Scholarlink

<p>Various Remote Sensing (RS) models have been developed to estimate evapotranspiration (ET) using different physical principles; however, their performance varies spatiotemporally. This variability makes it challenging to reliably estimate ET using RS based ET models always. Recently, ensemble modelling techniques have shown promise in improving the accuracy of modelled ET and this study aims to develop ensemble ET models over the Indian region. Three popular RS based ET models were used to create the ensemble: Priestley-Taylor Jet Propulsion Lab (PT-JPL), Soil Plant Atmosphere and Remote Sensing Evapotranspiration (SPARSE – Layer and Patch), and Surface Temperature Initiated Closure (STIC). The ensembles were created using simple averaging, Bayesian Model Averaging (BMA), k-Nearest Neighbor (k-NN), Random Forest (RF), and Support Vector Machine (SVM) techniques. The study was conducted at both the in-situ scale for daily ET using data from seven sites, and at a 1&#xa0;km scale for 8-day periods across India. The results showed that the ensembles outperformed individual models at both the in-situ and 1&#xa0;km scales. The Root Mean Square Error (RMSE) of the individual ET models ranged between 60 and 88 Wm<sup>–2</sup> and the RMSE of different ensembles was in the range of 36–52 Wm<sup>–2</sup> at field scale. Similarly, with MODIS data, the RMSE of individual models and ensembles were in the range of 46–52 Wm<sup>–2</sup> and 26–39 Wm<sup>–2</sup> respectively.\ The machine learning (ML) based ensembles performed better by significantly reducing the RMSE compared to simple averaging and BMA highlighting the potential of the technique to estimate accurate ET in a consistent manner.</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

Multi-model ensemble improves evapotranspiration estimation over India

  • 期刊论文

  • 发表日期：

    2025年11月28日

查看全文
  • K. V. Athira,
  • Eswar Rajasekaran,
  • Gilles Boulet,
  • Rahul Nigam,
  • B. K. Bhattacharya

摘要

Various Remote Sensing (RS) models have been developed to estimate evapotranspiration (ET) using different physical principles; however, their performance varies spatiotemporally. This variability makes it challenging to reliably estimate ET using RS based ET models always. Recently, ensemble modelling techniques have shown promise in improving the accuracy of modelled ET and this study aims to develop ensemble ET models over the Indian region. Three popular RS based ET models were used to create the ensemble: Priestley-Taylor Jet Propulsion Lab (PT-JPL), Soil Plant Atmosphere and Remote Sensing Evapotranspiration (SPARSE – Layer and Patch), and Surface Temperature Initiated Closure (STIC). The ensembles were created using simple averaging, Bayesian Model Averaging (BMA), k-Nearest Neighbor (k-NN), Random Forest (RF), and Support Vector Machine (SVM) techniques. The study was conducted at both the in-situ scale for daily ET using data from seven sites, and at a 1 km scale for 8-day periods across India. The results showed that the ensembles outperformed individual models at both the in-situ and 1 km scales. The Root Mean Square Error (RMSE) of the individual ET models ranged between 60 and 88 Wm–2 and the RMSE of different ensembles was in the range of 36–52 Wm–2 at field scale. Similarly, with MODIS data, the RMSE of individual models and ensembles were in the range of 46–52 Wm–2 and 26–39 Wm–2 respectively.\ The machine learning (ML) based ensembles performed better by significantly reducing the RMSE compared to simple averaging and BMA highlighting the potential of the technique to estimate accurate ET in a consistent manner.