Unveiling the Spatial Distribution and Temporal Trends of Total Phosphorus in the Yangtze River: Towards a Predictive Time-Series Modeling for Environmental Management

The accurate prediction of total phosphorus in water quality is crucial for monitoring ecosystem stability and eutrophication status. However, the distribution of natural environmental data such as water quality total phosphorus (TP) often undergoes complex changes over time. Stable and reliable predictive outcomes not only necessitate a degree of stability and periodicity within the natural data, but also require that TP prediction models exhibit strong adaptability to the random fluctuations and distribution drifts of environmental data. Therefore, adapting predictive models to accommodate distribution drifts in natural environmental data presents a challenge. This study provides a detailed description of the spatiotemporal variations of TP in the Yangtze River from 2019 to 2023. Utilizing data cleaning and data mining techniques, time series data were analyzed to generate a predictive dataset, with a particular emphasis on investigating the stability and periodicity of TP fluctuations. By comparing various time series forecasting models, the MTS-Mixers was ultimately selected as the experimental baseline model, and different modes were employed for time series prediction. The results demonstrate that the model maintains relatively high prediction accuracy within 20 time steps. The research findings not only offer a comprehensive description and reliable prediction of TP variations in the Yangtze River, but also provide effective methods and tools for water quality monitoring and management. They serve as a scientific basis for environmental protection and water quality improvement in the Yangtze River Basin, facilitating the formulation and implementation of relevant policies and advancing the sustainable development of the Yangtze River water environment. Furthermore, the study also confirms the applicability of machine learning in hydrological forecasting, which can be utilized for addressing environmental changes. Future research directions include ensuring the stability of critical monitoring data and exploring time-domain sub-band reconstruction methods to better understand the frequency characteristics of time series data, revealing hidden information and features.