Scholarlink

<p>The accurate estimation of precipitation is important in hydrological modeling and climate impact assessment. This is especially true in regions with challenging topography, such as the Andes in Peru. Unfortunately, satellite data derived from precipitation often have considerable biases and fail to delineate minute variations produced by elevation and topography. Existing techniques, such as FY4A-AGRI, TMPA v7, and TA-UNet, have struggled with these problems and have produced less reliable data for water resource planning. This research has developed a new deep learning framework called Preference Relation Recurrent Residual Graph Network (PRRRGN), integrated with hybrid Dargo Lizard–Fishing Cat Optimizer (DLFCO), which is directed towards enhancing precipitation estimation accuracy by combining spatiotemporal features, optimizing hyperparameters, and improving training convergence. When evaluating against TA-UNet, the suggested PRRRGN framework's performance with TA-UNet model is remarkably superior in all main performance metrics, with an RMSE of 0.10&#xa0;mm, MAE of 0.07&#xa0;mm, and an NMAE of 12%, in addition to an NSE of 0.96 in annual assessments for the Peruvian region. The PRRRGN model provides correction for satellite precipitation data that is scalable, fast, and highly accurate, thus indicating promising future for such applications in environmental monitoring and early warnings in regions with difficult topography. PRRRGN: A Deep Learning Framework for Bias-Corrected Satellite Precipitation Estimation in Complex Terrains.</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

PRRRGN: A Deep Learning Framework for Bias-corrected Satellite Precipitation Estimation in Complex Terrains

  • 期刊论文

  • 发表日期：

    2026年1月24日

查看全文
  • G. Kalyan Kumar,
  • P. Sankar

摘要

The accurate estimation of precipitation is important in hydrological modeling and climate impact assessment. This is especially true in regions with challenging topography, such as the Andes in Peru. Unfortunately, satellite data derived from precipitation often have considerable biases and fail to delineate minute variations produced by elevation and topography. Existing techniques, such as FY4A-AGRI, TMPA v7, and TA-UNet, have struggled with these problems and have produced less reliable data for water resource planning. This research has developed a new deep learning framework called Preference Relation Recurrent Residual Graph Network (PRRRGN), integrated with hybrid Dargo Lizard–Fishing Cat Optimizer (DLFCO), which is directed towards enhancing precipitation estimation accuracy by combining spatiotemporal features, optimizing hyperparameters, and improving training convergence. When evaluating against TA-UNet, the suggested PRRRGN framework's performance with TA-UNet model is remarkably superior in all main performance metrics, with an RMSE of 0.10 mm, MAE of 0.07 mm, and an NMAE of 12%, in addition to an NSE of 0.96 in annual assessments for the Peruvian region. The PRRRGN model provides correction for satellite precipitation data that is scalable, fast, and highly accurate, thus indicating promising future for such applications in environmental monitoring and early warnings in regions with difficult topography. PRRRGN: A Deep Learning Framework for Bias-Corrected Satellite Precipitation Estimation in Complex Terrains.