Xiang Tang

In heterogeneous river basins, the temporal dynamics of water quality vary substantially across indicators, monitoring stations, and hydrometeorological conditions, making basin-scale multi-indicator water quality prediction inherently challenging. Although data-driven approaches have been widely used for water quality forecasting, existing studies have focused primarily on single indicators, single sites, or limited validation scenarios, and the applicability of existing forecasting frameworks to multi-indicator prediction in heterogeneous basin settings therefore remains insufficiently assessed. Against this background, this study applies and validates a Fredformer-based frequency-domain forecasting framework for multi-indicator water quality prediction in the Weihe River Basin, with particular emphasis on task-oriented adaptation, performance under multiple validation settings, and potential for prototype-level integration. Using long-term observations from seven monitoring stations, the study conducted baseline comparison, input-output step-size analysis, forward temporal validation, cross-site transfer under limited local fine-tuning, hydrometeorological data fusion, and prototype system integration. The results show that the framework achieved strong overall predictive performance under the validated settings and remained stable in forward temporal validation and cross-site transfer after limited local fine-tuning. These findings indicate that, after task-specific adaptation and systematic validation, an existing frequency-domain forecasting framework can be effectively used for multi-indicator water quality prediction under heterogeneous monitoring conditions within the Weihe River Basin and can also be embedded into a monitoring-oriented prototype workflow for forecasting display, trend-based warning presentation, and sampling support.