Scholarlink

Fluorescent dissolved organic matter (FDOM) is one of the most critical water quality variables. Despite the significant improvement in water quality modelling and forecasting over the last few years, accurate forecasting of FDOM concentration is research challenge. In order to improves our knowledge skills, the present chapter presents a forecasting study based on the application of various machine learning models namely: (i) Random Forest Regression (RFR), (ii) Regularized Greedy Forest (RGF), (iii) Extremely Randomized Trees (ERT), (iv) Support Vector Regression (SVR), and (v) Relevance Vector Machine (RVM), for forecasting FDOM concentration at three different horizons: t, t + 1 and t + 7, measured at daily time scale. All models were developed using sig input lag from (t −1) to (t − 6). All models were compared suing numerical and graphical comparison, i.e., the root mean squared error (RMSE), the mean absolute error (MAE), the coefficient of correlation (R), and the Nash–Sutcliffe efficiency (NSE), and graphical visualization, i.e., scatterplot, boxplot, violinplot and Taylor diagram. From the obtained results, we can draw the following conclusions. At the horizon (t), the best numerical performances were obtained by ERT equally with RFR with R, NSE, RMSE and MAE of approximately 0.936, 0.876, 1.367, and 0.733, respectively, while the RVM was found to bez the poorest model. At the horizon (t + 7), the performances of the models were significantly decreased reaching the values of 0.464, 0.161, 3.555, and 2.269, in terms of R, NSE, RMSE and MAE, indices.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Surface Water and Groundwater Quality Forecasting Using Machine Learning Models

  • 章节

  • 发表日期：

    2025年

查看全文
  • Salim Heddam

摘要

Fluorescent dissolved organic matter (FDOM) is one of the most critical water quality variables. Despite the significant improvement in water quality modelling and forecasting over the last few years, accurate forecasting of FDOM concentration is research challenge. In order to improves our knowledge skills, the present chapter presents a forecasting study based on the application of various machine learning models namely: (i) Random Forest Regression (RFR), (ii) Regularized Greedy Forest (RGF), (iii) Extremely Randomized Trees (ERT), (iv) Support Vector Regression (SVR), and (v) Relevance Vector Machine (RVM), for forecasting FDOM concentration at three different horizons: t, t + 1 and t + 7, measured at daily time scale. All models were developed using sig input lag from (t −1) to (t − 6). All models were compared suing numerical and graphical comparison, i.e., the root mean squared error (RMSE), the mean absolute error (MAE), the coefficient of correlation (R), and the Nash–Sutcliffe efficiency (NSE), and graphical visualization, i.e., scatterplot, boxplot, violinplot and Taylor diagram. From the obtained results, we can draw the following conclusions. At the horizon (t), the best numerical performances were obtained by ERT equally with RFR with R, NSE, RMSE and MAE of approximately 0.936, 0.876, 1.367, and 0.733, respectively, while the RVM was found to bez the poorest model. At the horizon (t + 7), the performances of the models were significantly decreased reaching the values of 0.464, 0.161, 3.555, and 2.269, in terms of R, NSE, RMSE and MAE, indices.