Exploring the Dominant Roles of Vapor Pressure Deficit and Soil Moisture on Vegetation Productivity Using Explainable Machine Learning

Authors

Lingcheng Li¹* (lingcheng.li@pnnl.gov), Yilin Fang¹, Shiqin Xu², Zhonghua Zheng³, Nate McDowell^1,4, Jeffrey Chambers^5,6, Ruby Leung¹

Institutions

¹Pacific Northwest National Laboratory, Richland, WA; ²College of Geography and Environmental Science, Northwest Normal University, Lanzhou, China; ³Department of Earth and Environmental Sciences, The University of Manchester, Manchester, UK; ⁴School of Biological Sciences, Washington State University, Pullman, WA; ⁵Lawrence Berkeley National Laboratory, Berkeley, CA; ⁶Department of Geography, University of California–Berkeley, CA

URLs

• https://ngee-tropics.lbl.gov/

Abstract

With the increasing frequency and severity of droughts projected under future climate scenarios, the impact of drought on vegetation productivity is expected to intensify. Drought stress on ecosystem production is commonly characterized by low soil moisture (SM) and high atmospheric water demand (i.e., vapor pressure deficit; VPD). The relative dominance of VPD versus SM on vegetation production is still debatable. This study presents an explainable, machine learning approach to disentangling the dominant role of VPD and SM in vegetation gross primary production across global ecosystems, with a particular focus on tropical regions such as the Amazon basin, using multisource datasets. The findings reveal that, globally, vegetation productivity in most landscapes is dominated by SM rather than VPD. Moreover, the relative importance of VPD and SM varies with climate and plant functional types. The study provides a large-scale benchmark for modeling the drought impacts on terrestrial ecosystems in Earth system models like E3SM, with vital implications for understanding global water, carbon, and energy cycles in the face of climate change.