Long-term trend forecast of chlorophyll-a concentration over eutrophic lakes based on time series decomposition and...

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Long-term forecasting of chlorophyll-a (Chla) concentration is vital for managing eutrophication and pollution in lakes, especially with climate change.
• Time series analysis of Chla is complex due to trend, seasonal, and residual components, alongside nonlinear hydro-environmental factor relationships.

Purpose of the Study:

• To develop and validate a hybrid approach for long-term Chla trend forecasting in lakes.
• To investigate the relationships between Chla and hydro-environmental factors in Lake Taihu.

Main Methods:

• Seasonal and Trend decomposition using Loess (STL) for time series component separation.
• Wavelet coherence for identifying time-frequency domain relationships and time lags.
• Convolutional Neural Network with Bidirectional Long Short-Term Memory (CNN-BiLSTM) for multivariate forecasting.

Main Results:

• STL effectively separated long-term trends from seasonal and residual components.
• Wavelet coherence accurately identified resonance patterns and time lags between Chla and factors like Total Phosphorus (TP) and Water Temperature (WT).
• The CNN-BiLSTM model significantly outperformed other methods in forecasting accuracy (R², RMSE, IOA) and peak capture.

Conclusions:

• The proposed hybrid deep learning approach provides an effective solution for long-term Chla trend forecasting in eutrophic lakes.
• This method can handle the complex attributes of hydro-environmental data for improved algal bloom prediction and management.