Disentangling vegetation physiological responses under extreme drought in the Amazon Rainforest: A multispectral remote sensing approach with insights from ET, SIF, and VOD

Extreme drought has profound effects on global vegetation, shaping carbon and water cycles and drawing significant research attention. Physiological responses and structural adaptations are two main aspects when vegetation dealing with drought. Traditional remote sensing methods, relying on indicators like Leaf Area Index (LAI), Solar-Induced Fluorescence (SIF), and Near Infrared reflectance of vegetation (NIRv), face challenges in disentangling mixed signals and capturing fine-scale physiological changes. To address this issue, we proposed a multi-spectral remote sensing approach to construct models that disentangle remote sensing signals only representing vegetation's physiological response to drought. To achieve that, two separate random forest models were constructed using vegetation structural variables and hydro-meteorological variables to predict total and structural components of functional anomalies, quantified using SIF, Evapotranspiration (ET), and Vegetation Optical Depth (VOD) ratio. Subsequently, model residuals were calculated from the two models and used to disentangle the physiological component in observed remote sensing signals. The results in Amazon rainforest revealed that the physiological component explained the majority of functional anomalies during drought, with the physiological contributions of photosynthesis, transpiration, and water regulation functions accounting for 74.1%, 64.2%, and 71.8% of the anomalies in wet regions, and 67.7%, 62.6%, and 66.2% in dry regions, respectively. Attribution analysis indicated that regional hydro-meteorological conditions and vegetation types contributed to shaping the spatial patterns of vegetation physiological responses to drought, explaining 75.28% and 82.17% of the spatial variability in the physiological components during drought development and recovery phases. Structural equation modeling further elucidating causal pathways linking key environmental drivers to these physiological responses. The uncertainty of model predictions was quantified using the leave-one-out approach, yielding interquartile ranges of 0.72, 0.41, and 0.82 for the physiological component proportions of the three functional variables. This research disentangles physiological and structural responses with finer spatial and temporal resolution, providing a clearer view of vegetation dynamic changes and adaptation mechanisms. These findings emphasize the value of multi-spectral remote sensing in understanding vegetation functions under extreme drought conditions, offering a more detailed and accurate representation of vegetation dynamics.