Soil Evaporation Stress Determines Soil Moisture‐Evapotranspiration Coupling Strength in Land Surface Modeling

Abstract

Model-based estimates of soil moisture (SM)-evapotranspiration (ET) coupling strength (ρ) vary widely and are prone to bias. Here we apply numerical modeling and remote sensing to identify the process-level source of modeled ρ bias with the goal of improving the fidelity of current Earth system models. Results illustrate that modeled ρ is most strongly determined by soil evaporation (E) stress, and (generally positive) ρ modeling bias is attributable to the oversimplification of soil texture impacts on E stress. Based on new remotely sensed estimates of ρ, we demonstrate that removing ρ bias via a single optimized E stress parameter leads to improved ET accuracy and resolves a well-known modeling bias in the partitioning of ET into E and T. As such, we highlight the importance of the stress function relating E and SM and its central role in regulating land-atmosphere coupling processes impacting local climate. Plain Language Summary The interaction of soil moisture (SM) and evapotranspiration (ET) has been identified as a key process for accurately capturing land-atmosphere coupling. However, recent work suggests that the representation of SM-ET coupling in existing Earth system models (ESMs) is biased and can reduce the reliability of ESM-projected climate change scenarios. Based on remote sensing and land surface modeling, we identify soil evaporation (E) stress as the key contributor to SM-ET coupling uncertainty, even in areas where transpiration dominates total ET, and note that the representation of E stress is typically oversimplified in ESMs. Therefore, our analysis suggests that a more realistic representation of land E stress should be a priority for future ESM development.

Can Surface Soil Moisture Information Identify Evapotranspiration Regime Transitions?

“…Likewise, Crow et al (2015) combine multisensory soil moisture and evapotranspiration products to estimate the overall coupling strength of evapotranspiration and soil moisture, which implicitly reflects the occurrence frequency of water-limited regimes at a given location. This method is further employed for diagnosing numerical weather forecasting systems (Crow et al, 2020), land surface models (Dong et al, 2020), and Earth system models (Dong et al, 2022). Recently, demonstrate that L-band retrieved soil moisture drydown patterns can be decoded to quantify daily evapotranspiration regime transitions.…”

An Enhanced MOD16 Evapotranspiration Model for the Tibetan Plateau During the Unfrozen Season

“…This study is also based on empirical fitting relationships. Furthermore, empirical formulations of r s s have developed to estimate evaporation from bare soil in LSMs for long‐term ET simulation with large spatial scale (Chang et al., 2019; Dong et al., 2020; Lu et al., 2020; J. Tang & Riley, 2013; Yang et al., 2011). Although the MOD16‐STM algorithm based on empirical parameters performs better at five independent grassland sites, it is still necessary to improve the proposed algorithm by considering the physical processes of soil evaporation and with the help of more observations to enhance the performance of the model in the future studies.…”

“…This study is also based on empirical fitting relationships. Furthermore, empirical formulations of r s s have developed to estimate evaporation from bare soil in LSMs for long-term ET simulation with large spatial scale (Chang et al, 2019;Dong et al, 2020;Lu et al, 2020;J. Tang & Riley, 2013;.…”

Observed Landscape Responsiveness to Climate Forcing

“…This ultimately results in divergence of future projections of the land and atmosphere states (Berg et al., 2015; Berg & Sheffield, 2018; Guo et al., 2006). Land surface model outputs and observation‐driven reanalysis data will have prescribed linkages between energy fluxes and their environment which may both contribute to these model differences as well as confound any analysis attempting to gain fundamental insights into natural land surface responsiveness from these models (Dong et al., 2020). As such, in this study we use only observations for two main reasons.…”