Scholarlink

<p>The rising global demand for water has increased the need for groundwater studies, especially in arid regions such as the Araripe Sedimentary Basin (ASB) in northeastern Brazil, where groundwater is the main source of freshwater supply. This study aims to predict groundwater piezometric levels (PLs) in the ASB, using machine learning models to support sustainable water management. The dataset consisted of time series of PLs for 22 groundwater wells from 2013 to 2024, along with independent variables such as well ID, year, month, longitude, latitude, altitude, normalized difference vegetation index (NDVI), rainfall, water consumption, and the number of nearby wells with active water use licenses. The dataset was divided into a training set (70%) and a test set (30%). Three prediction models of different complexity were used to predict PLs: multivariate polynomial regression (MPR), support vector machine (SVM), and random forest (RF). In general, the prediction results showed minimal variability in performance between the different piezometers and models during the training and testing phases. The SVM models showed an average <i>R</i><sup>2</sup> of 0.85, a root mean square error (RMSE) of 1.33&#xa0;m, a mean absolute error (MAE) of 0.95&#xa0;m, a King-Gupta efficiency (KGE) of 0.84, and a Nash–Sutcliffe efficiency (NSE) of 0.76. It is important to note that the simplest model, MPR, achieved satisfactory performance compared to the more complex models, SVM and RF, demonstrating consistency and accuracy. These results highlight the potential of machine learning approaches for effective groundwater level prediction in geologically complex regions with limited data.</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

Predicting piezometric levels in the Araripe Sedimentary Basin, Brazil, using regression and machine learning models

  • 期刊论文

  • 发表日期：

    2025年8月15日

查看全文
  • Mirelle T. V. Lima,
  • Fabrício S. Terra,
  • Juliana P. Silva

摘要

The rising global demand for water has increased the need for groundwater studies, especially in arid regions such as the Araripe Sedimentary Basin (ASB) in northeastern Brazil, where groundwater is the main source of freshwater supply. This study aims to predict groundwater piezometric levels (PLs) in the ASB, using machine learning models to support sustainable water management. The dataset consisted of time series of PLs for 22 groundwater wells from 2013 to 2024, along with independent variables such as well ID, year, month, longitude, latitude, altitude, normalized difference vegetation index (NDVI), rainfall, water consumption, and the number of nearby wells with active water use licenses. The dataset was divided into a training set (70%) and a test set (30%). Three prediction models of different complexity were used to predict PLs: multivariate polynomial regression (MPR), support vector machine (SVM), and random forest (RF). In general, the prediction results showed minimal variability in performance between the different piezometers and models during the training and testing phases. The SVM models showed an average R2 of 0.85, a root mean square error (RMSE) of 1.33 m, a mean absolute error (MAE) of 0.95 m, a King-Gupta efficiency (KGE) of 0.84, and a Nash–Sutcliffe efficiency (NSE) of 0.76. It is important to note that the simplest model, MPR, achieved satisfactory performance compared to the more complex models, SVM and RF, demonstrating consistency and accuracy. These results highlight the potential of machine learning approaches for effective groundwater level prediction in geologically complex regions with limited data.