Widespread and complex drought effects on vegetation physiology inferred from space

Soil moisture droughts are increasing globally in both duration and intensity, posing a significant threat to terrestrial ecosystems. Droughts affect vegetation functioning by increasing the risk of carbon starvation and hydraulic failure, potentially leading to plant mortality. Because terrestrial vegetation plays a crucial role in regulating carbon and water fluxes at the Earth's surface, plant drought responses have important feedbacks to the climate system and may exacerbate global warming. Understanding these responses requires disentangling the effects on vegetation structure (e.g., leaf area index, LAI) and physiology (e.g., stomatal conductance, photosynthesis). While satellite-based estimates of vegetation structure (e.g., using greenness indices or LAI) are readily available, direct observations of vegetation physiology at large spatial scales have been limited. This study addresses this gap by using a combination of advanced remote sensing data and machine learning techniques to investigate the physiological response of vegetation to drought globally. This integrated approach allows for a more comprehensive understanding of the complex interactions between drought, vegetation, and climate.

Previous research has extensively utilized satellite-based vegetation greenness indices and LAI products to study vegetation responses to drought. However, these studies primarily focus on structural changes and often implicitly include physiological changes. Direct assessment of vegetation physiology, such as maximum carboxylation rate and stomatal conductance, has mainly been limited to site-level studies. Consequently, there's a need for improved methods to monitor and quantify large-scale physiological responses. Recent advancements in satellite remote sensing, particularly the availability of global solar-induced chlorophyll fluorescence (SIF) data from TROPOMI, offer new opportunities to address this limitation. SIF, along with land surface temperature (LST) and vegetation water content (VOD) derived from microwave remote sensing, provide valuable indicators of photosynthetic activity, evapotranspiration (ET), and vegetation hydraulics respectively. Combining these data sources can improve our ability to separate and understand the impacts of drought on both vegetation structure and physiology.

This study uses a multi-faceted approach combining remotely sensed data and modeling. The researchers use data from March 2018 to October 2021 at 8-daily temporal and 0.25° spatial resolution. The data includes: * **TROPOMI SIF:** Used as an indicator of ecosystem photosynthesis. The researchers use relative SIF (SIFrel) to account for the effects of photosynthetically active radiation and sun-view angular variability. * **MODIS LST:** Used to estimate ET via a simplified surface energy balance (SSEB) model. * **AMSR2 VOD:** Used to assess vegetation hydraulics, specifically the ratio of midday and midnight VOD (VOD ratio) to monitor ecosystem hydraulics. * **MODIS LAI and NIRv:** Used as proxies for vegetation structure. * **ERA5-Land reanalysis:** Used to provide hydrometeorological data (temperature, radiation, VPD, precipitation, soil moisture). The study employs a random forest-based machine learning technique to separate the physiological components of SIFrel, ET, and VOD ratio from structural changes captured by LAI. Structural response is defined as the variability explained by LAI, while physiological response is attributed to the variability explained by hydrometeorological variables. This process assumes that LAI captures all relevant structural changes, and physiological responses are primarily driven by hydrometeorological factors. The Soil Canopy Observation of Photochemistry and Energy flux (SCOPE) model is further used to simulate vegetation drought responses and validate the findings from the observation-based analysis. The SCOPE model considers vegetation physiology and produces radiance spectra associated with vegetation functioning and biophysical properties. Aridity is used to classify regions as wet or dry, further distinguishing drought responses across different climate zones. An attribution analysis using SHAP values is conducted to identify the relative importance of different drivers in explaining spatial variations of the physiological components.

The study reveals contrasting vegetation drought responses between wet and dry regions. In dry regions, LAI and NIRv show below-normal anomalies due to water stress, while wet regions often exhibit positive anomalies, possibly related to increased photosynthetic capacity under sunny drought conditions. SIF shows continuous decreases preceding drought peaks, stronger in dry regions. SIFrel shows similar trends but with weaker decreases in wet regions. Midday and midnight VOD anomalies both decrease during drought, with a greater decrease in midnight VOD in dry regions resulting in positive VOD ratio anomalies, indicating stomatal closure. ET anomalies are negative in dry regions and positive in wet regions. The strongest reductions in vegetation variables occur one time step after the soil moisture minimum. The study finds that overall and physiological patterns of vegetation anomalies are largely similar. Physiological changes explain 60–97% of overall functional drought responses, with the strongest downregulation observed in sub-humid and semi-arid areas. Physiological changes emerge earlier in dry regions, leading to severe decreases in SIFrel and ET while increasing VOD ratio (indicating stomatal closure). The attribution analysis reveals aridity and tree cover fraction as the most important controls for spatial variations of vegetation physiology during drought development, alongside meteorological anomalies such as radiation (for SIFrel), precipitation (for ET), and VPD (for VOD ratio). During drought recovery, soil moisture and VPD become the dominant controls. SCOPE simulations support the observation-based findings, showing that downregulation of stomatal conductance and light use efficiency are key determinants of vegetation response to water stress.

The findings highlight the dominance of physiological changes in shaping vegetation responses to drought, particularly in water-limited regions. The study successfully separates the influence of vegetation structure and physiology on drought responses using a novel approach combining remote sensing and machine learning. The contrasting responses between wet and dry regions emphasize the importance of considering climate context when assessing drought impacts on vegetation. The close agreement between observations and model simulations enhances the credibility of the results and provides a mechanistic understanding of the underlying processes. These findings have significant implications for improving land-climate models by better representing vegetation physiology and its feedbacks to the climate system.

This study presents a significant advance in understanding global vegetation drought responses by successfully disentangling the effects of physiology and structure. The findings emphasize the importance of physiological downregulation, especially in semi-arid regions, in driving vegetation responses to drought. The use of multiple remote sensing data streams and a machine learning approach strengthens the robustness of the results. Future research should focus on incorporating these findings into Earth system models to improve their ability to simulate land-climate feedbacks accurately and refine drought impact projections. This might include further investigations into specific plant functional traits and their role in determining drought sensitivity across different ecosystems.

The study acknowledges potential limitations. The method simplifies vegetation structural changes and cannot fully separate the direct meteorological effects on ET from physiological changes. The random forest model used might not accurately predict extreme values, potentially affecting the assessment of structural changes. Finally, uncertainties in the observational datasets (LAI, SIF, ET, VOD) and reanalysis data could impact the results. The SCOPE model does not account for drought stress through soil moisture deficits and their legacy effects, which may impact the accuracy of drought recovery simulations. The spatial resolution of the data also limits the study's ability to resolve fine-scale variability in vegetation responses.