Scholarlink

<p>In this study, a novel hybrid modeling framework combining the reptile search algorithm (RSA) with the outlier robust extreme learning machine (ORELM) was developed to enhance the prediction accuracy of groundwater equivalent thickness derived from the Gravity Recovery and Climate Experiment (GRACE) satellite observations. The investigation focuses on the Lake Urmia Basin, located in northwestern Iran. A total of 163 monthly GRACE datasets spanning the period from April 2002 to June 2017 were employed. To isolate the groundwater signal, hydrological components simulated by the GLDAS model were subtracted from the GRACE-derived total water storage variations. The output of the satellites includes 6 points located in the chosen basin in which the results show the decreasing trend of the groundwater equivalent thickness with the changing domain of −&#xa0;50 to + 50 for the certain mentioned points within the basin. The satellite output results obtained from the six study points are compared with the data of piezometric wells existing in each point zone. The comparison of the output of the GRACE satellites with the observed data displays that the correlation coefficient value in six points is 0.57 on average. In addition, the values of <i>RMSE</i>, <i>MARE</i>, and <i>RMSRE</i> are on average 8.8, 1.4, and 3.1, respectively, showing the appropriate performance of the GRACE satellites in approximating the groundwater equivalent thickness in the study region. After that, the time series obtained by the GRACE satellites for the six points are modeled by the RSA–ORELM model. Regarding the same trend in all points, the data are modeled simultaneously for them. The obtained results prove the suitable performance of the RSA–ORELM model so that the correlation coefficient in the training and testing stages is 0.97 and 0.87, respectively. The <i>RMSE</i> values in both stages are very close to each other. Also, the <i>RMSRE</i> value in the testing stage has a better performance than the training stage, showing the high efficiency of the model in simulating the GRACE 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

Prediction of groundwater equivalent thickness using the integration of the improved RSA–ORELM model and GRACE satellite images

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Mahtab Badkoubeh Hezaveh,
  • Mohsen Najarchi,
  • Mohammad Reza Jalali,
  • Hosein Mazaheri,
  • Saeid Shabanlou

摘要

In this study, a novel hybrid modeling framework combining the reptile search algorithm (RSA) with the outlier robust extreme learning machine (ORELM) was developed to enhance the prediction accuracy of groundwater equivalent thickness derived from the Gravity Recovery and Climate Experiment (GRACE) satellite observations. The investigation focuses on the Lake Urmia Basin, located in northwestern Iran. A total of 163 monthly GRACE datasets spanning the period from April 2002 to June 2017 were employed. To isolate the groundwater signal, hydrological components simulated by the GLDAS model were subtracted from the GRACE-derived total water storage variations. The output of the satellites includes 6 points located in the chosen basin in which the results show the decreasing trend of the groundwater equivalent thickness with the changing domain of − 50 to + 50 for the certain mentioned points within the basin. The satellite output results obtained from the six study points are compared with the data of piezometric wells existing in each point zone. The comparison of the output of the GRACE satellites with the observed data displays that the correlation coefficient value in six points is 0.57 on average. In addition, the values of RMSE, MARE, and RMSRE are on average 8.8, 1.4, and 3.1, respectively, showing the appropriate performance of the GRACE satellites in approximating the groundwater equivalent thickness in the study region. After that, the time series obtained by the GRACE satellites for the six points are modeled by the RSA–ORELM model. Regarding the same trend in all points, the data are modeled simultaneously for them. The obtained results prove the suitable performance of the RSA–ORELM model so that the correlation coefficient in the training and testing stages is 0.97 and 0.87, respectively. The RMSE values in both stages are very close to each other. Also, the RMSRE value in the testing stage has a better performance than the training stage, showing the high efficiency of the model in simulating the GRACE data.