Interpretable Machine Learning Reveals the Crucial Role of Water Availability in Regulating Thermal Optimality of Terrestrial Ecosystems

Abstract Gross primary productivity (GPP) is the total carbon dioxide plants fix in terrestrial ecosystems through photosynthesis. Air temperature directly influences photosynthesis and, consequently, ecosystem‐level GPP. However, air temperature is not the sole driver of GPP; ecosystem‐level GPP is a complex, multidimensional phenomenon where soil and atmospheric water availability play crucial roles. This study employs process‐based interpretable machine learning to investigate the thermal optimality of terrestrial ecosystems multidimensionally and evaluate the role of water availability in regulating thermal behavior and productivity. Our innovative data‐driven approach transcends traditional photosynthesis‐temperature response curves, visualizing the controlling effects of water availability in a three‐dimensional temperature‐moisture‐productivity space. We analyze 112,683 daily data samples of carbon, water, and energy fluxes alongside auxiliary micrometeorological variables from 108 eddy‐covariance sites across North America. Our multifaceted, observation‐driven approach quantifies the coupled influence of water availability and temperature on productivity. Findings highlight the critical role of long‐term ecosystem wetness and daily water availability in shaping terrestrial ecosystems' thermal behavior and optimality. Arid ecosystems tend to reach their optimum productivity at lower temperatures, with water availability as the primary productivity driver. Conversely, air temperature is the main productivity driver in wet ecosystems, accompanied by higher values for optimum temperature. Additionally, we observe an increasing air temperature trend in North America, which could cause a decline in productivity. However, thermal acclimation may counteract or mitigate this process.