Forecasting chlorophyll-a concentration using empirical wavelet transform and support vector regression

Accurate forecast of chlorophyll-a concentration of water bodies is important for aquatic management because it can support management decisions with future information. In this regard, this paper proposes a new chlorophyll-a forecast method that combines empirical wavelet transform, support vector regression, and sine cosine algorithm. Chlorophyll-a concentration data are decomposed by empirical wavelet transform, and then, support vector regression is employed to predict decomposed components, and finally, forecast value is obtained by reconstructing the predicted values for the decomposed components. Hyperparameters of support vector regression models are optimized by sine cosine algorithm. Our model is evaluated by chlorophyll-a concentration data of Lake Kasumigaura, Japan. For the purpose of comparison, several other models are also developed. Result indicates that our method shows better forecast performance than other competitor models. This study demonstrates that data processing by empirical wavelet transform can significantly improve forecast accuracy and our method is a promising new forecast method for lake chlorophyll-a concentration.