Volume 225, Issue 1 p. 105-112
Tansley insight
Free Access

Constraining estimates of terrestrial carbon uptake: new opportunities using long-term satellite observations and data assimilation

William K. Smith

Corresponding Author

William K. Smith

School of Natural Resources and the Environment, University of Arizona, Tucson, AZ, 85719 USA

Author for correspondence:

William K. Smith

Tel: +1 970 449 2949

Email: [email protected]

Search for more papers by this author
Andrew M. Fox

Andrew M. Fox

School of Natural Resources and the Environment, University of Arizona, Tucson, AZ, 85719 USA

Search for more papers by this author
Natasha MacBean

Natasha MacBean

Department of Geography, Indiana University, Bloomington, IN, 47405 USA

Search for more papers by this author
David J. P. Moore

David J. P. Moore

School of Natural Resources and the Environment, University of Arizona, Tucson, AZ, 85719 USA

Search for more papers by this author
Nicholas C. Parazoo

Nicholas C. Parazoo

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 91109 USA

Search for more papers by this author
First published: 12 July 2019
Citations: 41

Abstract

Summary

The response of terrestrial carbon uptake to increasing atmospheric [CO2], that is the CO2 fertilization effect (CFE), remains a key area of uncertainty in carbon cycle science. Here we provide a perspective on how satellite observations could be better used to understand and constrain CFE. We then highlight data assimilation (DA) as an effective way to reconcile different satellite datasets and systematically constrain carbon uptake trends in Earth System Models. As a proof-of-concept, we show that joint DA of multiple independent satellite datasets reduced model ensemble error by better constraining unobservable processes and variables, including those directly impacted by CFE. DA of multiple satellite datasets offers a powerful technique that could improve understanding of CFE and enable more accurate forecasts of terrestrial carbon uptake.

Contents
  Summary 105
I. Introduction 105
II. Satellite observations can constrain direct and indirect effects of CFE 106
III. Data assimilation 109
IV. Conclusions 110
  Acknowledgements 110
  References 110

I. Introduction

Terrestrial ecosystems provide a critical service to humanity by absorbing nearly half of anthropogenic CO2 emissions each year, limiting the rate and overall magnitude of climate warming (Le Quéré et al., 2018). This important ecosystem service probably depends on the positive responses of photosynthesis, namely gross primary productivity (GPP), to increases in atmospheric [CO2], namely the CO2 fertilization effect (CFE) (Norby et al., 2005; Cooley et al., 2018). CFE can influence vegetation growth through both direct and indirect pathways, making the full quantification of CFE challenging (Fig. 1). CFE can increase light use efficiency (LUE) and/or water use efficiency (WUE) through well-known physiological pathways, leading to direct increases in photosynthetic uptake (Long et al., 2004; Norby et al., 2016). CFE can also increase leaf area index (LAI) and in turn the fraction of absorbed photosynthetically active radiation (fAPAR), leading to indirect increases in photosynthetic uptake (Norby and Zak 2011; Tor-ngern et al., 2015). Adding complexity, the long-term persistence of these CFE pathways depends on interactions with multiple biogeochemical (e.g. nutrient limitation; Luo et al., 2004; Norby et al., 2010; Wieder et al., 2015) and biophysical (e.g. temperature and moisture stress; Reich et al., 2014; Gray et al., 2016; Buermann et al., 2018) constraints on photosynthesis. Indeed, recent evidence suggests we may be approaching a shift from a period dominated by the positive effects of CFE on GPP to a period dominated by the combined negative effects of nutrient limitation and moisture stress on GPP (Peñuelas et al., 2017; Green et al., 2018). Complete and accurate quantification of CFE requires approaches capable of quantifying both the direct (e.g. LUE and WUE) and the indirect (e.g. LAI and fAPAR) CFE pathways and their interactions with multiple biophysical and biogeochemical factors across global ecosystems and over decades.

Details are in the caption following the image
Schematic of the pathways by which CO2 fertilization effects (CFE) can increase gross primary productivity (GPP) and the potential ways satellite observations could be combined to constrain the CFE pathways. We define CFE pathways to include increases in light use efficiency (LUE), increases in water use efficiency (WUE), and increases in the fraction of photosynthetically active radiation (fAPAR). Satellite indices include leaf area index (LAI) and fAPAR, land surface temperature (LST), vegetation optical depth (VOD) and solar-induced Chl fluorescence (SIF). We show different combinations of these satellite records and indicate their potential to globally constrain (green ticks), regionally constrain (yellow ticks) or fail to constrain (red crosses) a given CFE pathway. Regional constraints (yellow ticks) are most often limited by atmospheric effects and/or signal saturation in dense canopies, such as tropical forest ecosystems.

Satellite remote sensing provides an important integrated perspective on ecosystem responses to changing biogeochemical and biophysical vegetation growth constraints, and has helped shape our current understanding of CFE (Lu et al., 2016; Schimel et al., 2015; Smith et al., 2016; Zhu et al., 2016). Yet, our understanding of what exactly different satellite observations represent, and how to best utilize long-term records to constrain CFE, remains an area of active research and debate (De Kauwe et al., 2016; Smith et al., 2016; He et al., 2017). This is partly because our current understanding has been surpassed by the rapid availability of diverse satellite records that span a wide spectral range, including visible and near-infrared (optical), thermal-infrared and microwave (Guan et al., 2016; Stavros et al., 2017). Here we focus on long-term records from each of these spectral ranges, including optically based LAI, fAPAR, optically based solar-induced fluorescence (SIF), thermal-infrared-based land surface temperature (LST) and microwave-based vegetation optical depth (VOD) (Fig. 2). A main strength of these diverse and independent records is that each offers unique global-scale insight into CFE, mediated by multiple interacting biophysical and biogeochemical factors. Satellite observations can thus complement experimental frameworks, such as the foundational free-air [CO2] enrichment (FACE) experiments, which isolate CFE, but often lack concomitant warming or nutrient treatments, are necessarily limited to one location or ecosystem, and are only rarely active for decades (Norby et al., 2016; Obermeier et al., 2016; Cooley et al., 2018; Walker et al., 2019).

Details are in the caption following the image
A timeline of satellite observations of leaf area index (LAI) and fraction of photosynthetically active radiation (fAPAR), land surface temperature (LST), vegetation optical depth (VOD) and solar-induced fluorescence (SIF). Observation timelines are provided for context and are not meant to provide a comprehensive overview of all available sensors. This timeline clearly demonstrates the availability of diverse, multidecadal satellite observation records that are rapidly increasing in spatiotemporal resolution.

The key questions addressed here are: ‘How can ecologists interpret the diversity of satellite observational records available?’; and ‘How can ecologists best extract and integrate information across the diversity of satellite observational records to systematically constrain CFE?’. We argue that long-term satellite observations remain an under-exploited data source in understanding CFE. We illustrate how independent, multidecadal satellite observation records can be integrated in new ways to quantify different aspects of CFE at high spatiotemporal resolution for the entire terrestrial biosphere. We then demonstrate the use of model data assimilation (DA) as an effective way to utilize these independent satellite data streams, that is the systematic combination of information from observations and models to achieve greater process understanding than possible from observations or models alone (Fox et al., 2009; Zobitz et al., 2011).

II. Satellite observations can constrain direct and indirect effects of CFE

Multiple existing long-term satellite observation records have the potential to provide new insights into CFE (Figs 1, 2). Existing merged, long-term LAI and fAPAR, LST and VOD records have been used recently to evaluate trends in long-term carbon cycling (e.g. Liu et al., 2011, 2015; Zhu et al., 2016). Merged, long-term SIF records are also becoming available to evaluate trends and variability in long-term carbon cycling (Joiner et al., 2013; Gentine & Alemohammad, 2018; Sun et al., 2018). We propose below new ways to integrate these independent, long-term observational data to systematically constrain the multiple CFE pathways.

LAI and fAPAR constraints

Satellite observations of reflectance in the optical wavelength range are used to estimate vegetation state factors LAI and fAPAR. Long-term satellite records indicate that growing season integrated LAI has significantly increased over 25–50% of the vegetated land area, and CFE has been suggested to explain 70% of this trend (Zhu et al., 2016). Satellite fAPAR and LAI data are very useful because they can be directly incorporated into satellite-based GPP algorithms (Smith et al., 2016), used to directly evaluate model outputs (Buermann et al., 2018) and/or directly assimilated to constrain model outputs (MacBean et al., 2016; Fox et al., 2018).

Yet, changes in LAI and fAPAR capture only the indirect CFE pathway (Fig. 1). CFE could also directly stimulate GPP through positive effects on LUE and/or WUE, leading to increases in GPP that may be undetectable by satellite LAI or fAPAR observations alone (Norby et al., 2005; De Kauwe et al., 2016). LAI and fAPAR satellite observations are also prone to saturation in forests with relatively dense canopies, and thus could underestimate even indirect CFE for these critical ecosystems, including tropical forests (Liu et al., 2013; Smith et al., 2016). Overcoming these limitations requires integration of independent satellite proxies that are sensitive to changes in LUE and/or WUE and less prone to saturation. Fortunately, as discussed below, new opportunities exist for such joint constraint of CFE using independent satellite proxies.

LUE constraints

LUE is generally defined as the amount of vegetation production (GPP) per unit radiation absorbed (fAPAR) (Monteith, 1977; Running et al., 2004). Rising CO2 increases LUE directly by increasing the velocity of carboxylation and by reducing photorespiration (Farquhar et al., 1980; Long et al., 2004). For forest ecosystems in the longest running FACE experiments, Oak Ridge and Duke Forest, enhanced [CO2] was found to drive predominantly large increases in LUE (Norby et al., 2005; Walker et al., 2019). LUE cannot be directly observed, but can be approximated by measuring changes in fAPAR and GPP. It follows then that satellite LAI and/or fAPAR estimates could be combined with independent records of vegetation production derived from satellite VOD and/or SIF to directly constrain LUE (Stavros et al., 2017; Fischer et al., 2019).

Long-term satellite VOD observations most directly track changes in the water content of above-ground vegetation, but they can also be applied to estimate changes in above-ground biomass due to the tight relationship between water content and biomass (Liu et al., 2015; Brandt et al., 2018; Konings et al., 2019). Long-term VOD datasets have been previously used to quantify multidecadal trends in above-ground biomass at the global scale (Liu et al., 2015) (Figs 2, 3). Recent work reinforces the near-linear relationship between VOD and above-ground biomass across a range of ecosystem types (Brandt et al., 2016, 2018). Thus, together satellite LAI and/or fAPAR estimates and VOD-based above-ground biomass estimates capture changes in above-ground biomass per unit light absorbed, a proxy for LUE (LUEVOD) that could be used to constrain this direct CFE pathway (Fig. 1).

Details are in the caption following the image
Global model-derived and satellite-derived carbon stock and flux estimates. Model outputs are from the Community Land Model (CLM 4.5 in CESM 1.2) and include above-ground carbon (AGCCLM) for July 2007 (a), and gross primary productivity (GPPCLM) for July 2007 (b). Corresponding satellite-derived estimates include AMSR-E VOD-derived above-ground carbon (AGCVOD) for July 2007 (c), and GOME-2 SIF for July 2007 (d).

Importantly, LUEVOD cannot capture shifts in below-ground C allocation, which could be prioritized when nutrients are the dominant constraint on GPP (Norby et al., 2004; DeLucia et al., 2005). Additionally, VOD estimates derived from high-frequency microwaves (e.g. NASA AMSR-E; Fig. 2) are still prone to saturation in dense canopies, thus limiting their ability to monitor changes in biomass across important tropical regions (Brandt et al., 2018). Looking forward, evidence suggests that more recent VOD records derived from relatively low-frequency microwaves (e.g. NASA SMAP, ESA SMOS; Fig. 2) are not as prone to saturation because they penetrate deeper into the canopy (Brandt et al., 2018). Progress on this frontier relies on the continued development and improvement of merged long-term records of VOD-derived above-ground biomass (Konings et al., 2019). New independent efforts, including the NASA Global Ecosystem Dynamics Investigation (NASA GEDI) – a high-resolution laser ranging system on the International Space Station (launched 2018) – will produce new high-resolution estimates of above-ground biomass, which will enable a more complete understanding of the VOD : above-ground biomass relationship, including a more complete quantification of uncertainty (Konings et al., 2019; Schimel et al., 2019).

Long-term satellite SIF observations are a direct measure of the light re-emitted from Chl during the light reactions of photosynthesis, and theory and observations show that SIF is correlated with photosynthetic activity in a dynamic fashion that is directly linked to APAR and LUE (Porcar-Castell et al., 2014; Yang et al., 2015; Smith et al., 2018; Magney et al., 2019) (Figs 2, 3). Thus, together satellite fAPAR and SIF estimates capture changes in photosynthesis per unit light absorbed (i.e. GPP/fAPAR), a proxy for LUE (LUESIF) that could be used to constrain this direct CFE pathway (Stavros et al., 2017; Schimel et al., 2019; Shiklomanov et al., 2019) (Fig. 1). LUESIF can be interpreted as a constraint on GPP, whereas LUEVOD more closely acts as a constraint on above-ground biomass, and thus together these estimates could be jointly applied to evaluate biomass allocation dynamics. Because SIF is a measurement of re-emitted light, it is less prone to signal saturation in dense canopies (Frankenberg & Berry, 2018), and thus has the potential to provide improved coverage of tropical forest ecosystems. More recent satellite efforts to measure SIF, including NASA OCO-2 and ESA TROPOMI, show high fidelity compared to ground observations (Köhler et al., 2018; Magney et al., 2019), providing increased confidence in the use of satellite SIF to accurately measure the timing and magnitude of GPP across diverse global ecosystems. The potential for satellite SIF observations as a constraint on CFE depends on continued development of merged, long-term SIF records with well-characterized estimates of uncertainty (Sun et al., 2018; Schimel et al., 2019); and improved logic for direct incorporation of SIF observations into models (Lee et al., 2015; MacBean et al., 2018; Norton et al., 2018; Gu et al., 2019; Schimel et al., 2019).

WUE constraint

WUE is generally defined as carbon uptake per unit water loss through transpiration (De Kauwe et al., 2013; Fisher et al., 2017). CFE can increase WUE directly because stomatal conductance (gs) has been observed to decrease in roughly inverse proportion to CFE, even as CO2 uptake remains roughly unchanged (De Kauwe et al., 2013). Long-term satellite records of thermal infrared (TIR) are available (Fig. 2), and can be used to estimate LST, which can be integrated with measurements of air temperature to infer changes in ecosystem evapotranspiration (ET) (Fisher et al., 2017; Schimel et al., 2019; Shiklomanov et al., 2019). Most directly, SIF-based GPP and LST-based ET estimates could be combined to constrain WUE by capturing changes in photosynthesis per unit water loss (i.e. GPP/ET) (Stavros et al., 2017; Schimel et al., 2019; Shiklomanov et al., 2019) (Fig. 1). Again, progress on this front will require continued development of merged, long-term SIF and LST observational records with well-characterized estimates of uncertainty (Fisher et al., 2017).

III. Data assimilation

DA provides a statistical framework to use observations to estimate model states and parameters, evaluate alternative model structures, and quantify and reduce uncertainties in model predictions. In particular, confronting ecosystem models with satellite observations using DA offers many benefits that could help constrain CFE, including: initialization of model states and parameters with high spatiotemporal frequency observations, which thus inform and constrain predictions (Dietze et al., 2018; Fox et al., 2018; Reichstein et al., 2019); implicit approximation of unobservable variables, for example LUE and WUE, that are constrained through process-based relationships within the model (Moore et al., 2008; Richardson et al., 2010; Fox et al., 2018); integration of multiple data streams at different spatial and temporal resolutions to provide constraints greater than the sum of individual data streams (Bacour et al., 2015; MacBean et al., 2016; Peylin et al., 2016); and systematic confrontation of models with observational data to drive cyclical and rapid model development (Parazoo et al., 2014; Scholze et al., 2017; Fischer et al., 2019; Reichstein et al., 2019).

DA is a powerful approach for integrating satellite data where there are many overlapping observations that inform CFE, but where gaps are introduced because of sensor limitations and/or where only some carbon pools can be credibly observed (Fig. 3). We can also reconcile alternative and sometimes conflicting estimates of CFE by simultaneously assimilating multiple independent satellite observation records using rigorous statistical methods to update model states. We highlight some of the key benefits of DA with an example using single point runs of the Community Land Model (CLM) 4.5 coupled with the Data Assimilation Research Testbed (DART) system (Fox et al., 2018) (Fig. 4). In this example, we assimilate simulated LAI and above-ground biomass observations using an Ensemble Adjustment Kalman Filter (EAKF). Assimilated LAI and above-ground biomass separately decreased model error by 40% and 29%, respectively, while assimilating these data jointly reduced model error by 51% (Fig. 4). By jointly assimilating LAI and above-ground biomass data, we introduce an emergent constraint of the amount of biomass produced per unit light absorbed (i.e. LUE), leading to the largest reduction in model ensemble error. Importantly, DA using EAKF as well as other common ensemble filters requires observational data with well-characterized uncertainty, thus reducing the likelihood of DA failure due to divergence between the model ensemble and the observations (Fox et al., 2018). Looking forward we suggest close collaboration between remote sensing and DA experts to rapidly improve our understanding of CFE and enable more accurate forecasts of terrestrial carbon uptake.

Details are in the caption following the image
Data assimilation (DA) of observational data decreases model ensemble error. DA of annual biomass observations (BIO) (a, b), DA of monthly leaf area index observations (LAI) (c, d), DA of both biomass and LAI observations (B&L) (e, f), and changes in root mean square error (RMSE) across these DA scenarios (g, h). The model ensemble without DA (Free) and the observational data (Truth) are shown across multiple panels for comparison. The DA system used here is based on single point runs of the Community Land Model (CLM) 4.5 forced with atmospheric variables from the Community Atmosphere Model (CAM), and coupled with the Data Assimilation Research Testbed (DART) system. Plots represent the ensemble mean ± 1 SD. Adapted from Fox et al. (2018).

IV. Conclusions

Here we show that independent, long-term satellite observation records exist and can be combined in new ways to more fully understand and quantify the CFE (Figs 1-3). We argue that an effective way to utilize these independent satellite observation records is through DA to bridge between the differences in observation uncertainty and spatial and temporal coverage of multiple independent satellite observation records (Fig. 3). We show that DA of multiple simulated observational data streams reduced ensemble model error by better constraining unobservable processes and variables, including those directly impacted by CFE (Fig. 4). Moving forward, we recommend DA of the identified combinations of satellite datasets (Fig. 1) to effectively constrain CFE, thus enabling more accurate forecasts of terrestrial carbon uptake.

Acknowledgements

We thank Dr Rich Norby for encouraging us to write this paper. WKS, AMF and DJPM acknowledge funding from NASA Terrestrial Ecosystems Grant 80NSSC19M0103. We also acknowledge support of the data assimilation analysis by DOE Regional and Global Climate Modeling DE-SC0016011 and NSF Macrosystems 1241851. Finally, we acknowledge the useful and stimulating discussions during the Integrating CO2 Fertilization Evidence Streams and Theory (ICOFEST) meeting September, 2018, part of the FACE model Data-Synthesis project funded by the US Department of Energy, Office of Science, Office of Biological and Environmental Research. A portion of this research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA. Support from the Earth Science Division MEaSUREs program is acknowledged. All rights reserved.