual value (does not always result in climate change mitigation or C sequestration in soils)

• C sequestration in soils may not always lead to climate change mitigation, depending on past sink strength or past GHG emissions.

3 | Current use of the term C sequestration

• just some statistics of use of terminology in literature

4 | Pitfalls of using the term C sequestration in soils

4.1 | C sequestration in soils versus C loss mitigation

• Reducing C losses is no sequestration
• Just a relative SOC stock increase wrt some BAU scenario that already loses C is not enough
• The absolute C stock increase is SOC accrual
• climate mitigation is an active measure to reduce losses compared to a BAU scenario

4.2 | C sequestration in soil, SOC storage or SOC stocks?

• C sequestration is a net flux of C from the atmoshpere to soils and should have dimensions of mass/area/time
• ** Remark: That’s basically an instantaneous net flux, that’s ridiculuous. To make climate change sense out of it, you do not only need to integrate with respect to time once, but twice from here!**
• technically, “storage” means both the process of stroring something and the process of being stored
• SOC accrual does not require teh atmosphere to be the C source, C sequestration in soils, however, does

4.3 | C flux or global warming potential

• The common unit to express the effect of GHGs on the climate is CO$_2$-equivalents (CO$_2$ -eq). This converts N$_2$O and CH$_4$ emissions into equivalent units relative to the cumulative radiative forcing of CO$_2$ over a given period, usually 100 years.

5 | From C sequestration in soils to negative emissions

5.1 | Permanence of addition soil C storage

• Finally: The time period of stored C is pivotal for its climate impact.

5.2 | Leakage can prevent C sequestration in soils from achieving climate change mitigation

• Leakage occurs if a measure to enhance SOC stocks leads to an increase in GHG emissions either on site (i.e., from the soil where SOC stocks are increased) or off site.

• on-site: soil N fertilization might completely offset the additional C sequestration in terms of GHG emissions
• off-site: for instance, additional energy required, or lower yield that has to be produced somewhere else
• global scale view is required to establosh true effect of C sequestration on climate

5.3 | The temporal dimension of C sequestration: C stocks versus GHG fluxes

• A permanent measure might lead to initial C accrual, but when the new steady state is reached, now new accrual occurs
• How long does the initial effect last?
• Different time scales (30yr vs 100yr) might lead to very different conclusions in term of climate change mitigation potential

6. | Conclusions

• nature-based solutions can mitigate climate change
• cross-discipline communinication important
• can only work if terms are clearly defined

My remarks

• All in all, “mitigation” is here seen as an active measure to mitigate losses
• C sequestration (removing C from the atmosphere) is the holy grail
• It’s ridiculuous for this holy grail to ignore the time component and consider instantaneous net fluxes instead. The holy grail should twice be integreated over time.
• However, they do not see C sequestration as the holy grail for it ignores other GHGs. Climate change mitigation potential includes them.
• For climate change mitigation, GPW is important.

Modified source-sink dynamics govern resource exchange in ectomycorrhizal symbiosis

Summary

Symbiosis between tree roots and fungi as a trade explains mutualism development on evolutionary timescale but not experimental results. Instead think of it as source-sink dynamics.

I. Introduction

• sometimes C invest and N return are uncorrelated or even negatively correlated
• reward schemes depend on the scale

II. Source-sink dynamics remain a useful model for resource movement

• source-sink dynamics: flow from high concentration to low concentration
• gradient and conductivity govern speed
• fungal biomass drives C sink strength which can lead to positive feedback
• fungi can monompolize N if it’s available at high level, induce artificial N scarcity
• nutrients and water also move according to source-sink dynamics
• plant, fungi, and environment change rates making thinkgs more complex

III. Beyond source-sink: modifications to resource movement

• plant and fungi have different interests, modify simple source-sink dynamics
• plant defense might change resource transfer
• C transfer sometimes mostly explained by expression of genes involved in defense and stress response
• high variation of fungi and immune responses lead to a variety of altered source-sink dynamics
• other functional diversities have also influence
• molecular tools not yet completely understood
• fungal competition also a factor in favor of the plant

IV. Conclusions

• source-sink dynamics a good first approximation
• altered by a lot of factors

Box 1 - Future research directions

1. How does strength of resource gradient affect dynamics?
   • Physics?
2. Extent of fungal diversity improving N access of the plant
   • Plant benefits from fungal competition
3. At what level of resource do fungi stop helping?
   • Tragedy of the commons?
4. Plant sensitivity to shift from helpful to unhelpful fungus
   • Makes only sense if there are more fungi or fungus-independent nutrient uptake pathway.
5. Spatial scale on which plant can discriminate between fungal partners. How?
   • Plays a role for C allocation to different fungi.
6. Can we predict plant-fungal compatibility from genomic data?
   • How fast are those genomic gata changing? Does it make sense to identify genes if they change too fast?
7. Differences in resource transfers with adult trees and seedlings.
   • Also adult fungi and youg ones? Have different levels of need?

My remarks

• no citation of “The mycorrhizal tragedy of the commons”
• So what do we learn? That it’s complex, lots of interests, strategies, external factors. Thanks.
• We probably learn about the future research directions. This thought is supported by the relatively new references.

The mycorrhizal tragedy of the commons

Abstract

• trees trade C for N with mycorrhizal network
• this network is the “commons”, part of the tragedy
• trees gain additional N et the expense of neighbours by supplying more C to fungi
• this can lead to increased N immobilization (in fungal biomass)

Introduction

• ectomycorrhizal fungi (EMF) hold N back from trees when they are in need
• when N supply is amended by fertilization, also EMF give away greater proportion
• further supply of EMF with C makes them diminish N returns
• fungus competes with other EMF symbionts of the same plant, can gain more C by exporting more N, until personal N matches its received C
• receiving more C makes EMF use more N for itself
• enhanced EMF growth initially increases N uptake, but eventually N immobilization in fungal biomass; negative feedback on plant’s N uptake

• How can this negative feedback made it through evolution?

• dual scale of ectomycorrhizal symbiosis: one tree, many fungi and many trees, one fungus
• trees have optimize individual benefit, this has consequences for other trees connected through the fungal network
• combined efforts of trees to increase individual N gain aggravates N immobilization in fungal biomass

Hypothesis

• fungus shares greater N proportion with trees that deliver more C
• tree delivers more C to fungi that supply more N
• tragedy from tree viewpoint: total C export to all fungi so high that it leads to N immobilization
• tragedy from fungus viewpoint: exporting more N would reduce own growth, exporting less N reducec competitiveness for plant C
• plants share common N pool, optimizing for individual gain hampers comunity gain
• experiments with plant strangling and shading

Materials and methods

Seedling experiment

Field experiment

Model description

Results

• shaded less biomass than sun, strangled no effect
• shaded plants more N than sun plants, strangled plants way less: decreased C export to fungi mobilizes soil N to the plant and vice versa
• maximum network-scale N mobilization occurs at intermediate level of network-scale C export
• at individual tree level: 1 more C supply leads to 0.95 more N return, hence there is competition among trees which will eventually make network-scale N mobilization drop

Discussion

• tragedy of the commons: individual gain reduces community gain
• there is one common N pool all trees draw from (common EMF network, in which multiple fungi connect the host plants)
• elevated atmospheric CO$_2$ levels should lead to higher N immobilization, more fungal biomass growth, maybe less tree growth

The carbon costs of global wood harvests

Abstract

• counting newly grown wood from harvested forests as extra is wrong because this growth would also happen without harvesting
• harvesting forests leads until 2050 to additional eCO$_2$ in the realms of land use change due to agricultural expansion

Main text

• depending on how you count, harvesting wood might not be just C neutral but even benefit the climate
• reporting at national level allows countries to look C neutral
• harvesting should decrease this benefit, but is not reported separately
• net increase in forest C makes countries appear to have no emissions
• reporting net effects of new wood harvests and regrowth from previous harvests leads to similar effects (no identification of effects of new wood harvests alone)
• appears that harvesting in temperate countries is beneficial and in tropical countries it is costly (in terms of climate)
• wrong accounting: Growth that would occur anyway cannot offset harvesting costs.
• so far: mostly spatial offsetting, or offsetting with past growth
• possible: offsetting with what grows in the same place after the harvest
• harvests lead to short-term emissions and undermine Paris Agreement

Accounting for time in estimation of GHG costs

• CHARM (carbon harvest model): new global forest C model
  • live vegetation, roots, slash, different wood products and landfills
• computes annual difference in scenarios: forest would have grown, forest is harvested
• emissions discounted with 4% per year: the earlier the heavier
• SCC (social carbon costs): for instance cost for mitigation might decreaste with better technology
• mitigation now is more valuable (costly), already included in the discount
• discount can be seen as interest rate for companies on today’s emissions, they will have to pay back more later

Growing wood demand

• long-lived wood products (LLP): sawn wood, wood panels, and other industrial roundwood
• short-lived wood products (SLP): paper and papaerboard products
• very-short-lived products-wood fuel (VSLP-WFL): wood harvedted deliberately for energy
• very-short-lived products-industrial (VSLP-IND): waste from manufacture of other wood products, burned for energy
• wood harvests will increase by 54% globally between 2010 nd 2050, mostly SLP; VSLP-WFL most uncertain
• substituton effects for steal and concrete
• “clear-cut-equivalents”: area necessary by clearcutting to get the same amount of wood
• nice flow-chart of wood products

Robustness of results

• 3-5 GtCO$_2$e per year
• probably conservative: effects of harvests on soil C not counted: meta-analyses show quite some soil C loss
• indirect effects such as road building ignored (can be several times direct effects)

Insensivity to discount rate

• robust if society has a small preference for short-term over long-term mitigation

Meaning of economic effects

• Compare harvesting with no human activity instead of another human activity such as land use change, otherwise you are covolving numbers.

A potential mitigation option

• harvesting comes with carbon costs, offsetting is usually done in wrong ways
• comparing them with land-use change due to agricultural expansion makes costs disappear, because costs are about equal
• No harvesting mitigates climate change.
• Later, mature forests might sink in C slower, but no harvesting now would buy us time.

Acorn review: The persistent mystery of declining growth in older forests

Abstract

Forests are very diverse all over the world, but they all share the decline in diameter growth rather after an early plateau (moment of full canopy development). Growth of trees and forests may be determined by other factors than carbon supply.

1. Introduction

It’s not the tall trees that slow down in diameter growth but the medium to small trees. They really decline in grwoth rate which reduces the forest’s growth rate. This can be caused at three levels:

• resource use
• efficiency
• partitioning

2. Resource use

• decrease in LAI:
  • due to collisions of branches, nutrient supply, hydraulic constraints on moving water up
  • LAI reduction by 30% leads to light interception reduction of 10-20% (less self-shading)
  • not enough to explain decreasing stem growth
• water flow in streams draining old forests smaller than in young forests: declining water use cannot explain declining growth
• not many studies on age vs nutrient supply
  • big trees could limit nutrient availability of smaller trees
  • fertilization increases growth
  • but: still the same pattern of decreasing growth

3. Efficiency: Declines in growth per unit of resource use

• efficiency seems more site-dependent than age-dependent

4. Partitioning

• tempting assumption: older forests have more tissue and respire more, less C available for growth
• but: most tissue is dead, half of respiration comes from growth, studies show lower respiration in older forests
• below-ground C flux usually declines in concert with stemwood growth; no universal pattern though

5. Beyond Carbon: Fitness may be about more than photosynthesis and C budgets

• seemingly: declining PH from declining resource use efficiency, then different partitioning
• Is this focus on C sufficient?
• growth often out of synchrony with C assimilation
• growth not always limited by rate of C uptake
  • trees might not maximize C uptake, might partition elsewhere to dispose of excess C (see Prescott et al., 2021)
  • REDUNDANCY???
• once hierarchy of trees is established, trees arrange with it
• older trees use resources less efficient, why?

6. Three key points for the future

• more data needed for aging forests, managed and unmanaged
• do trees sustain their established social status?

Understanding the roles of nonstructural carbohydrates in forest trees – from what we can measure to what we want to know

Abstract

New isotopic tools allow direct quantification of timescales involved in NSC dynamics, and show that NSC-C fixed years to decades previously is used to support tree functions.

I. Introduction

Although these approaches may be much more insightful for tree C relationships than concentration or pool size measurements (Ryan, 2011), they still infer fluxes from changes in NSC concentration over time and may require the assumption of steady state on longer timescales.

One major theme we emphasize throughout this review is that measured concentrations of NSC in tree organs cannot be interpreted explicitly in terms of ‘storage’ functions, but that storage is the result of asynchronies in the supply and demand for C currency compounds that occur across a range of timescales and, in part, reflect distances between source and sink tissues.

Especially as the combinations of stresses experienced by trees are altered in the context of climate change and enhanced atmospheric CO2 (Niinemets, 2010; Trumbore et al., 2015a), questions about the central role of NSC are critical for predicting resilience of trees that differ in life stage (ontogeny) and life strategies (phylogeny). In this paper, we give particular emphasis to the regulation of NSC storage in trees.

In forests, NSC pools are large enough for their changes to be important in annual stand-level C balance (Richardson et al., 2013) and given the central role of NSC in plant functioning, it is not surprising that many vegetation models are based on the analogy of mobile C as currency or cash flow (McDowell et al., 2011; Klein & Hoch, 2015).

• try several models and check NSC age?

II. NSC in plant function: synthesis, classes, roles and responses to drought

Plants lack enzymes for cellulose degradation (Pallardy, 2008), so C in these compounds is not available to the plant for future use.

Amylose and amylopectin are the constituents of the most common storage carbohydrate, starch, and have the advantage over other carbohydrates in being osmotically inactive, allowing plants to accumulate them in large quantities. Besides NSC, lipids are important storage compounds and are used by plants as substrates for respiration but also for plant defence and communication.

1. Photosynthesis and metabolism

During photosynthesis plants transpire hundreds of molecules of water for each molecule of CO2 they assimilate (Taiz & Zeiger, 2002). Most of this water is transpired and only a small fraction is used to provide electrons for the light-dependent reactions of photosynthesis. Vascular plants control water loss from transpiration via regulation of stomatal aperture when soil water availability is reduced or when water vapour deficit is high. Regulation in response to declining soil water availability is triggered by root signalling via phytohormones causing stomata to close (Brodribb & McAdam, 2011); however, stomatal closure also reduces CO2 diffusion into the leaf.

• for Umeå

2. Defence

Although it is commonly believed that plants under stress are more vulnerable to additional disturbances (Manion, 1991), stresses such as drought usually result in increased concentrations of secondary metabolites (Gershenzon, 1984; Mattson & Haack, 1987) which likely reflects the accumulation of NSC from decreased growth sink activity (Herms & Mattson, 1992). However, accumulation of NSC may only occur during early phases of drought, and declining NSC availability during longer droughts may cause a reduction of allocation to defence compounds (Steele et al., 1995).

6. Storage

Storage in plants has been defined as ‘resources that build up in the plant and can be mobilized in the future to support biosynthesis’ (Chapin et al., 1990, p. 424) so as to buffer any asynchrony of supply and demand which may occur on diel, seasonal or decadal (or longer) temporal scales and across plant organs.

In their early work on plant storage economy, Chapin et al. (1990) differentiated three distinct processes: (1) accumulation (build-up of resources when supply exceeds demand); (2) reserve formation (metabolically regulated synthesis of storage compounds, competing with other sinks like growth and defence); and (3) recycling (reutilization of compounds involved in growth or defence during later metabolization). This definition includes, therefore, both an overflow process (accumulation) and an actively regulated component of storage (reserve formation). Interestingly, even some 30 yr after this seminal work, there is still discussion about whether C storage may be either ‘passive’ (sensu accumulation) or ‘active’ (sensu reserve formation) or both (Sala et al., 2012; Wiley & Helliker, 2012).

III. Tools and approaches for quantifying NSC dynamics

Such problems severely limit the usefulness of any metric derived from concentrations measurements for mass-balance approaches for whole-plant C balance estimation, for example, by comparison with measures of net photosynthesis and respiration fluxes (Hartmann et al., 2015b). Direct assessments of whole-plant C balance are already challenging for smaller individuals (Zhao et al., 2013) but are exceedingly difficult in mature trees (Ryan, 2011) and hence new approaches are needed.

IV. What is the spatial and temporal distribution of NSC in trees?

Sugars produced in the leaf are converted to starch and stored in the chloroplast during daytime (i.e. when supply > demand) to be remobilized and used for growth during the night (Geiger et al., 2000).

Stems and coarse roots comprise the main woody volume of trees. Thus, even though their NSC concentrations are relatively low, they account for most of a tree’s NSC-stock.

NSC concentrations usually decrease from the outer towards the inner sapwood zone in stems but remain constant from the sapwood–heartwood transition into the heartwood (Hoch et al., 2003).

• Fig. 6 shows C ages in different trees (stem and root effluxes, NSC, resprouts)

V. Studies on the use of NSC in plant functioning –

progress towards answering longstanding questions

Similarly, tracking how the age of both respired C and NSC change in various organs during tree death can shed light on whether residual NSC might be inaccessible for metabolism or simply not used in processes when trees approach mortality.

Fischer et al. (2015) used such measures of RQ and 13C in respired CO2 to show that trees under C limitation (from shading) switched from progressively declining carbohydrates to stored lipids to fuel respiration but this was not observed during drought.

Allocation to storage was maintained even though the NSC pool size decreased (Hartmann et al., 2015b).

Corroborative evidence from radiocarbon analysis of springtime ascending xylem sap in sugar maple (Acer saccharum) indicates that the sugars mobilized to fuel leaf-out have been assimilated during several preceding growing seasons, likely by regular allocation of NSC to a well-mixed 3–5 yr ‘deep’ functional storage pool (Muhr et al., 2016). (Compare with E age.)

The storage pool size may decrease but sustained allocation can be traced with isotopic markers added to the atmosphere (Hartmann et al., 2015b) and should be accompanied by distinct transcript patterns of genes encoding for storage processes, like starch or lipid synthesis (Koch, 1996) and associated enzymatic activities. (Should be seen as decreasing $C_S$ age during starvation.)

6. What is the role of symbionts in plant C allocation strategies?

Radiocarbon measurements of mycorrhizal fruiting bodies and parasitic plants indicate that C being transferred is mostly fixed within the last year (Gaudinski et al., 2009), which would seem to indicate that this is a high priority.

Interestingly, nutrient uptake did not decrease even though shaded or low-CO2 plants decreased the absolute amount of C transferred to mycorrhiza. Instead, plants optimized internal resource distribution by allocating roportionally more C and N to aboveground tissues to maximize the potential for CO2 assimilation (Zhang et al., 2015).

A similar resource limitation experiment applied atmospheric N2 removal to force rhizobia to ‘cheat’ on their hosts, thereby addressing plant sanctions for nonrewarding symbionts (Kiers et al., 2003).

Dynamics of non-structural carbohydrates in terrestrial plants: a global synthesis

Abstract

• high NSC sign for oversupply or necessary for survivial or inaccessible?
• NSC about 10% dry plant biomass, higesht in leaves, lowest in stem
• strong depletion during growing season
• high in conifers with relatively high seasonal minimum
• starch as future reservoir, soluble sugars for immediate functions (osmoregulation)

Introduction

• non-structural carbon compounds (NCC)
  • NSC: starch, soluble sugars, fructants (in some herbs)
  • neutral lipids (in some taxa)
• role as storage: support night metabolism, support in stressful periods
• starch might be depleted, soluble sugars need to be above a certain threshold for everyday functions like phloem transport, turgor stability, …
• size of soluble sugars threshold unknown
• NSC lower in stems because of more lignins and non-living tissue

Methods

Results

• NSC(leaves) > NSC(belowground) > NSC(stem)
• NSC varies with plant functional type
• seasonal NSC oscillations
• minimum NSC between 30% and 50% of seasonal maximum

Discussion

• imbalance between photosynthesis and C demands not only reason for NSC seasonal patterns
• soluble sugars seldom depleted suggesting that it serves important immediate physiological functions
• neutral lipids (not measured) can comprise almost half of total NSC in some species

Conclusion

• dual function of NSC as storage (starch) and performing immediate physiological functions (soluble sugars)
• plants exhibit relatively high NSC thresholds, mortality should occur when NSC falls below these thresholds, even in non-stressed plants
• dual function with soluble sugar to starch conversion would have to be explicitly modeled

My remarks

• in ACGCA NSC to biomass ratio is constant for fine roots and leaves and cannot be used for regrowth, so we assume it is needed for metabolism somehow
• in stem, NSC lever is supposed to be lower, nevertheless, some threshold should not be undershot- let’s say half of the leaves-fine root average is the normal stem NSC average, then say half of it is needed as soluble sugars: \begin{equation} \nonumber \min\delta_S = \frac{1}{4}\,\frac{\delta_L+\rho_{RL}\,\delta_R}{1+\rho_{RL}} \approx 0.02 \end{equation}

Demonstrating and Evaluating Teaching Proficiency

SLU guidelines

• provide support to demonstrate teaching skills or assess them

1. Teaching proficiency - a short background

• proficiency: ability to perform instruction and carry out examination
• late 20th century: paradigm shoft from “providing instruction” to “producing learning”
• teaching proficiency: proficiency in planning, performing and evaluating teaching, including ability to motivate teaching methods and refklect on them
• teaching skills requirement for professor or lecturer, equal weight to scientific skills
• good practice in education:
  1. Encourages contacts between students and faculty
  2. Develops reciprocity and cooperation among students
  3. Uses active learning techniques
  4. Gives prompt feedback
  5. Emphasizes time on task
  6. Communicates high expectations
  7. Respects diverse talents and ways of learning
• 4 dimensions: Context –> Knowledge and approaches – Planning –> Teaching
• 3 progressive levels of early stage teachers’ views:
  1. Who the student is: learning is primarily defined by the students’ personal characteristics: some ar gifted and some are not
  2. What the teacher does: learning is primarily defined by what the teacher does: how she teaches
  3. What the student does: learning is a result of the learning activities the students engage in, and it is determined by the students’ previous experiences and the learning environment they find themselves in
• also 3 progressive levels for proficient teachers:
  1. excellent: proficient in the classroom, teaching is experienced as effective by colleagues and students
  2. expert:
     • excellent plus theoretical knowledge about teaching and learning
     • combine subject knowlege and practical knowledge of teaching
     • continuous self-development and adaptation in teaching methods
  3. scholarship of teaching and learning: excellent + expert + sharing and promoting knowledge on the general development of teaching: conferences, articles, etc.

2. Documenting teaching skills - the teaching portfolio

• two types of teaching portfolios:
  • personal:
    • extensive documentation together with reflections
    • course evaluations, testimonials from course leaders, material produced, certificates, etc.
    • all kind of situations you helped people, being a coach, private support
    • include negative experiences
    • everything is valuable
  • specific: assembled for special purpose
  • What have I done? How did I do it? Why did I do it this way? What was the result?

What serves as evidence of teaching proficiency?

• pedagogical self-reflection:
  • What I teach
  • How I teach
  • Why I teach the way I do
  • What results do I achieve
• teaching philosophy (pedagogical standpoint):
  • what is knowledge
  • existing limits for what is possible to know
  • how learning happens
  • what is important to know
  • how should we teach
  • should match your actual way of teaching

• A well-structured philosophy can be formulated citing the literature.

• most important: philosophy in action
• self-reflection:
  • penetrate all aspects of a teacher’s role
  • trace own development as a teacher over time
  • include merits
    • for example self-development during participation in a teaching course
      • Constructive alignment
      • important is what the students do, not what the teacher does
      • deep knowledge needs personal activity
      • active learning way more useful than just bein able to preach
      • rather teach less but deeper
    • mere participation does not demonstrate teaching skills
    • desciption of significance of the course for own development
    • expansion of vision as a teacher, influence on student learning
  • ideas for future development (personal, at the department)

    3. Requesting or writing a testimonial of teaching skills

    To be done.

4. Evaluating teaching proficiencey

To be done.

My questions and comments

• In the seven principles for good teaching (Table 1), what does “Emphasizes time on task” mean?
• My teaching was long ago, I did not care about certificates, course evaluations, or testimonials for in Germany nobody cares about. How does this impede future applications in Sweden?

Biodiversity as insurance: from concept to measurement and application

Abstract

• biodiversity will make aggreagate ecosystem properties vary less
• insurance and portfolio theory connected to biodiversity
• distinction between effects on mean and variability
• application to ecosystem management

I. Introduction

• decrease variation (buffering), increase mean (performance enhancing)
• often happen at the same time

II. Insurance and portfolio theory in economics

• trade-off between decreasing variance and increasing mean
  • portfolios: different stocks
  • options: permission to delay a decission until more information is available
  • insurance: an insurer is paid to cover the highest risk scenarios

(1) Portfolios

• maximize expected returns by a given level of risk by choosing not highly positively correlated assets
• mean-variance trade-offs not yet applied in ecosystem functioning

(2) Options

• option value determined by Black and Scholes
• increased return if ill-suited species are lost, cynical toward biodiversity and stability theory

(3) Insurance

• paying the insurance premium lowers wealth but saves from disaster
• ecology: conservation strategies?

III. Insurance and portfolio theories in ecology

(1) Similarities and differences between insurance and portfolio theories in ecology

• a zero-correlation assumption is ridiculous, apart from being separeted in space maybe
• the shared ecological driver (climate) by all species will have an increasingly negative impact which makes portfolio and insurance theory less likely to increase mean returns, no matter how it’s done; planting species more suitable for a new climate might be useless for trees that grow to slowly compared to current climatic changes

(2) Mechanisms of biological insurance

• important: asynchronous fluctuations

• while competition does contribute to increase the level of asynchrony of population fluctuations, which has a stabilising effect on ecosystem properties, it simultaneously increases the amplitude of population fluctuations, which has a destabilising effect

• one should expect reduction of competition, i.e. niche complementarity, not competition, to favour ecosystem stability

(3) Distinguishing between the effects of biodiversity on the mean and variability of ecosystem properties

• in ecology often trade-off between mean and variablity not yet consindered

• Therefore, for clarity’s sake, we propose that the performance-enhancing effect be renamed a selection effect

  • Who selects better than nature itself?

• It may be worth recalling here that the selection effect does not conflict with the positive effects of biodiversity. Not only does it require the maintenance of biodiversity at larger spatial and temporal scales (Loreau, 2000), it even turns into functional complementarity when considered at larger scales because selection of the best-performing species under each environmental condition tends to increase the average level of ecosystem properties across space or time

  • Large scale spatial diversity is not the same as small scale spatial diversity for, let’s say small animals living in such ecosystems, who may depend on different species and cannot just wander around between different ecosystems on a daily basis

(4) Spatial insurance

• large spatial scales, but low diversity on small scales

IV. Applications of biological insurance in ecosystem management

(1) Agriculture

• local diversity for single farmers as biological insurance instead of economical insurance against extreme events
• on country-wide scale: spatial diversity/insurance

(2) Fisheries

• high species diversity allows simpler adaptation to market requests

(3) Forestry

• under current future climate, mixed-species stands seem inevitable
• portfolio theory over the last 20 years shows that mixed stands are both more stable and more ecomicially useful
• pure poductivity without looking at the economic aspect does not make much sense

• Ecological knowledge on the mechanisms that provide biological insurance has yet to be integrated into applications of biological insurance and portfolio theory to forest management. For example, García-Robredo (2018) recently demonstrated that reduced competition and facilitation between mixtures of two tree species can lead to overyielding, increased economic return and reduced financial risk. Most studies so far, however, have focused on demonstrating the positive effects of managing different types of stands (often monospecific) and have ignored complementarity effects within stands. Moreover, portfolio studies have not considered the variability in site conditions across the managed forest landscape and have disregarded spatial ecological insurance.

V. Synthesis

(1) Shared features across disciplines

• ecology: fixed species value, variable abundance
• economy: changing stock value, fixed abundance (on first investment)

(2) Contrasts between ecological and economic concepts

• economics:
  • no spatial insurance effect for a single asset
  • automatic mean-stability trade-off
• ecology:
  • spatial insurance also for single species
  • increase of performance and decrease of risk possible by complementary species effects

VI. Future challenges

(1) Linking the effects of biodiversity on ecosystem functioning and stability

(2) Incorporating multiple functions and feedbacks

• Future developments of biological insurance theory may require a clearer formulation and justification of the assumed objective function.

(3) Developing new approaches to partition biodiversity effects across scales

(4) Extending biological insurance theory to complex interaction networks

VII. Conclusions

• biological insurance theory becoming mature
• need to distinct between effects on mean and variablity of ecosystem properties
• much more to do in this growing research field

Personal questions and remarks

• What corresponds to “the market” and its fluctuation?
• Scale (time oe space) must be large enough, a stand is probably not enough.
• One should put infinite costs on the extinction of a species (policy-wise).
• Portfolios: many pure spruce stands yield high productivity, but are susceptible to bark beetle outbreaks: mix the stands with different species
• Option pricing in ecology seems cynical to me. Wait a while to learn whether or how much an intact ecosystem if beneficial before we decide to gradually destroy it opens Pandora’s box, in particular at incomplete information of species’ values and their interactions. Insurance: store seedlings and genes in huge protected databases in case a species goes extinct
• an insurance strategy: reduce plant density to avoid spread of infections
• It seems to me that when they mention an example of a quantity to be stabilized in the first place, they speak about stock (productivity). It’s not about species conservation or bioderversity in itself. Biodiversity is only considered as means to help ensure productivity (biomass, yield, salmon catches).
• Best-performing species might deplete soil nutrients or other resources that are simply not part of the equations here.

Multidimensional tropical forest recovery

Tropical forests disappear rapidly because of deforestation, yet they have the potential to regrow naturally on abandoned lands. We analyze how 12 forest attributes recover during secondary succession and how their recovery is interrelated using 77 sites across the tropics.

Network analysis shows three independent clusters of attribute recovery, related to structure, species diversity, and species composition. Secondary forests should be embraced as a low-cost, natural solution for ecosystem restoration, climate change mitigation, and biodiversity conservation.

• tropical forest regrowth on many agricultural sites left because of loss of fertitlity, migration, or alternative livelihhod options
• resilience: ability to resist disturbance and to recover from disturbance
  • resistance: difference between the value of a specific forest attribute at the start of succession and the average old-growth forest value
  • recovery: ability to return to old-grown forest attribute values after succession
• succession: change in vegetation structure, species composition and ecosystem functioning over time after a disturbance
• secondary succession: on previously vegetated lands when a disturbance removes most of the above-ground vegetation and can proceed at fast rates due to legacy effects of previous forest or previous land use

Pace of recovery

• soil resistance was high, indicating that agricultural use does not disturb it too much; fast recovery

Wood density (WD) is the stem-wood dry mass divided by stem volume, and it increases tissue longevity and carbon residence time in trees and forests. Specific leaf area (SLA) is the leaf area divided by the leaf mass. It reflects leaf display cost and scales positively with photosynthetic capacity and forest productivity and negatively with leaf longevity. WD and SLA change during secondary succession because pioneer species are typically replaced by later-successional species with opposite trait values.

Nitrogen fixation is generally high early in succession when irradiance is high and trees can support their nitrogen-fixing symbionts with carbohydrates and declines over time as forests regrow, light availability in the stand drops, and nitrogen fixation becomes too costly.

• fast plant functional recovery after short-lived pioneer species, with similar traits

Structural heterogeneity (SH) refers to the tree size variation in a plot; it increases light capture and ecosystem productivity and contributes to biodiversity conservation by providing a habitat for different species.

• forest structure recovers at intermediate pace
• species diversity and composition recovers at intermediate to slow pace

Network properties and proxies for multidimensional recovery

The first network analysis was based on pairwise correlations among all 12 attributes and showed that recovery of attributes occurred in parallel, with the highest expected influence (i.e., many links with other attributes) for SC, followed by the three structural attributes and soil C.

The clustering of forest attributes into multiple groups suggests that recovery of different forest attributes is shaped by different drivers or processes. For example, recovery of biodimeversity attributes may be driven by the land-scape context, land-use history, and the availability of seed trees and dispersal vectors, whereas recovery of structural attributes may be driven by resource availability [i.e., water availability, soil fertility (16), and remnant trees].

We hypothesized that AGB would be the best predictor of multidimensional recovery because ecosystem processes and flux rates strongly depend upon the amount of vegetation. Instead, we found that recovery of Dmax had the highest influence.

Resilience

• overall return time about 120 years
• fast forest recovery during secondary succession because of many legacies and productive, warm, and wet conditions

Applied implications

we urge the embrace of SFs as a low-cost, nature-based solution to meet the United Nations’ Sustainable Development goals and the United Nations’ Decade on Ecosystem Restoration goals (where needed with some assistance from management: control of invasive species, seed availability)

Tips from neuroscience to keep you focused on hard tasks

Understanding cognitive control can help your working life, says David Badre.

Hard tasks

• solve a methodological hurdle
• design an elegant experiment
• make sense of a puzzling result
• work on a new model
• write a paper or grant proposal

Make space

In practice, returning to a hard task in this way comes with a ‘restart’ cost.

Switching frequently between tasks makes producing quality work harder.

Be consistent. We should try to reserve a consistent time and place for our hard work and be protective of it.

Minimize distraction and never multitask

Even cues that we simply associate with other tasks, such as seeing our phones on the table, can distract us. As much as possible, we should keep our space and time for hard work clear of other distracting tasks.

Beware the allure of easy tasks.

Engage in good problem-solving habits

In general, we can get better at structuring hard problems with experience.

Interact with others. Just like taking a break, interacting with others can help us conceptualize a problem in new ways. Talking to people with diverse backgrounds, perspectives and viewpoints that differ from our own can be a powerful way to break out of a rut and make progress, as well as get some perspective.

Personal comments

• don’t get overwhelmed by the entire task, do it step by step
• maybe step back a level of abstraction and ask yourself, what is the goal

Carbon Sequestration in Forests - Addressing the Scale Question

Whether young or old forests sequester or store more carbon, is a heated debate. Depending on the consideres scale in time, space, and involved proces, there are arguments for either side. This controversy is resolved at the landscape scale.

Young forests are growing faster, while old forests have more dead trees and decomposition. On the other hand, replacement of older forests by young forests will result in a net release of carbon.

• process of defining a problem should include specifying a spatial, temportal and process level
• processes involved:
  • tree growth
  • photosynthesis
  • plant respiration
  • tree death
  • litter production
  • decomposition
  • formation of stable organic matter in soil
  • disturbances (e.g., harvesting and fire)
  • manufacture use
  • disposal of forest products
  • substitution of fossil fuels
• young forests often have a high amount of slash: high decomposition
• detritus, soil, and forest products can be long-term storages
• a single dead tree decomposes and loses mass, but an accumulation of many dead trees can accumulate mass
• NPP = gross growth (in forestry terms)

• If one considers the average production over the length of a rotation, then older forests may be lust as productive as younger ones. This is because no forest can be $X$ years old without having been $X-1$ years old.

• given enough time, ecosystems’ carbon accumulation ability vanishes

• Although old forests have a substantial amount of dead and dying material, these losses are roughly offset by the production of this material. Ironically, it is the very production of that dead and dying material that prevents the older forest ecosystem from being a net carbon source to the atmosphere.

• at large spatial cases with many age classes, effects might cancel out
• disturbances less severe and less frequent: higher carbon storage (and vice versa)
• increased NPP (by climate change) far too small to offset losses caused by conversion of older forests to younger forests

Personal comments

• all the analyses are done in terms of carbon (tons/ha, tons/ha/yr)
• we can do similar analyses in terms of transit time

On the value of preprints: An early career researcher perspective

The publication of preprints, publicly available scientific manuscripts posted on dedicated preprint servers prior to journal-managed peer review can play a key role in addressing these ECR challenges such as timely publication and increased interdisciplinarity in life sciences research.

Introduction

• research output comes in very many varieties: research articles, reviews, commentaries, perspectives, theory manuscripts, methods, data, reagents, model organisms, computational models, patents, drugs, vaccines, software, and highly trained researchers
• still the only “currency” is published articles in peer reviewed journals
• preprints: online, freely available (open-access) scientific manuscripts posted by authors on dedicated servers prior to peer review and publication in an academic journal
• sometimes concurrently to journal submission, sometimes sole way of publication
• no peer review, only screening for appropriate topic
• some journals do not accept preprinted articles

Values of preprints for ECRs

Preprints accelerate science communication that facilitates ECR career progression

• long duration of traditional journal publishing can negatively impact ECRs seeking funding, promotion, and hiring
• added benefit of encouraging collaboration, informal discussion, and sharing and receiving data
• some funding agencies take preprints into account, in order to evaluate the researcher based on quality of work not only publications

Preprints increase ECR visibility and facilitate networking

• increase networking
• discussion in social media
• higher citation rates of later publications

Preprints can help ECRs accelerate training time and optimize research design and quality

• faster knowledge and data dissemination with all its benefits like steeper learning cuves, reduction of costs, avoidance of redundancy

Preprints allow ECRs with limited funds to publish their findings with open access

• low costs open access publishing

Preprints in public health and medical research can boost ECR research

• Was it used much to fight the pandemic?

Preprints can accelerate the peer-review process to make ECRs more efficient

• useful if the author has no experts at hand in the field
• feedback through email or social media

1. researchers can begin to respond to preprint comments before journal-solicited reviews are received
2. researchers can submit higher-quality articles to journals after getting feedback from preprint readers
3. with the exception of a few journals [1], the journal peer review process remains largely opaque and confidential.

If open preprint peer review were to become common practice, rereviewing of the same article could be avoided.

Preprint commenting can help ECRs develop their reviewer skills

Only 20% of scientists perform 69% to 94% of the all journal-solicited peer reviews culminating to 63.4 million review hours a year, 15 million of which are spent rereviewing rejected papers [3,44].

• Commenting on preprints by ECRs is an opportunity to sharpen their reviewing skills and to give them a voice in academic publishing that can expand and diversify the pool of peer reviewers.
• more and more interdisciplinary research needs more and more reviewers
• there are platforms for preprint reviews

Preprints helps ECRs perform corrections via revisions

• easy corrections instead of retractions

Publishing all research findings and conditions in preprints can benefit ECRs

• negative results, replication studies

Perceived concerns by ECR on preprinting

Preprinting leads to scooping

• preprints come with a DOI
• depends on whether journals allow citing preprints and whether they accept being second

Preprinting prevents publication

• unclear publisher and journal policies
• closer collaboration and reverse links between journals and preprint servers could help

Preprints have low visibility

• any publishing option will benefit ECRs who need to prove productivity over a short period of time
• some results might not be published at all otherwise
• preprint search engines are improving

Conclusions

Preprints are already benefiting ECRs and life scientists at large, but we argue that they are underutilized and can be used in new ways to aid ECR development and increase the efficiency of scientific research.

Personal comments

• I remember a case in which I read a paper on entropy rates in marked Poisson processes and got totally stuck for quite some time. Only to learn much later that this was a preprint and they had several errors in the equations… That was rather annoying.
• Recently I reviewed a paper and got stuck with some formulas. But it was a review, I knew there might be something wrong, noted it down in my report, and the authors corrected it. So the actual reader later does not have to go through this process anymore, maybe without knowing, it has not been double checked.
• Sometimes it takes forever to find reviewers, but PhDs need the publications. What does it help however, if preprints are not accepted by the university?

• Preprints empower authors to decide when their work is ready to be shared with the scientific community.

  • This might be too early.
• another means for networking, in particular now in Corona times

• For example, in the biophysics and fluorescence microscopy fields, preprinted methods were used well in advance of the peer-reviewed publication in sample labeling [19–20], instrument design [21–22], and image analysis [23].

  • This is kind of dangerous.
• might be a first or one of many steps away from the ridiculous current publication system
• I might consider my Landau constants paper for preprint for I have no expert in the field at hand. Furthermore, I have a negative result that could go on arXiv.
• How does it help the system if every ECR can publish more and faster?
• I know that at Imperial College London, they have an internal preprint server.

Quantum Theory

I am not sure where this project belongs. It is not like classically reading a paper and it is not about classically citing some quotes of enjoyable books. For starters, I would like to note my impressions. I need to study physics to understand the simplest of Bohm’s arguments, I do not have the time, and the energy to do so fainted as well. So I try to make the best of it and note down what I feel most interesting.

Chapter 1

• in three dimensions, a transverse wave (the link helped me a lot to understand the concept of polarization) has two options for polarization
  • Are there always $d-1$ options for polarization if $d$ is the space dimension?
    • No! Only $d-1$ basis directions, any combination is possible, the wave could also be tilted.
  • What is a longitudinal wave?
• Maxwell’s equations and Fourier mathematics lead to a perfect blackbody radiation theory, as long as the frequency is not too high (Rayleigh-Jeans law)
  • Fourier mathematics, as is mentioned in a footnote on page 10, works as long as the function is piecewise continuous, which I find a pretty remarkable footnote in the realms of the borderline between classical physics and quantum mechanics
• both Maxwell’s equations and Fourier mathematics are undisputable, but something is wrong when it comes to measurements at high frequencies
• Max Planck (Einstein is also mentioned, Nobel Prize?) comes with the idea, that quantized energy packets might do the trick
• it was not ovious earlier noted because the frequency-dependent package size $h\,\nu$ is small enough to make quantum theory look continuous at not too high frequencies
• Maxwell’s distribution reconciles both the Rayleigh-Jeans law and the Wiener law, his new law interpolates even everything in between very well
• most of the things are just greek to me but I hope I got the gist

Extinction risk depends strongly on factors contributing to stochasticity

Extinction depends on four kinds of stochasticity. Models that do not incorporate all four kinds might greatly undererstimate extinction risks.

General situation

Four categories of stochasticity:

• demographic stochasticity: birth and death at the level of individuals (stochasticity underlying equal production rates)
• environmental stochasticity: variation in population-level birth and death rates among times or locations
• the sex of individuals

• demographic heterogeneity: variation in vital rates among individuals within a population (e.g. individuals of different size might have different mating chances/reproduction rates)

• often demographic stochasticity dominates environmental stochasticity (see Southern fulmar example in George’s paper)
• the first stochastic models showed that populations could become extinct even if deterministic models concluded they would persist indefinitely
• long term growth rate of stochastic population differs from deterministic population
• in small populations demographic stochasticity becomes more important because of potential dramatic effects in individual level that would cancel out in larger populations
• in large populations environmental stochasticity might be dominant because the whole population is affected at once
• strictly speaking: stochastic isex determination $\subset$ demographic heterogeneity $\subset$ demographic stochasticity
• most current (as of 2008) models of extinction risk include only females
  • but: stochastic sex ratio can increase variation in population growth above the demographic effects of femals alone
  • males contribute to density regulation; lack of males reduces female mating success
• demographic variance $\supset$ demographic stochasticity $+$ sex ratio stochasticity $+$ demographic heterogeneity
• demographic heterogeneity can affect demogrpahic variance in either direction and thus increase or decrease the extinction risk

Modeling

• usually deterministic skeleton (often Ricker model) plus noise
• here: stochastic Ricker model, all four stochastic categories directly involved (no determinstic plus noise, but immediately stochastic)
  • compare: exponential lifetime distribution nearly impossible to model by determinstic mean plus a stochastic noise
• variance of individuals in next generation increases with additional stochasticity sources
• classical Poisson-Ricker model (pure demographic stochasticity) has lowest variance: great potential for underestimation of extinction risk
• if total variance is held constant:
  • environmental stochasticity: maximum variance in number of individuals in next generation as in the deterministic case (equilibrium), variance parameter is density independent
  • demographic stochasticity: maximum variance further left, at lower population size; higher risk of extiction; variance parameter density dependent
• not only increases the variance of number of individuals in the next generation with increasing stochasticity, the point of maximum variance moves further left and sets smaller populations at even higher risk of extinction
• higher reproduction rate $R$ also increases contribution of nonlinear dynamics to variance in population fluctuations which has a negative effect on survival (if $R$ greater than Ricker bifurcation point, even drastically)

Incubation experiment

• model with all four sources of stochasticity fitted best
• classical model with only demographic stochasticity fitted worst
• demographic heterogeneity dominated environmental stochasticity
  • models without demographic heterogeneity absorb it in the variance parameter of environmental stochasticity and attribute the effect wrongly

Results

• species currently regarded at risk from environmental stochasticity might be at much higher (undetected) risk from demographic stochasticity
• specifically important for small populations

Using heterogeneity indices to adjust basal area – Leaf area index relationship in managed coniferous stands

Abstract (excerpt)

The structure of contemporary managed forests is complex and deviates from experimental forests which are usually even-aged monocultures and single-storied. To apply theoretical growth and yield functions on managed forests, adjustments are required, especially for leaf area index (LAI) which is a key biophysical variable in process-based growth models. To asses this, the performance of canopy LAI in modelling the basal area (BA, measured at breast height = 1.30m) of managed boreal forests dominated by Norway spruce (Picea abies (L). Karst) and Scots pine (Pinus sylvestris L.) was investigated by heterogeneity analysis. The study was based on the assumption that canopy LAI and BA are strongly related and are vital for estimating stand productivity and growth. Managed forests were represented by field data from the 2016 and 2017 Swedish National Forest Inventory (NFI) campaigns.

• including ground-based stand structural heterogeneity (SSH, described by species decomposition, coefficient of tree diameter variation, tree social status, height-diameter ratio) improved LAI-BA and BA-LAI models
• including spectral hereogeneity (SPH, described by begetation and textural indices) improved BA-LAI fit as well, suggesting potential use of Sentinel-2 data (Google Earth Engine) in future grwoth models

1. Introduction

• stem growth per unit of absorbed light positively related to per unit grwoth of tree canopy LAI
• LAI: half of the total surface area of green leaves or needles per unit of ground horizontal surface area
• PAI: LAI + other light blocking elements (e.g. twigs, branches, stems)

In recent years, mixed-species forests have been recognised and a number of studies relating mixture effects on productivity (Pretzsch et al., 2015; Pretzsch and Schuetze, 2016), provision of multiple ecosystem services (Gamfeldt et al., 2013; Forrester and Pretzsch, 2015; Felton et al., 2016) and as an adaptation strategy to cope with climate change (Bolte et al., 2009; Kolstrom et al., 2011) have been published.

• lack of growth models resulted in theoretical growth functions, usually suited for even-aged, single-storied monocultures
• adaptions needed when applying LAI-conversion functions to managed stands

2. Materials and methods

Sweden’s forest covers about 28 million hectares representing 69% of the total land area of the country and it is characterized by a north–south gradient (from 55° N to 69° N) with temperate to boreal vegetation zones, respectively (Skogsdata, 2018). Productive forestland covers about 23.5 million hectares with coniferous species largely dominant (Pinus sylvestris L., 39% and Picea abies (L). Karst, 27%). Birch species (Betula pendula and Betula pubescens) also accounts for 12% of the total productive forest area. On the other hand, the share of the area by mixed-species forests have also increased with 14% and 8% of mixed conifers and mixed conifer/broadleaves on productive forest lands, respectively (Fridman and Westerlund, 2016).

• data restrictions: dbh $\geq 10$ cm, tree height $\geq 4$ m, forest age $\leq 100$ yr
• table with data from NFI (2016–2017):
  • basal area (m^2 / ha)
  • tree height (m)
  • stand age (yr)
  • stem density (trees / ha)
  • species proportion (% of basal area, % of stem number)
  • H-D ratio
  • coefficient of tree diameter variation (CVD)
  • tree social status
  • …
• stand structural variables assumed to influence relationship between measured BA and modelled LAI:
  • species composition and dominance
  • tree size heterogeneity
  • stem density
  • tree height
  • stand age
  • variables describing site characteristics: latitudinal gradient, temperature sum, humidity
• tree dominance: $\geq 65$% or $\geq 70$% proportion in stems or BA (not respectivelty but alternatively)
• site heterogeneity:
  • temperature sum: sum of average daily mean temperature exceeding 5$^{\circ}$C during vegetation period
  • humidity: difference between precipitation and loss of water by transpiration and evaporation during growing season period where temperature exceeds 5$^{\circ}$C
• BA(LAI) is Michaelis-Menten type relation, here LAI is the resource and BA the production variable

3. Results

For both species, general stand structural variables such as tree height, stand age, stem density and heterogeneity variables by CVD, tree social status and H-D ratio were the most significant variables in the final models. For Scots pine, the addition of temperature sum was also significant to the final model.

4. Discussion

Many literature references are mentioned here with classical relationships between LAI and stem diameter and so on. Probably all of those relationships hold in our model by construction.

Recent surveys from the Swedish NFI reports that about 20% of the productive forest-land is a mixture of Scots pine and Norway spruce (Fridman et al., 2014; Holmström et al., 2018). Similarly, multi-layered Scots pine stands with understory layer of Norway spruce and birch created by mass regeneration and size stratification have been observed in central and northern Sweden (Lundqvist et al., 2019).

For Norway spruce, the LAI was higher in the monocultures (100% of the total BA) than in the mixtures because of the lower tree heights observed in the mixtures (Fig. 1E).

Scots pine is a shade intolerant species and therefore, exhibits dominance in mixtures (Goude et al., 2019).

All the size heterogeneity variables had negative coefficients (Table 3) and this supports earlier reports of the negative correlation between tree size inequality and stand productivity (Zeller et al., 2018; Sun et al., 2018).

Species-specific sizes and crown allometry of dominant (larger) trees show greater light interception and light-use efficiency than for smaller trees in heterogenous stands (Pretzsch, 2009; Gspaltl et al., 2013; Binkley et al., 2013a).

Personal conclusions

We can use data from NFI (Table 1) for to initialize stands and to compare later stages of our stands with. In any case we would need to come up with temperature sums and humidities for our stands.

We can use provided formulas to compute stand characteristics of our modeled stands:

• species composition: Eq. (1)
• species dominance: Eqs. (2) and (3)
• tree size heterogeneity: Eqs. (4)–(6)

Moving beyond the incorrect but useful paradigm: reevaluating big-leaf and multilayer plant canopies to model biosphere-atmosphere fluxes – a review

Abstract

This article reviews the scientific debate of big-leaf vs multi-layer canopy modeling for land surface models. Comparisons with flux tower measurements show supremacy of good multi-layer models (5-10 layer suffice).

• big leaf: one-layer canopy

Important vertically resolved quantities:

• canopy air temperature
• specific humidity
• wind speed
• water leaf potential (for dry soils)

1. Introduction

George Box’s often quoted statement that “all models are wrong but some are useful” was part of a discussion regarding the process of building scientific models (Box, 1979). He argued for parsimonious models because “simplicity illuminates, and complication obscures” and because “indiscriminate model elaboration is in any case not a practical option because this road is endless”.

• dual source canopy: one leaf layer, separate fluxes for leaves and soil
• wind speed decreases with canopy depth
• relatively dense forests: daytime air temperature varies, mod-canopy maximum a few degrees warmer than in understort
• vertical profiles in leaf water potential can create water-stressed leaves in upper canopy with reduced stomatal conductance and photosynthesis
• heterogeneity in leaf temperature
• CLM5: one canopy layer but > 20 soil layers (despite poorly known necessary vertical thermal and hydraulic parameters)
• Raupach (1991):
  • canopy-atmosphere models (CAM, multi-layer): locally useful
  • simplified canopy-atmosphere models (SCAM): globally useful
• profound differences in microclimate between overstory and understory: important for ecological impact studies
• today:
  • better numerical solutions, CAM might also globally be useful
  • better surface flux modeling

2. Background

• multi-layer models compute generally: radiative transfer, stomatal conductance, leaf energy fluxes,turbulent diffusion; coupled to temperature and water vapor concentration, CO$_2$ concentration at each layer
• big leaf: Penman-Monteith equation for evaporation, atmosphere governed by
  • $R^{‘}_n$: net energy availabe to canopy after accounting for soil and biomass heat storage
  • $\delta^{‘}$: saturation deficient at some height above canopy
  • $g_a$: buld aerodynamic conductance for scalar transport (assumed equal for heat and water vapor, only for convenience)
  • $g_c$: bulk stomatal conductance (canopy conductance)
• relationships between multi-layer and single-layer parameters not uniquely determined
• physical meaning of Penman-Monteith equation gets lost if accumulated over several layers, in particular $g_c$ is impossible to be precisely specified
• leaf nitrogen gradient in canopy less steep than light gradient
• both types of models can be separated into sunlit and shaded leaves
• also leaf inclination angle and leave age cohorts added later to multi-layer models
• green leaf area index and brown stem area index:
  • leaf surface: transpiration and photosynthesis
  • leave and stem surface (plant area index): radiative transfer, canopy evaporation, sensible heat flux, momentum absorption
• plant canopy models must parameterize turbulent fluxes between vegetation and atmosphere
  • MOST: Monin-Obukhov Similarity Theory
  • RSL: roughness sublayer, region within and just above canopies where observed flux-gradient relationships depart from MOST
• Lagrangian approach to turbulent eddies
• localized near-field theory

CLM5 also requires a complex iterative calculation of surface fluxes and canopy temperature (Lawrence et al., 2019). Up to 40 iterations are allowed in CLM5, but convergence is not guaranteed, in which case arbitrary adjustments to fluxes are implemented to achieve energy conservation.

• vertical distribution of leaf area most important
• in data-rich world of today it can be measured (opposing to soils)
• also leaf angular inclination, vertical nitrogen profile

3. The multilayer canopy model

Model revised from Bonan et al (2018) to allow for a generalized continuum of 1 to N layers and is identified as CLM-ml v1.

• mesic habitat: moderate or well-balanced supply for moisture
• xeric habitat: dry

4. Comparison of one-layer and multilayer canopies

In general, the multilayer model’s fluxes come much closer to the observations.

Reasons

• sunlit leaves dominate upper canopy, most shaded leaves in lower canopy
• $V_{\text{cmax}}$ declines with canopy depth down to 1/2
• daytime air temperature peaks in mid-canopy and decreases by 1 degree or more in lower canopy
• nighttime minimum temperature in upper canopy
• daytime specific humidity less in upper canopy
• wind speed decreases with canopy depth
• much of solar radiation absorbed in upper canopy
• leaf transpiration peaks in upper canopy: lower mid-day leaf water potential in upper canopy
• leaves in upper canopy 2 degrees warmer than in lower canopy

Important variabales

• within-canopy air temperature, specific humidity, wind speed profiles
• sensible heat flux most sensitive to temperatire profile, sunlit and shaded proportion sensible to wind speed (not the total)
• latent heatflux sensible to temperature and humidity
• GPP differences between multi-layer and big leaf driven by their light profiles, low response to well-mixed assumption, temperature, humidity and wind profiles

Number of layers

• better results with increasing number of layers
• little immprovements beyond 10 layers
  • valid for simply-structured canopies (not tropical ones)
  • might be model dependent
• proportion of absorbed solar radiation between sunlit and shaded leaves independent of number of layers

5. Discussion

The results presented in this study show that resolving vertical profiles of air temperature, specific humidity, and wind speed in forest canopies reduces surface flux biases compared with a single canopy layer.

6. Conclusions

The modeling analyses presented herein show that the dismissal of multilayer canopy models as too complex is no longer valid. Multilayer canopy models are not a complication that obscures; they are instead a simplification that illuminates the processes controlling surface flux calculations compared with land surface models such as CLM5. Multilayer canopy models are a necessary model elaboration, not the indiscriminate model elaboration that Box (1979) believed was endless and argued against.

Indeed, a comparison of technical descriptions reveals the multilayer canopy (Bonan et al., 2018) to be the simpler, not more complex, implementation of surface fluxes compared to CLM5 (Lawrence et al., 2019).

CLM5 has an ecosystem demography component (FATES; Fisher et al., 2015; Koven et al., 2020), but in coupling FATES to CLM5 the vertically-structured canopy, utilized for photosynthesis and light competition in FATES, is reduced to a canopy leaf area index and a canopy conductance for CLM5’s big-leaf surface flux formulation (Lawrence et al., 2019). A more consistent coupling between the models would utilize the vertical structure of FATES in a multilayer model of surface fluxes.

Personal conclusion

In general, multilayer canopy models are much better than big leaf models. However, for our purpose of computing GPP, we can omitt many of the overwhelmingly complicated processes and concentrate on the light profile.

How to avoid having your manuscript rejected: Perspectives from the Editors-in-Chief of Soil Biology and Biogeochemistry

In this editorial piece the editors in chief of SBB explain to what authors should pay attention before submitting papers in general, and to SBB in particular.

1. Introduction

• papers are for the readers, not the authors (which in fact is not true): do readers gain novel insights and new knowledge?

Checklist

• journal’s target audience
• methods and analyses support interpretations and conclusions
• conclusions well-grounded in the data and conclusive
• speculation is clearly identified as such
• clear writing

2. SBB’s criteria are based on our aims and scope

Soil Biology & Biochemistry publishes scientific research articles of international significance which describe and explain fundamental biological and biochemical features and processes occurring in soil systems.

1. Audience
   • what does the reader already know: leave it out and take it for granted
2. Questions
   • no case studies, only broadly relevant questions
3. Focus and scale
   • drivers of processes of soil biochemistry, not just soil as a “brown box”
4. Novelty and importance
   • well-studied processes in new circumstances or with new methods
   • incremental science: larger synthesis, new insights, deeper understanding
   • show why and how the work is novel
5. Language
   • English: AE tends to shorter sentences (laconic authors such as Mark Twain and Ernest Hemmingway), BE tends to longer and more elaborate prose (Jane Austen and Thomas Hardy)

3. An Editor’s approach: How we apply these standards when we get a manuscript

• first check for: international relevance, fundemantality, soil systems
• language check: title, highlights, opening paragraphs
• first impressions matter
• first check happens in short blocks of free time, if there are problems the paper is put on the backburner

Common issues that lead to a “Desk Reject”

1. Telling the wrong story
   • message must meet the journal’s scope
   • opening paragraph defines a problem to solve
   • conclusions will come back to this problem
   • opening problem defines target audience
   • for SBB: be general here!
2. No real question or hypothesis
   • define specific objectives at end of introduction: What question will you answer?
   • identify a knowledge gap
   • be as precise as possible
3. Poor presentation
   • read the *Guide to Authors**
   • use little color, good palettes, large font sizes, line numbers
4. Weak conclusions
   • show what the work has contributed and has it has advanced understanding
   • avoid “more research is needed…”: point out what you have done, not what you have not done
   • discuss limitations earlier
   • be precise, what are the exact implications?
   • “Show, don’t tell”.
   • first and last words are most powerful

4. Conclusions

• check journal’s scope
• read as an editor/reviewer/reader
• ironically, typo “concecus” in last words of the paper

The fate and transit time of carbon in a tropical forest

The paper is on means and quantiles of transit times of C in the Porce forest (Columbia). Furthermore, the concept of $\mathrm{NPP}/\mathrm{GPP}$ as $\mathrm{CUE}$ is questioned and an interpretation as $\mathrm{NPP}/\mathrm{GPP} = R_h/\mathrm{GPP}$ is suggested.

• The time that C fixed as $\mathrm{GPP}$ spends in an ecosystem is relevant to understand feedbacks between ecosystems and the climate system.
• Plot level estimates of
• Used average of parameters from a parameter set obtained by MCMC.
  • Average okay for linear models?
  • What about nonlinear models?
• autotrophic pools: foliage, wood, fine roots, coarse roots
• heterotrophic pools: fine litter, CWD, soil carbon
• Most metabolic processes operate on an intra-annual time-scale.

We obtained an average value of 0.3 for the ratio NPP:GPP for the forests at equilibrium, a ratio that is often called carbon use ef- ficiency ($\mathrm{CUE}$) (Chambers et al., 2004; DeLucia et al., 2007; Gifford, 2003; Malhi et al., 2015). According to common interpretation, this ratio would suggests that 30% of the photosynthetically fixed car- bon is used for biomass production. Similar values for $\mathrm{CUE}$ with simi- lar interpretations are also given by Chambers et al. (2004) and Malhi et al. (2013), although larger variability in $\mathrm{CUE}$ is reported in Doughty et al. (2018). However, we believe that this common interpretation of $\mathrm{CUE}$ has problems since, as our transit time distribution showed, au- totrophic respiration is composed of carbon that spends some time in biomass before being respired. The amount of time carbon stays in plant cells can vary from hours to decades, but photosynthates have to be metabolized from living cells (biomass) for CO2 production to occur. Thus, autotrophic respiration originates from biomass already produced; however, most of this metabolism occurs very quickly as the transit time distribution suggests, giving the false impression that a large proportion of carbon was not used to produce biomass. As other authors have shown (DeLucia et al., 2007; Gifford, 2003), estimates of $\mathrm{CUE}$ depend largely on whether estimates are made on short or long periods of time, and the transit time distribution pro- vides good support for avoiding an interpretation of this ratio out of the context of the time-­scales involved.

\begin{equation} \nonumber \frac{\mathrm{NPP}}{\mathrm{GPP}} = \frac{\mathrm{GPP}-R_a}{\mathrm{GPP}} = \frac{R_e}{\mathrm{GPP}} \end{equation}

• $\mathrm{NPP}/\mathrm{GPP} = 0.3$ means that 70% of total photosynthates are respired by autoptrophs and only 30% by heterotrophs.
• no relation to biomass production (as $\mathrm{CUE}$), but interpretation in terms of different pathways of fixed C

Communicating scientific uncertainty

This paper tries to characterize, asses, and convey the uncertainties relevant to each of the following decision classes:

• Decisions about action thresholds: Is it time to act?
• Decisions with fixed options: Which is best?

• Decisions about potential options: What is possible?

• problem of too high or too little trust in science
• scientists may overemphasize unimportant uncertainties and leave out others because of routine in the field
  • the right level depends on the decision to be made
• nonpersuasive (informing choices) and persuasive (advocating for a choice) communication

Decisions About Action Thresholds: Is It Time to Act?

• threshold passed: time to act
• uncertainty about the threshold

Characterizing Uncertainty

• science judged by apparent wisdom of recommendations
• outcome bias: decisions judged by outcomes rather than by their wisdom
• hindsight bias: foreseeability of outcomes exaggerated
• distrust can be raised if messages are misaligned with decision maker’s values

Assessing Uncertainty

• states must be precisely defined: what is a tumor, what is a flood?
• states/eventc can be observed directly (counting) or indirectly (through biomarkers)
• make decision thought process transparent

Conveying Uncertainty

• standard words may have special meaning in certain fields
• content of science-based knowledge should be based on what recipients know already

Decisions with Fixed Options: Which Is Best?

• recommendation or informing of choices

Characterizing Uncertainty

• probability distributions or parts of it or several distributions for different cases
• scientific uncertainty might increase over time as research reveals unforeseen complications

Assessing Uncertainty

• uncertainty in data, in how data was collected, how it is treated
• uncertainty of method or model
• all kinds of assumptions made

A recent experiment reduced this uncertainty for electricity field trials, finding a 2.7% reduction in consumption during a month in which residents received weekly postcards saying that they were in a study. That Hawthorne effect was as large as the changes attributed to actual programs in reports on other trials.

Conveying Uncertainty

When uncertainties arise from limits to the science, decision makers must rely on the community of scientists to discover and share problems, so as to preserve the commons of trust that it enjoys.

For example, there is wide variation in how laypeople interpret the expressions of uncertainty improvised by the IPCC, in hopes of helping nonscientists.

Decisions About Potential Options: What Is Possible?

• decision makers try to create options

When they choose to act, they may wish to create options with more certain outcomes in order to know better what they will get, or less certain ones in order to confuse rivals.

Characterizing Uncertainty

• graph of variables and their relations (influence diagram), run scenarios on it
• uncertainties in both variables and their relationships
• uncertainty from missing variables (knowingly or unknowingly)

Assessing Uncertainty

• run model with values sampled from probability distributions, compute sensitivity of predictions
• assess uncertainty of factors that science typically ignores or takes for granted

Conversely, aviation has reduced uncertainty by addressing human factor problems in instrument design (52) and cockpit team dynamics (53). Decision makers need to know which factors a field neglects and what uncertainty that creates.

Conveying Uncertainty

To create options, people need to know the science about how things work.

• problem with unintuitive integration (dynamics, nonlinearities)

Eliciting Uncertainty

Science communication is driven by what audiences need to know, not by what scientists want to say.

• standard format required that scientists can create and decision makers can rad

Variability

• use numerical values instead of unclear words (see IPCC)

Internal Validity, External Validity

• effects on credible intervals: they differ from confidence intervals because of internal (evaluation of studies) and external validity (extrapolation of rsults) and pedigree of scientific results

Conclusion

Performing these tasks demands commitment from scientists and from their institutions. It also demands resources for the direct costs of analysis, elicitation, and message development, and for the opportunity costs of having scientists spend time communicating uncertainty rather than reducing it (through their research).

Model–data synthesis in terrestrial carbon observation: imethods, data requirements and data uncertainty specifications

The focus of this paper is observation of the carbon cycle, and in particular its land-atmosphere compo- nents, as one part of an integrated earth observation system.

Introduction

• model data synthesis: the combination of the information contained in both observations and models through both parameter-estimation and data-assimilation techniques
  • model testing and data quality control
  • interpolation of spatially and temporally sparse data
  • inference from available observations of quantities which are not directly observable
  • forecasting
• 3 themes:
  • model-data syntheis based on terrestrial biosphere models as essential component of a terrestrial carbon observation ssystem (TCOS)
  • data uncertainties as as important as data values
  • sound uncertain specifications

Purposes and attributes of a TCOS

A succinct statement of the overall purpose of a TCOS might be: to operationally monitor the cycles of carbon and related entities (water, energy, nutrients) in the terrestrial biosphere, in support of comprehensive, sustained earth observation and prediction, and hence sustainable environmental management and socio- economic development.

• a TCOS needs:
  • scientific credibility
  • respect carbon budgets
  • high spatial resolution
  • high temporal resolution
  • large number of entities
  • sufficient range of processes
  • partitioninong of net fluxes
  • quantification of uncertainty
  • altogether: swiss army knife (eierlegende Wollmilchsau)

Model–data synthesis: methods

Overview

All applications rest on three foundations: a model of the system, data about the system, and a synthesis approach.

Model

• ODE or difference equation, including a noise term
• noise accounts for imperfect model formulation and stochastic variability in forcings and parameters

Data

• two types:
  • observations and measurements
    • $z=h(x, u) + \text{noise}$, where $h$ specifies the deterministic relationship between the measured quantities $z$ and $u$ and the system state $x$; z accounts fot mesurement error and representation error
  • prior estimates for model quantities

Synthesis

• finding optimal match between data and model
• 3 kinds of output:
  • optimal estimates for model properties to be adjusted (target variables)
  • uncertainties about these estimates
  • assessment of fitting the data, given the uncertainties
• 3 basic choices:
  • target variables
  • cost function
  • search strategy for optimal values
    • nonsequential: all data treated at once
    • sequential: data incorporated step by step

Target variables

• model parameters $p$, forcing variables $u$, initial conditions $x^0$m state vector $x^n$ itself: all collected in vector $y$
• parameter estimation problems: $y=p$
• data assimilation problems: target variables can be any model property, with emphasis on state variables

Cost function

• common choice:

\begin{equation} \label{eqn:cf} J(y) = (z-h(y))^T[\operatorname{Cov}\,z]^{-1}(z-h(y)) + (y-\hat{y})^T[\operatorname{Cov}\,\hat{y}]^{-1}(y-\hat{y}) \end{equation}

• $\hat{y}$ vector of priors (a priori estimates for target variables)
• model–data synthesis problem: vary $y$ to minimize $J(y)$, subject to the constraint that $x(t)$ must satisfy the dynamic model
• $y$ at the minimum is the a posteriori estimate of $y$, including information from the observations as well as the priors
• Eq. \eqref{eqn:cf} is minimum-variance estimate for $y$
  • for any error distributed it is unbiased
  • minimum error covariance among all in $z$ linear and unbiased estimates
  • if error distributions Gaussian, even a maximum likelihood estimate for $y$, conditional on data and model dynamics
• other choices for other problems or other error distributions

Search strategies for nonsequential problems

Example

Thus the cost function, and thence the entire minimization, takes a form in which neither the observations nor the prior estimates appear; they are replaced by quantities a and b scaled by the square roots of the inverse covariance matrices, which are measures of confidence. This is no mathematical nicety; rather it demonstrates that the data and the uncertainties are completely inseparable in the formalism. To put the point provocatively, providing data and allowing another researcher to provide the uncertainty is indistinguishable from allowing the second researcher to make up the data in the first place.

Algorithms for nonsequential problems

A high condition number of the Hessian of $J$ indicates that some linear combination(s) of the columns are nearly zero, that is, that the curvature is nearly zero in some direction(s), so that the minimization problem is ill-conditioned, as in the case of a valley with a flat floor.

• analytical solution: only possible of $h(y)=H\,y+\text{noise}$ linear- gradient descent: simple and low cost, but tend to find local minima near starting value rather than global minimum
• global search: find global minimum by searching through the whole $y$ space: overcome local minimum problem, but high costs
  • for example, simulated annealing finds the vicinity of a global minimum
  • then apply gradient descent from there

Search strategies for sequential problems

• Kalman filter, genetic methods
• adjoint methods (backward integration)

Discussion of model–data synthesis methods

Differences between nonsequential and sequential strategies.

• advantages sequential:
  • optimal state can differ from that embodied in model equations
  • $x^n$ required to be included in $y$ (leads to intractable dimensionality in nonsequential models
  • size does not grow with length of model integration
  • can easily handle incremental extensions to time series observations
• advantages nonsequential:
  • treat all data at once (see impacts of data at different points in time)

Model and data error structures

• often assumend Gaussian, no temporal correlations
• generalizations active area of research

Nonsequential and sequential parameter estimation

• usually done nonsequantially (LS)
• but: one can incorporate $p$ as part of $y$ and do sequential analysis
  • allows change of parameters by data, caused by catastrophic events

Model–data synthesis: examples

• parameter estimation
• atmospheric inversion methods to infer surface-atmosphere fluxes from atmospheric composition observations and atmospheric transport models
• combination, advantages:
  • different observations constrain different processes
  • different observations have different resolutions in space and time (also a problem)
• weather forecast by atmospheric and ocean circulation models

Data characteristics: uncertainty in measurement and representation

We have emphasized that data uncertainties affect not only the predicted uncertainty of the eventual result of a model–data synthesis process, but also the predicted best estimate.

• scale mismatches between measurement and models is part of the representation error

An analogous temporal representation error arises when flask measurements (actually grab samples in time) are interpreted as longer-term means…

A further contribution to representation errors for most atmospheric inversion studies to date has been the projection of possible source distributions to a restricted subspace, usually by dividing the earth into a number of large regions. This is done both for computational reasons and to reduce the error amplification arising from under-determined problems. Errors in the prescription of flux distributions within these regions give rise to a so-called aggregation error, described and quantified by Kaminski et al. (2001). This error can be avoided by using adjoint representations of atmospheric transport that do not require aggregation (Rodenbeck et al., 2003a,b).

There are few experiments where representation errors can be evaluated, since this requires simultaneous knowledge of sources and atmospheric transport. However, one can use the range of model simulations as a guide (e.g. Law et al., 1996; Gurney et al., 2003).

• GPP = [net assimilation]
• net primary productivity (NPP) = [GPP - autotrophic respiration]
• net ecosystem productivity (NEP) = [NPPheterotrophic respiration]
• net biome productivity (NBP) = [NEP - disturbance flux]
• disturbance flux: grazing, harvest, and catastrophic events (fire, windthrow, clearing)

Mistmatches of measurement and model scale

Observational issue

There are several options to relate fine-scaled measurements to a coarse-scaled model:

• $z_{\text{fine}}$ is a noisy sample of $z_{\text{coarse}}$, variability in $z_{\text{fine}}$ (covariance $R_{\text{fine}}$) treated as contribution to representation error
• direct aggregation: $z_{\text{coarse}}$ is weighted average of $z_{\text{fine}}$
• $z_{\text{fine}} = g(x_{\text{coarse}}, a_{\text{fine}})$: relate fine-scale observations to coarse-scale state variables and additional fine-scale sncillary data such as topography

Scaling of dynamic model

Translate $\mathrm{d}x/\mathrm{d}t=f(x, u, p)$ between scales:

fine-scale and coarse-scale equations are different (for instance, biased with respect to each other) because of interactions between fine-scale variability and nonlinearity in the fine-scale function f(x, u, p)

Summary and conclusions

Critical error properties include:

• the diagonal elements $[\operatorname{Cov}(z)]_{mm}=\sigma_m^2$ of the measurement error covariance matrix (where $\sigma_m^2$ is the error magnitude for an observation $z_m$)
• the correlations between different observations, quantified by the off-diagonal elements of the covariance matrix
• the temporal and the spatial structure of errors
• the error distribution
• possible scale mismatches between measurements and models
• the representation of the observations in the model

Model Selection and Multimodel Inference

We wrote this book to introduce graduate students and research workers in various scientific disciplines to the use of information-theoretic approaches in the analysis of empirical data. These methods allow the data-based selection of a “best” model and a ranking and weighting of the remaining models in a pre-defined set. Traditional statistical inference can then be based on this selected best model. However, we now emphasize that information-theoretic approaches allow formal inference to be based on more than one model (multimodel inference). Such procedures lead to more robust inferences in many cases, and we advocate these approaches throughout the book.

Preface

We recommend the information-theoretic approach for the analysis of data from observational studies. In this broad class of studies, we find that all the various hypothesis-testing approaches have no theoretical justification and may often perform poorly. For classic experiments (control–treatment, with randomization and replication) we generally support the traditional approaches (e.g., analysis of variance); there is a very large literature on this classic subject. However, for complex experiments we suggest consideration of fitting explanatory models, hence on estimation of the size and precision of the treatment effects and on parsimony, with far less emphasis on “tests” of null hypotheses, leading to the arbitrary classification “significant” versus “not significant.” Instead, a strength of evidence approach is advocated.

Introduction

1.1 Objectives of the Book

• model parameters can provide insights even if not linked to directly observable variables
• AIC used routinely in time series analysis
• marriage of information theory and mathematical statistics: Kullback’s (1959) book
• Akaike considered AIC an extension of R. A. Fisher’s likelihood theory
• estimates of model selection uncertainty: inference problems that arise from in using the same data for both model selection and parameter estimation and inference; if irgnored, precision overestimated
• multimodel inference (MMI): model averaging, confidence sets on models
• small sample size: AIC$_c$ instead of AIC (Mina)

1.2 Background Material

1.2.1 Inference from Data, Given a Model

• Fisher’s likelihood theory assumes that the model structure is known and correct, only the parameters are to be estimated

1.2.2 Likelihood and Least Squares Theory

• LS and likelohood yield identical estimators if structural parameters of residuals are normal and independent
• LS:
  • $y_i = \beta_0 + \beta_i\,\cdot x_i + \epsilon_i$ with $\epsilon_i\sim\mathcal{N}(0,\sigma^2)$ independent
  • LS gives $\hat{\beta_0}$ and $\hat{\beta_i}$ such that the residual square sum $\operatorname{RSS} = \sum_i \epsilon_i^2$ is minimized
• in likelihood function, data are given and model is assumed; interest in estimating unknown parameters: likelihood is function of only the parameters
• $\mathcal{L}(\theta\,|\,data,\,model) = \mathcal{L}(\theta,|\,x,g)$ is the likelihood of the unknown parameter $\theta$, given data $x$ and model $g$
• likelihood is a relative value
• LS is a special case of ML
• $\sigma^2$ is to be considered as a parameter, $\hat{\sigma}^2$ differs by a multiplicative constant (depending on number of parameters and samplie size) for LS and ML
• in LS, $\operatorname{RSS}=n\hat{\sigma}^2$ is minimized, which is for all parameters other than $\sigma^2$ equivalent to maximizing $-1/2\cdot n\,\log \hat{\sigma}^2$

1.2.3 The Critical Issue: “What Is the Best Model to Use?”

As Potscher (1991) noted, asymptotic properties are of little value unless they hold for realized sample sizes.

• model selection based on parsimony, information-theoretic criteria, selection uncertainty

1.2.4 Science Inputs: Formulation of the Set of Candidate Models

Building the set of candiate models is partially a subjective art; […] The most original, innovative part of scientific work is the phase leading to the proper question.

• lots of exploratory work necessary
• large datasets are likely to support more complexity: one hase to correct for this
• Freedman’s paradox: If number of variables $\sim$ number of observations, high $R^2$ and so on possible even if $y$ independent of data
• “An inference from a model to some aspect of the real world is justified only after the model has been shown to adequately fit relevant empirical data.”
• careful thinking rather than brute force

1.2.5 Models Versus Full Reality

• fundamental assumption: none of the candidate models are the “true model” for the “true model” is infinite-dimensional

Models, used cautiously, tell us “what effects are supported by the (finite) data available.” Increased sample size (information) allows us to chase full reality, but never quite catch it.

If we were given a nonlinear formula with 200 parameter values, we could make correct predictions, but it would be difficult to understand the main dynamics of the system without some further simplification or analysis. Thus, one should tolerate some inexactness (an inflated error term) to facilitate a simpler and more useful understanding of the phenomenon.

[…] there are often several large, important effects, followed by many smaller effects, and, finally, followed by a myriad of yet smaller effects. […] Rare events that have large effects may be very important but quite difficult to study.

Conceptually, the role of a good model is to filter the data so as to separate information from noise.

1.2.6 An Ideal Approximating Model

It is important that the best model is selected from a set of models that were defined prior to data analysis and based on the science of the issue at hand.

There are many cases where two or more models are essentially tied for “best,” and this should be fully recognized in further analysis and inference, especially when they produce different predictions. In other cases there might be 4–10 models that have at least some support, and these, too, deserve scrutiny in reaching conclusions from the data, based on inferences from more than a single model.

• good tool to assess model quanlity: small-sized confidence intervals with high confidende ($\geq 0.95$) for parameter values

1.3 Model Fundamentals and Notation

Introduction of model classes and notation.

1.3.1 Truth or Full Reality $f$

Truth is denoted by $f$ as an abstract concept because it is unknown.

1.3.2 Approximating models $g_i(x\,|\,\theta)$

Ideally, the set of $R$ models will have been defined prior to data analysis. These models specify only the form of the model, leaving the unknown parameters ($θ$) unspecified.

1.3.3 The Kullback–Leibler Best Model $g_i(x\,|\,\theta_0)$

• “K-L best” means relative to the unkown truth $f$

The parameters that produce this conceptually best single model, in the class $g(x\,|\,\theta)$0, are denoted by $\theta_0$. Of course, this model is generally unknown to us but can be estimated; such estimation involves computing the MLEs of the parameters in each model ($\theta$) and then estimating K-L information as a basis for model selection and inference. The MLEs converge asymptotically to $\theta_0$ and the concept of bias is with respect to $\theta_0$, rather than our conceptual “true parameters” associated with full reality $f$.

1.3.4 Estimated Models $g_i(x\,|\,\hat{\theta})$

In a sense, when we have only the model form $g(x\,|\,\theta)$ we have an infinite number of models, where all such models have the same form but different values of $\theta$. Yet, in all of these models there is a unique K-L best model. Conceptually, we know how to find this model, given $f$.

1.3.5 Generating Models

One should not confuse a generating model or results based on Monte Carlo data with full reality $f$.

1.3.6 Global Model

Ideally, the global model has in it all the factors or variables thought to be important. Other models are often special cases of this global model. There is not always a global model. If sample size is small, it may be impossible to fit the global model. Goodness-of-fit tests and estimates of an overdispersion parameter for count data should be based (only) on the global model. The concept of overdispersion is relatively model-independent; however, some model must be used to compute or model any overdispersion thought to exist in count data.

• In compartmental systems, a global model with $n$ pools is then the models where all pools are connected, and all pools have external inputs and outputs
• other models or special cases are models with missing connections.
• In terms of AIC and finding the right dimension ($n$, number of pools), should one first compare several “global” models with one another? If so, how (pure likelihood, AIC)?

The advantage of this approach is that if the global model fits the data adequately, then a selected model that is more parsimonious will also fit the data (this is an empirical result, not a theorem).

1.3.7 Overview of Stochastic Models in the Biological Sciences

• linear and nonlinear regression (based on LS and ML)
• log-linear and logistic models (mostly for count data)
• compartmental models as state transitions (more advanced: “random effects”, Kreft and deLeeuw 1998)
• differential equations in general
• open and closed capture-recapture, band recovery, distance sampling
• spatial models (Kriging can be viewed as LS technique), Gibbs sampler
• spatiotemporal (MCMC methods)

There are general information-theoretic approaches for models well outside the likelihood framework (Qin and Lawless 1994, Ishiguo et al. 1997, Hurvich and Simonoff 1998, and Pan 2001a and b). There are now model selection methods for nonparametric regression, splines, kernel methods, martingales, and generalized estimation equations. Thus, methods exist for nearly all classes of models we might expect to see in the theoretical or applied biological sciences.

1.4 Inference and the Principle of Parsimony

1.4.1 Avoid Overfitting to Achieve a Good Model Fit

“How many parameters does it take to fit an elephant?”

[Wel] concluded that the 30-term model “may not satisfy the third-grade art teacher, but would carry most chemical engineers intopreliminary design.”

Wel’s finding is both insightful and humorous, but it deserves further interpretation for our purposes here. His “standard” is itself only a crude drawing—it even lacks ears, a prominent elephantine feature; hardly truth. A better target would have been a large, digitized, high-resolution photograph; however, this, too, would have been only a model (and not truth). Perhaps a real elephant should have been used as truth, but this begs the question, “Which elephant should we use?”

1.4.2 The Principle of Parsimony

Statisticians view the principle of parsimony as a bias versus variance tradeoff. In general, bias decreases and variance increases as the dimension of the model increases.

If we believe that truth is essentially infinite-dimensional, then overfitting is not even defined in terms of the number of parameters in the fitted model.

Instead, we reserve the terms underfitted and overfitted for use in relation to a “best approximating model”.

Here, an underfitted model would ignore some important replicable (i.e., conceptually replicable in most other samples) structure in the data and thus fail to identify effects that were actually supported by the data. In this case, bias in the parameter estimators is often substantial, and the sampling variance is underestimated, both factors resulting in poor confidence interval coverage. Underfitted models tend to miss important treatment effects in experimental settings.

Overfitted models, as judged against a best approximating model, are often free of bias in the parameter estimators, but have estimated (and actual) sampling variances that are needlessly large (the precision of the estimators is poor, relative to what could have been accomplished with a more parsimonious model). Spurious treatment effects tend to be identified, and spurious variables are included with overfitted models.

The goal of data collection and analysis is to make inferences from the sample that properly apply to the population. The inferences relate to the information about structure of the system under study as inferred from the models considered and the parameters estimated in each model. A paramount consideration is the repeatability, with good precision, of any inference reached. When we imagine many replicate samples, there will be some recognizable features common to almost all of the samples. Such features are the sort of inference about which we seek to make strong inferences (from our single sample). Other features might appear in, say, 60% of the samples yet still reflect something real about the population or process under study, and we would hope to make weaker inferences concerning these. Yet additional features appear in only a few samples, and these might be best included in the error term $(\sigma^2)$ in modeling.

The data are not being approximated; rather we approximate the structural information in the data that is replicble over such samples (see Chatfield 1996, Collopy et al. 1994). Quantifying that structure with a model form and parameter estimates is subject to some “sampling variation” that must also be estimated (inferred) from the data.

Some model selection methods are “parsimonious” (e.g., BIC, Schwarz 1978) but tend, in realistic situations, to select models that are too simple (i.e., underfitted). One has only a highly precise, quite biased result.

• precision: model results look similar for different datasets from the same source
• overfittied models replicate the very dataset at hand and are thus imprecise, meaning that model parameters are uncertain

This example illustrates that valid statistical inference is only partially dependent on the analysis process; the science of the situation must play an important role through modeling.

• mass balance is such a scientific fact

1.4.3 Model Selection Methods

Generally, hypothesis testing is a very poor basis for model selection (Akaike 1974 and Sclove 1994b).

• stepwise backwards (removing variables step by step): one misses parameter combinations and hence possible synergistic effects
• cross-validation:
  • data are divided into two partitions, first partition is used for model fitting and the second is used for model validation
  • then a new partition is selected, and this whole process is repeated hundreds or thousands of times
  • some criterion is as basis for model selection, e.g. minimum squared prediction error
  • computationally expensive

1.5 Data Dredging, Overanalysis of Data, and Spurious Effect

• data dredging: analzing data and searching for patterns without questions or goals
• resulting models overfitted and without predivtive power

1.5.1 Overanalysis of Data

• two versions of data dredging: iteratively adding variables and trying “all” models
• with increasing computer power become more poular
• better: think before doing data dredging

Journal editors and referees rarely seem to show concern for the validity of results and conclusions where substantial data dredging has occurred. Thus, the entire methodology based on data dredging has been allowed to be perpetuated in an unthinking manner.

We believe that objective science is best served using a priori considerations with very limited peeking at plots of the data, parameter estimates from particular models, correlation matrices, or test statistics as the analysis proceeds.

1.5.2 Some Trends

• more use of likelihood (compuationally more expensive, more flexible) then least squares
• less hypothesis thesting, more estimation of effects and confidence intervals
• no formal test theory for model selection exists, how to interpret diferent $P$-values from tests with different powers
• likelihoods require nested models

1.6 Model Selection Bias

• data are used to both select a parsimonious model and estimate the model parameters and their precision (i.e., the conditional sampling covariance matrix, given the selected model).
• large biases in regression coefficients are often caused by data-based model selection
• if a variable would be selected (model selection) into a model by only few of a large numble of samples, this variable’s importance will be vastly overestimated if one looks only at one of the datasets which would suggest to include the variable
• you actually don’t know that because you only have this particular dataset available
• even $t$-tests will tell you to include this variable

1.7 Model Selection Uncertainty

Denote the sampling variance of an estimator $\theta$, given a model, by $\operatorname{var}(\theta\,|\,\text{model})$. More generally, the sampling variance of $\hat{\theta}$ should have two components: (1) $\operatorname{var}(\theta\,|,\text{model})$ and (2) a variance component due to not knowing the best approximating model to use (and, therefore, having to estimate this). Thus, if one uses a method such as AIC to select a parsimonious model, given the data, and estimates a conditional sampling variance, given the selected model. Then estimated precision will be too small because the variance component for model selection uncertainty is missing.

• problems for inference, probably not for mere data description
• proper model selection is accompanied by a substantial amount of uncertainty
• bootstrap techniques can allow insights into model uncertainty
• choosing a model completely independent of the data has hidden costs in lack of reliability

1.8 Summary

• model selection includes scientific understanding: which models to include in the candidate sets and which not
• data dreding weakens inferences
• information-theoretic can be used to select a model
• multimodel inference: models are ranked and scaled to understand model uncertainty

Data analysis is taken to mean the entire integrated process of a pri- ori model specification, model selection, and estimation of parameters and their precision. Scientific inference is based on this process.

• databased selection of a parsimonious model is challenging
• rewards: valid inferences
• dangers: underfitting or overfitting, model selection bias and model selection uncertainty

Mapping the deforestation footprint of nations reveals growing threat to tropical forests

The authors provide a fine-scale representation of spatial patterns of deforestation associated with international trade. They find that many developed countries have increased the deforestation embodied in their imports Consumption patterns of G7 countries drive an average loss of 3.9 trees per person per year. The results emphasize the need to reform zero-deforestation policies through strong transnational efforts and by improving supply chain transparency, public–private engagement and financial support for the tropics.

Current situation

• deforestation is permanently increasing
  • driven by international trade
  • negative impact on global climate and biodiversity
• many developed/developing countries with net domestic forest gain but imported deforestation
• spatial distribution of deforestation embodied in imports not well known

Questions

• Which deforestation hotspots are driven by which consumer countries?
• Which forest ecosystems, tropical rain forests or other forest types are the top targets of global supply chains?

Results

• using a global supply chain model, high resolution maps- of deforestation footprints of various nations were built
  • Germany: cocoa in Ivory Coast and Ghana, coffee in Vietnam
  • Japan: cotton and sesame seed in Tanzania
  • China: timber and rubber in Indochina
  • USA: Cambodia, Madagascar, Liberia, Central America, Chile, Amazon through timber, rubber, beef, fruits, nuts
• many developed/developing countries increase their imported damage faster than their domestic mitigation
• tree loss per capita and year
  • G7: 4
  • USA: 8
  • Sweden 22: biomass for energy supply
• different usage requires different tree types with different impacts on biodiversity
• tropical: USA, Germany, Singapore, China, Russia have increasing net imports from all biomes, but rapidly increasing from tropics
• tropical and mangrove rainforest deforestation increased GDPs per capita in developed countries, trade patterns remained the same, leading to even more deforestation (except for Norway and Sweden)

Discussion

• maps can help countries to improve their deforestation footprint and its ecological impact (climate and biodiversity)
• to maintain net forest gains, G7 and China and India outsourced their deforestation with increasing tendency
• no subnational analysis included
• maps can be used by each country to think about their personal consumptional behavior and supply chains
• international strategies, collaboration of private and public sector, financial support for exporting countries, and transparent supply chains necessary

My comments

The 22 trees for Sweden are shocking given that only recently I watched a documentary praising Sweden for their smooth and successful transition to renewable energies.

Global maps of twenty-first century forest carbon fluxes

The authors introduce a geospatial monitoring framework that integrates ground Earth observation data to map annual forest-related greenhouse gas emissions and removels from 2001 till 2019. They estimate that global forests were a carbon sink of $-7.6\,$GtCO$_2$e yr$^{-1}$ ($-15.6+8.1$). The final goal is to support forest-specific climate mitigation with both local detail and global consistency.

Current situation

• land use change patterns change faster than modelled
• distinguishing anthropogenic from non-anthropogenic effects possible only by direct observation
• different approaches lead to very different global net forest fluxes (projects, models vs inventories, countries, etc).
  $\to$ forests’ role in climate mitigation unclear
  $\to$ discouraging to take transformational actions

What’s new

• transparent, independent and spatially explicit global system for monitoring collective impact of forest-related climate policies by diverse actors across multiple scales
• separation of sources from sinks

Global distribution of forest emissions and removals

• most uncertainty in global gross removals
• tropical forests with highest gross fluxes, highest net sinks in temporal and boreal forests

Fluxes for specific localities and drivers of forest change

• Brazilian amazon forest a net source, greater Amazon River Basin a net sink
• smaller Congo River basin six times higher net sink due to lower emissions

A flexible data integration framework

• three tiers of methods, parameters, and data sources with different complexity and accuracy
• results most sensitive to data sources

Forest fluxes in the global carbon budget

• results not comparable to other global estimates (net vs gross, forests vs all, all GHGs vs CO$_2$)
• no way here to distinguish between anthropogenic and non-anthropogenic effects (data not available on small scales)
• net CO$_2$ forest sink larger than in Global Carbon Project because emissions might not be completely captured here by the medium resolution satellite observations used to underpin the analysis

Limitations and future improvements

• data spatially detailed with temporal inconsistencies:
  • lack of consistent time-series of forest regrowth
  • lack of consistent time-series for global loss product
• for many forests required long-term inventories do not exist

Conclusions

• reduce deforestation is important
• mitiagtion effects of intact (middle-)old forests often underestimated
• maps better than tables

The global forest carbon monitoring framework introduced here, and the main improvements identified above, allow for efficient prioritization and evaluation of how data updates and improvements influence GHG flux estimates and their uncertainties.

My comments

As far as I understood, the authors seem to see the problem that there is no time to install detailed and long-term monitoring systems for forests, so they propose to synthesize available data by means of general standards.

Ideas

Giulia asked whether a 30mx30m resolution is rather necessary or hindering. What about an AIC for spatial resolution?

Carbon cycle in mature and regrowth forests globally

The authors compile the Global Forest Database (ForC) to provide a macropscopic overview of the C cycle in the world’s forests. They compute the mean and standard deviation of 24 flux and stock variables (no soil variables) for mature and regrown (age < 100 years) forests. C cycling rates decrease from tropical to temperate to boreal forests. The majority of flux variables, together with most live biomass pools, increased significantly with the logarithm of stand age.

1. Introduction

• forests photosynthesize 69 GtC/year, leading to being a C sink accounting for 29% of fossil fuel emissions (problem: deforestation)
• regrowth (= secondary) forests become increasingly important
• biomes: categories for different climate and vegetation
• NEP = GPP - $R_{\text{eco}}$: net ecosystem production = gross primary production - total ecosystem respiration
• biomass accumulation increases rapidly in young forests, followed by a slow decline to near zero in old forests

2. Methods and design

• synthesis of many existing databases with the goal of understanding how C cycle varies depending on location and stand age
• R scripts and manual edits
• unit dry organic matter converted to C by C=0.47 OM (IPCC, 2018)
• 4 biome types (tropical broadleaf, temperate broadleaf, temperate needleleaf, boreal needleleaf) and 2 age classes (young, mature)
• C budget assumed closed if mean of components summed to within one standard deviation of the aggregate variable
• effect of stand age tested by using mixed effects models
• logarithmic fit also due to lack of sufficient data to use more parameters

3. Review results and synthesis

• mature forests:
  • fluxes: tropical > temperate > boreal
  • NEP: no significant trend
  • mean stocks: tropical > temperate > boreal
  • max. stocks in temperate biomes
• young forests:
  • fluxes and stocks increase with $\log_{10}$ of age
  • fluxes: tropical > temperate > boreal
  • NEP: temperate > boreal

4. Discussion

• variation in NPP in mature forests less controlled by climate, more by moderate disturbance and $R_{\text{soil}}$ vs C inputs
• organic layer (OL) highest in boreal forests due to slow decomposition
• NEP increases for first 100 years
• future forest C cycling will shape climate (Song et al. 2019, Schimel et al. 2015)
• ForC contains ground data for variables that cannot be measured (at least directly) remotely, such as respiration fluxes

5. Conclusions

• loss of biomass from mature forests cannot be recovered on time scales relevant for mitigating climate change
• conservation of mature forests most important

Ideas

By definition, future projections extend our existing observations and understanding to conditions that do not currently exist on Earth (Bonan and Doney 2018, Gustafson et al 2018, McDowell et al 2018). To ensure that models are giving the right answers for the right reasons (Sulman et al 2018), it is important to benchmark against multiple components of the C cycle that are internally consistent with each other (Collier et al 2018, Wang et al 2018).

What about applying information partitioning to ForC?

Calculating the effective permeability of sandstone with multiscale lattice Boltzmann/finite element simulations

When scaling up from microscale to macroscale, often one is not interested in a single global value such as a mean only, but rather in the variation of a continuum variable. The authors define a representative elementary volume (REV) at the microstate-macrostate boundary. Inside the REVs, a lattice Boltzmann (LB) method is used to compute the microdynamics. The result per REV is then used on global scale to solve global dynamics by a finite elements (FE) method.

I see a close link to Dynamic upscaling of decomposition kinetics for carbon cycling models. In the referenced paper, Eq. (16) describes the macroscale dynamics corresponding to FE. On macroscale the parameters $\sigma^2_{C_s}$, $\sigma^2_{C_b}$, $C’_s C’_b$, etc. are used and those can be obtained from microscale dynamics corresponding to LB.

Quantifying entropy using recurrence matrix microstates

The authors introduce a complexity measure for nonlinear time series data that bases on the reccurence plot (RP) and the Shannon information entropy of its microstates. This complexity measure is easy and efficient to compute and approximates the maximum Lyapunov exponent of the data. It can also be used to discriminate between deterministic, chaotic, and stochastic data.

For $x_i\in\R^d$, $i=1,\ldots,K$, the RP is a matrix $\tens{R}=(R_{ij})_{ij}$ given by

\begin{equation} \nonumber R_{ij} = \begin{cases} 1,\, & |x_i-x_j| \lt \varepsilon,\newline 0,\, & |x_i-x_j| \geq \varepsilon, \end{cases} \end{equation} where $\varepsilon>0$ is called vicinity threshold. Diagonal structures of $1$s parallel to the main diagonal display recurrence patters and are signs for determinism. iThe idea here is now to fix a small natural number $N$, typically $N\in\{1,2,3,4\}$, and look at ($N\times N)$-submatrices of $\tens{R}$. A fixed number $\bar{N}$ of such structures is selected randomly. The total number of possible microstates is $N^\ast=2^{(N^2)}$ and with $P_i=n_i/\bar{N}$, where $n_i$ is the number of occurences of the specific microstate $i$, we get the entropy

\begin{equation} \nonumber S(N^\ast) = -\suml_{i=1}^{N^\ast} P_i\,\log P_i. \end{equation}

Although $N^\ast$ grows quickly as a function of $N$, usually just a small number of microstates are effectively populated. So, the effective set of microstates needed to compute adequately the entropy can be populated by just $\bar{N}$ random samples obtained from the recurrence matrix, and a fast convergence is expected. In general, we found that usually $\bar{N} \ll N^\ast$ for $N > 3$ such that $\bar{N} \sim 10,000$ is enough. This makes the method extremely fast even for moderate values of microstate sizes $N$. This observation also points out that a microstate size $N = 4$ is sufficient for many dynamical and stochastic systems.

The maximum entropy occurs when all microstates are equally likely, i.e. $P_i=1/N^\ast$, and is given by

\begin{equation} \nonumber S(N^\ast) = N^2\,\log2 . \end{equation}

The closer $S(N)$ is to $S(N^\ast)$, the more stochastic are the data.