Scholarlink

<p>This work presents the assessment of drought conditions using multi-satellite and reanalysis precipitation products over the Murray-Darling Basin (MDB), Australia-a region characterized by high elevation and marked climatic variability. An ensemble framework that integrates multiple machine learning techniques was developed to enhance the accuracy of spatial precipitation estimation. It comprises four steps: (i) K-means on the MDB region from 1970 to 2020 for homogeneous areas to distinct groups on precipitation, (ii) An ensemble of three advanced ML will reduce the systematic biases and uncertainties of the multi-source precipitation products from CHIRPS, TRMM, and ERA5 products, while making relation for the estimation of precipitation in the data-scarce regions, and (iii) Validation implemented using the newly developed reinforcement Q-learning model, EnQLeM, for drought monitoring at different time scales across the MBD. In general, CHIRPS outperformed TRMM and ERA5 over the study area, especially for the highlands. These results showed that the precipitation in MDB clusters into four distinct regions. Also, the proposed relationships, evaluated through K-fold cross-validation processes, presented acceptable and reliable performance. Among the evaluated models, EnQLeM demonstrated the highest accuracy in assessing drought conditions, with an average improvement in correlation coefficient of 12.56% (SPI-3), 11.03% (SPI-6), and 7.12% (SPI-12) over CHIRPS; 20.09% (SPI-3), 17.61% (SPI-6), and 15.8% (SPI-12) over TRMM; 12.47% (SPI-3), 10.88% (SPI-6), and 8.04% (SPI-12) over MARS; and 12.85% (SPI-3), 11.17% (SPI-6), and 9.34% (SPI-12) over ANN across 19 RG stations. The greatest improvement was observed against ERA5, with an average increase of 30.2% for SPI-3, while the smallest enhancement was recorded against RF (1.08%) for SPI-12. Spatial assessment from 1998 to 2020 showed a general increase in drought severity, where SPI values are indicating extreme drought conditions during 2000–2010. The findings suggest the potential of the EnQLeM model in the integration of multi sources toward a robust platform for enhanced drought assessment.</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

An ensemble learning strategy for spatiotemporal drought assessment through the fusion of multi-satellite and atmospheric reanalysis precipitation products

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Mohammad Rezaie-Balf,
  • Lloyd H. C. Chua

摘要

This work presents the assessment of drought conditions using multi-satellite and reanalysis precipitation products over the Murray-Darling Basin (MDB), Australia-a region characterized by high elevation and marked climatic variability. An ensemble framework that integrates multiple machine learning techniques was developed to enhance the accuracy of spatial precipitation estimation. It comprises four steps: (i) K-means on the MDB region from 1970 to 2020 for homogeneous areas to distinct groups on precipitation, (ii) An ensemble of three advanced ML will reduce the systematic biases and uncertainties of the multi-source precipitation products from CHIRPS, TRMM, and ERA5 products, while making relation for the estimation of precipitation in the data-scarce regions, and (iii) Validation implemented using the newly developed reinforcement Q-learning model, EnQLeM, for drought monitoring at different time scales across the MBD. In general, CHIRPS outperformed TRMM and ERA5 over the study area, especially for the highlands. These results showed that the precipitation in MDB clusters into four distinct regions. Also, the proposed relationships, evaluated through K-fold cross-validation processes, presented acceptable and reliable performance. Among the evaluated models, EnQLeM demonstrated the highest accuracy in assessing drought conditions, with an average improvement in correlation coefficient of 12.56% (SPI-3), 11.03% (SPI-6), and 7.12% (SPI-12) over CHIRPS; 20.09% (SPI-3), 17.61% (SPI-6), and 15.8% (SPI-12) over TRMM; 12.47% (SPI-3), 10.88% (SPI-6), and 8.04% (SPI-12) over MARS; and 12.85% (SPI-3), 11.17% (SPI-6), and 9.34% (SPI-12) over ANN across 19 RG stations. The greatest improvement was observed against ERA5, with an average increase of 30.2% for SPI-3, while the smallest enhancement was recorded against RF (1.08%) for SPI-12. Spatial assessment from 1998 to 2020 showed a general increase in drought severity, where SPI values are indicating extreme drought conditions during 2000–2010. The findings suggest the potential of the EnQLeM model in the integration of multi sources toward a robust platform for enhanced drought assessment.