| Abstrakt: |
Monitoring and managing lake water quality are some of the most reminiscent of aquatic environmental challenges. Prediction of water quality parameters plays a major role in the enrichment of water resource management. This study creates and compares deep learning models such as ANFIS (adaptive neuro-fuzzy inference system), LSTM (long short-term memory), and NAR neural networks in water quality prediction and proposes the most effective prediction model. Obtaining adequate water quality data with high precision to train and test deep learning models is sometimes challenging due to cost or technological constraints. A solution to this problem was established by developing the best band regression empirical water quality (BREWQ) model, which was used to extract seasonal water quality dataset (n= 490) from multi-resolution satellite imagery (Sentinel 2A) for 2016–2021. The accuracy assessment parameters, such as coefficient of determination (R2), mean squared error (MSE), root mean squared error (RMSE), and mean absolute percentage error (MAPE), is confined to illustrate the performance accuracy of the incorporated models. The results from this study inferred that the ANFIS model approach of Gbellmf linear, Gaussmff with the fuzzy set combination of [3 3 3 3] has the potential to predict magnesium-87.03% (MSE:8.2436, RMSE:9.92, MAPE:4.50), potassium-94.34% (MSE:6.24, RMSE: 8.06, MAPE: 31.55), sodium-96.19% (MSE:5.82, RMSE:7.36, MAPE:4.82). Secchi Disc Depth (SDD)-99.87% (MSE:3.72, RMSE:6.14, MAPE:16.03), temperature-96.2% (MSE:2.19, RMSE:3.67, MAPE:45.33), total hardness-90.31% (MSE:8.09, RMSE:10.13, MAPE: 8.91) and turbidity-60.52% (MSE:6.24, RMSE: 8.08, and MAPE:1688.43) with the least error, whereas LSTM showed the lowest error in predicting parameters such as Chlorophyll-a (Chl-a)- 96.76% (MSE:8.78, RMSE:10.4502 & MAPE:16.83), and Total Suspended Solids (TSS)-84.31% (MSE:3.58, RMSE: 6.68, MAPE:35.69). On the other hand, it is revealed from this study that NARNET had less accuracy than LSTM and ANFIS in predicting water quality due to their simple network structure. In the ultramodern era, such deep-learning prediction models aid in continuously monitoring water bodies to prevent pollution. |
Full text from SpringerLink