Duong Hai Ha, Huong Thi Thanh Ngo, Phong Tran Van, Dam Nguyen Duc, Mohammadtaghi Avand, Duy Nguyen Huu, M. Amiri, Hiep Van Le, Indra Prakash, Pham Binh Thai
{"title":"Development and application of hybrid artificial intelligence models for groundwater potential mapping and assessment","authors":"Duong Hai Ha, Huong Thi Thanh Ngo, Phong Tran Van, Dam Nguyen Duc, Mohammadtaghi Avand, Duy Nguyen Huu, M. Amiri, Hiep Van Le, Indra Prakash, Pham Binh Thai","doi":"10.15625/2615-9783/17240","DOIUrl":null,"url":null,"abstract":"Groundwater potential assessment is essential for optimum utilization and recharge of groundwater resources for the proper development and management of an area. The main aim of this study is to develop an accurate groundwater potential map of the Dak Nong Province (Vietnam) using hybrid artificial intelligence models, which are a combination of Random Forest (RF) and its Ensemble Framework (AdaBoost - ABRF, Bagging - BRF and LogitBoost - LBRF). In this study, twelve conditioning factors, namely topography (aspect, elevation, Topographic Wetness Index - TWI, slope, and curvature), hydrology (infiltration and river density, rainfall, Sediment Transport Index - STI, Stream Power Index - SPI), land use, and soil were used to develop the models. Well, yield data was also utilized to develop and validate potential groundwater zones. \nOne Rule (R) feature selection method was utilized to prioritize the importance of groundwater potential affecting parameters. The results indicated that the Average Merit (AM) of the rainfall factor was the highest (68.039), and river density was the lowest (53,969). Performance evaluation of ML models was done using standard statistical indicators, including Area Under the Receiver Operating Characteristic (ROC) curve (AUC). The results showed that all the four models performed well in the training (AUC ≥ 0.967) and testing (AUC ≥ 0.734) phases, but the performance of the ABRF (AUC=0.992) model is the best in the training phase, whereas LBRF is the best in the testing phase (AUC=0.776). The present model study would be helpful in the proper groundwater potential assessment and management of groundwater resources for sustainable development. ","PeriodicalId":23639,"journal":{"name":"VIETNAM JOURNAL OF EARTH SCIENCES","volume":" ","pages":""},"PeriodicalIF":2.4000,"publicationDate":"2022-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"VIETNAM JOURNAL OF EARTH SCIENCES","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.15625/2615-9783/17240","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"GEOSCIENCES, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 3
Abstract
Groundwater potential assessment is essential for optimum utilization and recharge of groundwater resources for the proper development and management of an area. The main aim of this study is to develop an accurate groundwater potential map of the Dak Nong Province (Vietnam) using hybrid artificial intelligence models, which are a combination of Random Forest (RF) and its Ensemble Framework (AdaBoost - ABRF, Bagging - BRF and LogitBoost - LBRF). In this study, twelve conditioning factors, namely topography (aspect, elevation, Topographic Wetness Index - TWI, slope, and curvature), hydrology (infiltration and river density, rainfall, Sediment Transport Index - STI, Stream Power Index - SPI), land use, and soil were used to develop the models. Well, yield data was also utilized to develop and validate potential groundwater zones.
One Rule (R) feature selection method was utilized to prioritize the importance of groundwater potential affecting parameters. The results indicated that the Average Merit (AM) of the rainfall factor was the highest (68.039), and river density was the lowest (53,969). Performance evaluation of ML models was done using standard statistical indicators, including Area Under the Receiver Operating Characteristic (ROC) curve (AUC). The results showed that all the four models performed well in the training (AUC ≥ 0.967) and testing (AUC ≥ 0.734) phases, but the performance of the ABRF (AUC=0.992) model is the best in the training phase, whereas LBRF is the best in the testing phase (AUC=0.776). The present model study would be helpful in the proper groundwater potential assessment and management of groundwater resources for sustainable development.