One of the main concerns of environmental and ecological managers for rivers, lakes, reservoirs, and marine ecosystems is developing a reliable and efficient predictive model for chlorophyll a concentration. In this study, the online sequential extreme learning machine, M5 Prime tree, multilayer perceptron artificial neural network, response surface methodology, and multivariate adaptive regression spline models were investigated for daily chlorophyll a concentration prediction by assessing the relations between Chl-a and several water quality parameters, including water temperature, pH, specific conductance, and turbidity. Different scenarios based on TE, pH, SC, and TU were defined. Also, this study evaluated the influence of periodicity input as the last scenario to obtain more accurate predictions of Chl-a values. Daily data measured for 2009–2019 from USGS no. 14207200 and USGS no. 14211720 stations were used. For assessing the prediction performance of the proposed techniques, three different objective indicators were employed, namely RMSE, R2 , and NSE. Moreover, the Taylor diagram was employed for evaluating the accuracy and generalization capability of the applied models for the prediction of Chl-a. Results indicated that OS-ELM with input parameters of TE, pH, SC, TU, Y (year), M (month), and D (day) showed higher accuracy in predicting Chl-a with RMSE of 3.151, NSE of 0.798, and R of 0.894 for USGS no. 14207200 and with RMSE of 0.907, NSE of 0.820, and R of 0.912 for USGS no. 14211720 than the other models, respectively. Additionally, MLPNN ranked as the second best method for the estimation of Chl-a values at both stations. As an interesting point, it was quite evident that adding periodicity as an input parameter could significantly enhance the performance of all models in predicting the daily Chl-a concentration at both stations. Results proved that OS-ELM models can be a reliable tool for the prediction of the Chl-a values in aquatic environments, benefiting ecological and environmental management, and algal bloom control.
In this paper, the capacity of an Adaptive-Network-Based Fuzzy Inference System (ANFIS) for predicting salinity of the Tafna River is investigated. Time series data of daily liquid flow and saline concentrations from the gauging station of Pierre du Chat (160801) were used for training, validation and testing the hybrid model. Different methods were used to test the accuracy of our results, i.e. coefficient of determination (R2), Nash–Sutcliffe efficiency coefficient (E), root of the mean squared error (RSR) and graphic techniques. The model produced satisfactory results and showed a very good agreement between the predicted and observed data, with R2 equal (88% for training, 78.01% validation and 80.00% for testing), E equal (85.84% for training, 82.51% validation and 78.17% for testing), and RSR equal (2% for training, 10% validation and 49% for testing).
PL
W pracy badano zdolność systemu wnioskowania rozmytego opartego na adaptacyjnej sieci (ANFIS) do przewidywania zasolenia rzeki Tafna. Do trenowania, oceny i testowania modelu hybrydowego wykorzystano serie pomiarów dobowych przepływów płynu i stężeń soli ze stacji pomiarowej w Pierre du Chat (160801). Dokładność wyników testowano za pomocą: współczynnika determinacji (R2), współczynnika wydajności Nasha–Sutcliffe’a (E), pierwiastka średniego błędu kwadratowego (RSR) i technik graficznych. Model dał zadowalające wyniki i wykazywał dobrą zgodność między danymi obserwowanymi a przewidywanymi: R2 (88% w przypadku uczenia sieci, 78.01% walidacji i 80.00% testowania), E (85.84% w przypadku uczenia sieci, 82.51% walidacji i 78.17% testowania) i RSR (2% w przypadku uczenia sieci, 10% walidacji i 49% testowania).
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.