| Prediction of Total Dissolved Solids in the Groundwater of the Birjand Plain, Iran, Using Machine-Learning Models under Different Input Scenarios |
| کد مقاله : 1169-CONFIHA |
| نویسندگان |
|
مهران ملکی نیا * گروه عمران دانشگاه آزاد اسلامی واحد قاینات |
| چکیده مقاله |
| Total dissolved solids (TDS) indicates groundwater salinity and suitability for drinking and irrigation, but its measurement requires major-ion laboratory analysis or electrical conductivity. This study compares four machine-learning algorithms—random forest (RF), gradient boosting machine (GBM), support vector regression (SVR), and Bayesian-regularized multilayer perceptron (MLP)—and their equal-weight ensemble for predicting TDS in the arid Birjand Plain, eastern Iran. The dataset comprised 642 samples (TDS: 416–9292.8 mg L⁻¹; mean: 2968.2 mg L⁻¹) with K⁺, Na⁺, Mg²⁺, Ca²⁺, SO₄²⁻, HCO₃⁻, Cl⁻, pH, and groundwater level (GWL). Five input scenarios (S1–S5) progressively excluded dominant ions. Models were evaluated using five-fold cross-validation repeated three times. With all inputs (S1), models predicted TDS with high accuracy (R² = 0.938–0.994; ensemble RMSE = 129.6 mg L⁻¹), largely reflecting the additive mass-balance relationship between TDS and constituent ions. Excluding Na⁺ and Cl⁻ (S3) reduced ensemble performance to R² = 0.897 (RMSE = 549.5 mg L⁻¹), while the minimal scenario (S5), using pH, GWL, HCO₃⁻, Ca²⁺, and Mg²⁺, achieved R² = 0.837. The ensemble had the lowest RMSE in all scenarios, followed by GBM and RF; SVR performed weakest and systematically under-predicted TDS. In S3, Ca²⁺, Mg²⁺, and SO₄²⁻ contributed about 73% of RF importance. These findings indicate that TDS can be estimated with acceptable accuracy from fewer, inexpensive hydrochemical variables, supporting rapid salinity screening and gap-filling in arid-aquifer monitoring records. |
| کلیدواژه ها |
| Groundwater salinity; Total dissolved solids; Machine learning; Random forest; Gradient boosting; Support vector regression; Birjand Plain |
| وضعیت: پذیرفته شده برای ارائه شفاهی |
