Interpreting Vegetation-Climate Interaction with Explainable Artificial Intelligence (XAI)
Conferences
Overview
Overview
abstract
Climate change has considerably altered vegetation dynamics, affecting Earth’s biological systems. Elevated temperatures have contributed to changes in vital vegetation dynamics, such as high latitude greening, earlier leaf unfolding and delayed leaf senescence, with substantial implications for ecosystem functioning. These effects are particularly evident in high-latitude regions, leading to feedback responses such as altered surface albedo, energy budget, and plant-pollinator interactions.
This study aims to understand environmental forcings on vegetation dynamics on a global scale. We trained Extreme Gradient Boosting (XGBoost) models to forecast the Leaf Area Index (LAI) using a range of environmental covariates, including cloud cover, precipitation rate, temperature, frequency of wet days, incoming solar radiation, and soil moisture. Using the 1982-2014 half-degree, monthly, global environmental data, our model forecasts the 2015 LAI estimates.
Our LAI emulator achieved a Pearson correlation (R^2) of 0.63 and an LAI Root Mean Squared Error (RMSE) of 0.91. A subsequent SHAP (SHapley Additive exPlanations) Tree Explainer analysis revealed temperature, precipitation, and soil moisture as the most critical factors in predicting LAI. We will present the specific impacts of these factors, including situations where they may increase or decrease LAI. This work underscores the importance of climate factors in predicting global vegetation dynamics, contributing to our understanding of vegetation-climate interaction and climate change impacts.
