Hybrid AI Model Refines Tracking of Land-to-Atmosphere Water Movement
Researchers integrate machine learning into a physical framework to better distinguish between soil evaporation and plant transpiration.
A multi-institutional research team has developed a hybrid modeling approach that combines a physical framework with machine learning to more accurately track evapotranspiration. This process, which governs the movement of water from land to the atmosphere, is a critical driver of the Earth's water cycle.
According to the U.S. Department of Energy, the new model integrates machine learning into the established Penman–Monteith modeling framework. By using AI to estimate surface conductance values while maintaining strict physical constraints, the system can more effectively partition total water loss into its two primary components: soil evaporation and plant transpiration. To validate the approach, researchers evaluated the model using data from 47 National Ecological Observatory Network (NEON) sites, spanning a diverse range of vegetation types and environmental conditions.
The Challenge of Water Partitioning
Evapotranspiration is a fundamental component of global climate systems, significantly influencing the intensity and duration of heat waves and droughts. However, traditional computer models have historically struggled to separate the water lost through soil evaporation from the water released by plants via transpiration. This inability to distinguish between the two sources creates a gap in understanding that limits the precision of Earth system simulations. Without accurate partitioning, it is difficult for scientists to determine exactly how different landscapes respond to moisture stress.
Implications for Climate Resilience
Improving the accuracy of water flux predictions allows scientists to better understand the specific roles that radiation, soil moisture, and vegetation play in ecosystem water use. By bridging the gap between purely physics-based models and purely ML-driven models, this hybrid approach provides a more robust tool for Earth system simulations. These improvements are critical for providing the data necessary to predict and mitigate the impacts of climate-driven events. As severe droughts and heat waves become more frequent, the ability to precisely model water movement helps communities better prepare for extreme weather risks.
Future Outlook
While the hybrid model has demonstrated superior performance across the 47 NEON sites, the next phase of research will likely focus on scaling these predictions to a global level. Scientists continue to investigate how these refined conductance estimates can be integrated into larger climate models to improve long-term weather forecasting. Further verification is expected as the model is applied to a wider array of unseen environments to ensure consistent stability across all global biomes.