Flood-drought shifts monitoring on arid Xinjiang, China using a novel machine learning based algorithm

This study addresses the growing challenges of climate extremes and their impact on flood-drought shifts in Xinjiang, China, a region highly sensitive to climate variations. While existing classification models such as logistic regression (LR), support vector machines (SVMs), and geographically weighted logistic regression (GWLR) have been applied to spatial data, they exhibit limitations in handling spatial nonstationarity and balancing accuracy with interpretability. To fill this gap, we propose a novel least squares SVM (LSSVM)-based spatially varying coefficient logistic regression (LSSVM-SVCLR) model, which combines the flexibility of LSSVM with the interpretability of logistic regression and the spatial adaptability of spatially varying coefficient models. Through simulations under varying data sizes and complexity, the model achieved high accuracy, with area under the curve (AUC) values approaching 1 in simpler cases and around 0.8 in more complex scenarios. A case study analyzing the relationship between climate extremes and flood-drought shifts in Xinjiang demonstrated the model's applicability, achieving training and testing accuracies of 0.994 and 0.831, respectively, outperforming state-of-the-art machine learning models. Furthermore, the model revealed specific spatial effects of climate extremes on flood-drought shifts, providing probabilistic predictions across the study area. The findings highlight the potential of the proposed model to improve predictions of extreme climate-related events, offering valuable insights for disaster management and climate risk evaluation. This study provides a robust framework for analyzing the complexities of spatial nonstationarity in climate risk analysis.