Optimizing lakeside buffers and extending ecological redlines to key rivers: A multi-scale buffer and causal identification for diagnosing water quality drivers in a plateau lake, China.

Elucidating the spatiotemporal evolution of water quality and its driving mechanisms is essential for the sustainable management of basin water resources in plateau lakes. Taking Dianchi Lake-a typical plateau lake in the Yunnan-Guizhou Plateau, China-as the study area, monthly water quality data from 10 monitoring sites (2001-2019) were analyzed. A single-factor assessment combined with an optimized Water Quality Index (WQI) was used to characterize spatiotemporal variability. To systematically investigate the complex driving mechanisms involving natural (vegetation, hydrology, climate) and anthropogenic (socio-economic, landscape) factors, an integrated framework was established using machine learning (Random Forest, eXtreme Gradient Boosting, and SHapley Additive exPlanations), Optimal Parameters‑based Geographical Detector, causal inference (Peter-Clark Momentary Conditional Independence plus), and Partial Least Squares Path Modeling, with particular emphasis on spatial scale effects, factor interactions, causal relationships, and impact pathways. The main findings are as follows. (1) Pollution and eutrophication in Dianchi Lake remain severe; COD, Chl.a, TN, and TP frequently exceed Grade V standards. The WQI reveals better water quality in spring and winter than in summer and autumn, with overall quality predominantly at Grade IV-V (∼60% of observations). Water quality in Caohai is substantially worse than that in Waihai. (2) Water quality in Caohai is primarily driven by landscape (>48%) and climate (>33%), whereas that in Waihai is jointly influenced by landscape (>22%), climate (>31%), and hydrology (>17%). The relationships between natural/anthropogenic factors and water quality are complex and predominantly nonlinear. (3) Multi-factor interactions generally amplify the impacts on water quality, with interactions involving landscape factors being particularly critical. Many factors exhibit both contemporaneous and lagged (1-3 months) causal relationships with water quality. (4) The optimal buffer scales follow the order: landscape > socio-economic > vegetation, and the effective spatial extent is generally larger for rivers than for lakes. Based on these results, it is recommended to appropriately extend buffer zones around the lake in ecological protection management and to delineate ecological red lines in key inflowing river catchments, thereby achieving coordinated "Lake-River-Land" system management. The findings provide a scientific basis for precision pollution control and ecological spatial planning in Dianchi Lake, as well as a theoretical framework for the comprehensive protection and management of plateau lake basins.