Polarization of dryland vegetation response to spring flood

Dryland vegetation plays a pivotal role in the global ecosystem and is extraordinarily sensitive to water pulses, yet it is experiencing increasingly severe flooding. Floods have both significant beneficial and detrimental effects that may trigger extreme positive or negative responses in dryland vegetation compared to non-flooded areas, a phenomenon we call polarization, which threatens the stability of ecosystems. However, the characters and drivers of the polarized responses of dryland vegetation to spring floods remain unclear. Taking Central Asia—a typical and important dryland region—as a case, this study developed a Vegetation Response Index (VRI) and a threshold-based Polarization Index (PI) based on satellite remote sensing, and then integrated interpretable machine learning to investigate the drivers of vegetation polarization responses. The results indicate that vegetation in flood-affected areas exhibited much stronger positive and negative VRI compared to non-flooded areas, with an average of 28.6 % (positive) and 25.8 % (negative) of the pixels exceeding the polarization threshold. This polarization phenomenon was validated using two different vegetation indices (EVI and LAI) based on polarization magnitude, event occurrence rate, and consistency, all providing strong evidence for the polarized growth patterns of dryland vegetation after spring floods. The influencing factors have high interpretability for both positive and negative polarization (88.8 % and 90.2 % accuracy), with climatic factors being the determinant of whether polarization occurs. The key factors present three patterns of effect on polarization: intersection, threshold and range. These patterns characterize the environmental and climate conditions most likely to lead to polarization, explaining why some plants were able to adapt to flooding while others are succumbed. This study identifies and analyzes the polarized response of dryland vegetation to floods, providing vital insights for predicting and managing ecosystem vulnerability amidst increasing flood events.