Search for: All records

Drought is one of the most destructive and expensive natural disasters, severely impacting natural resources and risks by depleting water resources and diminishing agricultural yields. Under climate change, accurately predicting drought is critical for mitigating drought-induced risks. However, the intricate interplay among the physical and biological drivers that regulate droughts limits the predictability and understanding of drought, particularly at a subseasonal to seasonal (S2S) time scale. While deep learning has demonstrated the potential to address climate forecasting challenges, its application to drought prediction has received relatively less attention. In this work, we propose a new dataset, DroughtSet, which integrates relevant predictive features and three drought indices from multiple remote sensing and reanalysis datasets across the contiguous United States (CONUS). DroughtSet specifically provides the machine learning community with a new real-world dataset to benchmark drought prediction models and more generally, time-series forecasting methods. Furthermore, we propose a spatial-temporal model SPDrought to predict and interpret S2S droughts. Our model learns from the spatial and temporal information of physical and biological features to predict three types of droughts simultaneously. Multiple strategies are employed to quantify the importance of physical and biological features for drought prediction. Our results provide insights for researchers to better understand the predictability and sensitivity of drought to biological and physical conditions. We aim to contribute to the climate field by proposing a new tool to predict and understand the occurrence of droughts and provide the AI community with a new benchmark to study deep learning applications in climate science. 

Abstract Evapotranspiration (ET) plays a critical role in water and energy budgets at regional to global scales. ET is composed of direct evaporation (E) and plant transpiration (T) where the latter is regulated via stomatal conductance (g_sc), which depends on a multitude of plant physiological processes and hydrometeorological forcings. In recent years, significant advances have been made toward estimatingg_scusing a variety of models, ranging from relatively simple empirical models to more complex and data‐intensive plant hydraulic models. Using machine learning (ML) and eddy covariance flux tower data of 642 site years across 84 sites distributed across 10 land covers globally, here we show that structural constraints inherent in current empirical and plant hydraulic models ofg_sclimit their effectiveness for predicting ET. These constraints also prevent the models from fully utilizing the available hydrometeorological data at eddy covariance sites. Even if theseg_scmodels are calibrated locally, structural simplifications inherent in them limit their capability to accurately captureg_scdynamics. In contrast, a ML approach, wherein the model structure is learned from the data, outperforms traditional models, thus highlighting that there still is significant room for improvement in the structure of traditional models for predicting ET. These results underscore the need to prioritize improvements ing_scmodels for more accurate ET estimation. This, in turn, will help reduce uncertainties in the assessments of plants' role in regulating the Earth's climate. 

Abstract Atmospheric dryness (i.e., high vapor pressure deficit, VPD), together with soil moisture stress, limits plant photosynthesis and threatens ecosystem functioning. Regions where rainfall and soil moisture are relatively sufficient, such as the rainfed part of the U.S. Corn Belt, are especially prone to high VPD stress. With globally projected rising VPD under climate change, it is crucial to understand, simulate, and manage its negative impacts on agricultural ecosystems. However, most existing models simulating crop response to VPD are highly empirical and insufficient in capturing plant response to high VPD, and improved modeling approaches are urgently required. In this study, by leveraging recent advances in plant hydraulic theory, we demonstrate that the VPD constraints in the widely used coupled photosynthesis‐stomatal conductance models alone are inadequate to fully capture VPD stress effects. Incorporating plant xylem hydraulic transport significantly improves the simulation of transpiration under high VPD, even when soil moisture is sufficient. Our results indicate that the limited water transport capability from the plant root to the leaf stoma could be a major mechanism of plant response to high VPD stress. We then introduce a Demand‐side Hydraulic Limitation Factor (DHLF) that simplifies the xylem and the leaf segments of the plant hydraulic model to only one parameter yet captures the effect of plant hydraulic transport on transpiration response to high VPD with similar accuracy. We expect the improved understanding and modeling of crop response to high VPD to help contribute to better management and adaptation of agricultural systems in a changing climate. 

Abstract The spatio‐temporal variation of stomatal conductance directly regulates photosynthesis, water partitioning, and biosphere‐atmosphere interactions. While many studies have focused on stomatal response to stresses, the spatial variation of unstressed stomatal conductance remains poorly determined, and is usually characterized in land surface models (LSMs) simply based on plant functional type (PFT). Here, we derived unstressed stomatal conductance at the ecosystem‐scale using observations from 115 global FLUXNET sites. When aggregated by PFTs, the across‐PFT pattern was highly consistent with the parameterizations of LSMs. However, PFTs alone captured only 17% of the variation in unstressed stomatal conductance across sites. Within the same PFT, unstressed stomatal conductance was negatively related to climate dryness and canopy height, which explained 45% of the total spatial variation. Our results highlight the importance of plant‐environment interactions in shaping stomatal traits. The trait‐environment relationship established here provides an empirical approach for improved parameterizations of stomatal conductance in LSMs.