Residual multi-model anomaly analysis for climate-ecosystem disturbances in the Indian Sundarbans

The presence of strong seasonal variability and trends can hide signals of ecosystem stress, which can further make it difficult to identify disturbances in climate-ecosystem systems. Therefore, the need to distinguish the climatic structure from the variability representing the disturbance patterns to understand how the disturbance evolves, sustains itself, and appears before and after the catastrophic event becomes important in such analyses. The present study applies a residuals-driven analytical approach aimed at identifying disturbance-associated variability patterns in climate-ecosystem systems in the Indian Sundarbans for the period of 2000–2025. The necessary variables of temperature, rainfall, vegetation, and hydrological variables were subjected to time series decomposition along with the identification of structural residuals via Singular Spectrum Analysis, and a set of valid residuals was obtained by applying two different machine learning approaches for the classification of residuals in machine learning algorithms. For the 13 anomalous months distinguished by the Random Forest algorithm, 14 by the Gradient Boost algorithm, precipitation error was discovered to be the fundamental cause of the disturbance, while the presence of smaller-than-significant indicators included residuals for vegetation, as well as water shortage. For the evaluation of credibility regarding these anomalies, six-month cyclone reference systems were applied to verify that a large number of these disturbances were event-related while also implying non-specific incidents that were not traceable to a specific danger. Only the instance of Cyclone Sidr had a clearly identifiable major precursor, while the rest were of shorter periods with identifiable precursors. The findings show that, in line with increasing background variability and disturbance memory, disturbance expression in the Sundarbans has changed over time from more obvious pre-event precursors to post-event dominance. A framework for examining disturbance evolution in deltaic ecosystems subjected to recurrent hydroclimatic shocks is provided by residual-based anomaly chronologies, which allow interpretable disturbance attribution and precursor evaluation without depending on forecasts.