Scholarlink

<p>The deterioration of water quality in downstream dams/reservoirs may result from agricultural chemicals, industrial waste, sediment inflow due to soil erosion, and urban runoff. Therefore, conducting thorough quality analyses of dam/reservoir water is crucial to guarantee the provision of high-quality irrigation water. Keeping this in view, a study was carried out at Punjab Agricultural University, Ludhiana, to evaluate the water quality of the Dholbaha reservoir, which is a multi-purpose dam situated in the Hoshiarpur district of Punjab, India. Machine Learning techniques viz. Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest (RF), coupled with Remote sensing-derived indices and parameters, namely Chlorophyll-a (Chl-a) concentration, Normalized Difference Chlorophyll Index (NDCI), and Normalized Difference Turbidity Index (NDTI), were used for analyzing water quality of the reservoir for each month of the year 2022. Moreover, on-site investigations were carried out for water quality analysis of the reservoir during pre- and post-monsoon periods to develop a robust water quality assessment technique. Rigorous water sampling was done in the months of May and October and the collected samples were examined for electrical conductivity (EC), pH, chloride, calcium, magnesium, and total dissolved solids (TDS). Results showed a pH decrease from 8.3 (pre-monsoon) to 7.6 (post-monsoon), within the normal range. EC increased from 0.042 to 0.047 dS/m, while TDS decreased from 169.5 ppm to 139.3 ppm post-monsoon, all within suitable ranges. Chloride content decreased from 1.8 to 1.0 meq/l, and calcium and magnesium levels reduced substantially in post-monsoon, indicating improved water quality. Machine learning models effectively integrated remote sensing and field data to predict water quality, particularly turbidity. PLSR showed the highest predictive accuracy (R² = 0.99), capturing strong linear correlations between inputs and predicted values, while SVR (R² = 0.96) handled non-linear relationships, and RF (R² = 0.90) captured complex interactions with slightly higher variability. Furthermore, these models were applied to predict the water quality of the reservoir, confirming PLSR as the most efficient method, while SVR and RF also provided reliable predictions for more complex patterns. Both remote sensing and conventional analyses of water quality proved to be efficient inputs for the machine learning model, indicating effectively dynamic spatial and temporal fluctuations. Together, these methods provide comprehensive and complementary insights for effective water quality assessment and management.</p>

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

Unveiling Water Quality Dynamics of Dholbaha Reservoir, Punjab, Using Remote Sensing and Machine Learning Techniques with In-Situ Validation

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Mahesh Chand Singh,
  • Jaswinder Singh,
  • Koyel Sur,
  • Ramandeep Kaur

摘要

The deterioration of water quality in downstream dams/reservoirs may result from agricultural chemicals, industrial waste, sediment inflow due to soil erosion, and urban runoff. Therefore, conducting thorough quality analyses of dam/reservoir water is crucial to guarantee the provision of high-quality irrigation water. Keeping this in view, a study was carried out at Punjab Agricultural University, Ludhiana, to evaluate the water quality of the Dholbaha reservoir, which is a multi-purpose dam situated in the Hoshiarpur district of Punjab, India. Machine Learning techniques viz. Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest (RF), coupled with Remote sensing-derived indices and parameters, namely Chlorophyll-a (Chl-a) concentration, Normalized Difference Chlorophyll Index (NDCI), and Normalized Difference Turbidity Index (NDTI), were used for analyzing water quality of the reservoir for each month of the year 2022. Moreover, on-site investigations were carried out for water quality analysis of the reservoir during pre- and post-monsoon periods to develop a robust water quality assessment technique. Rigorous water sampling was done in the months of May and October and the collected samples were examined for electrical conductivity (EC), pH, chloride, calcium, magnesium, and total dissolved solids (TDS). Results showed a pH decrease from 8.3 (pre-monsoon) to 7.6 (post-monsoon), within the normal range. EC increased from 0.042 to 0.047 dS/m, while TDS decreased from 169.5 ppm to 139.3 ppm post-monsoon, all within suitable ranges. Chloride content decreased from 1.8 to 1.0 meq/l, and calcium and magnesium levels reduced substantially in post-monsoon, indicating improved water quality. Machine learning models effectively integrated remote sensing and field data to predict water quality, particularly turbidity. PLSR showed the highest predictive accuracy (R² = 0.99), capturing strong linear correlations between inputs and predicted values, while SVR (R² = 0.96) handled non-linear relationships, and RF (R² = 0.90) captured complex interactions with slightly higher variability. Furthermore, these models were applied to predict the water quality of the reservoir, confirming PLSR as the most efficient method, while SVR and RF also provided reliable predictions for more complex patterns. Both remote sensing and conventional analyses of water quality proved to be efficient inputs for the machine learning model, indicating effectively dynamic spatial and temporal fluctuations. Together, these methods provide comprehensive and complementary insights for effective water quality assessment and management.