Scholarlink

<p>Early prediction of agricultural drought helps in mitigating the impacts of drought, by developing reliable warning systems and proper resource allocation. This study investigates the strength of wavelet transform as a signal denoising tool to build a robust agricultural drought prediction model up to 6 months of lead time. Wavelet-coupled support vector machine (WSVM), convolutional neural network (WCNN) and long short-term memory (WLSTM) models were used for agricultural drought prediction in the Palakkad district of Kerala state, India. All the wavelet-coupled models showed reliable prediction performance during evaluation. The WLSTM performed exceptionally well for longer lead predictions making it a suitable choice for future drought forecasting. In comparison, WLSTM outperforms WCNN and WSVM, especially at higher leads of 5 and 6 months at all the grid points in the study area. The hybrid wavelet SVM model shows superior performance than simple SVM in predicting agricultural drought, exhibiting a 9% increase in the average <i>R</i><sup>2</sup> value at a 4-month lead time. This underscores the efficacy of wavelet denoising techniques. The wavelet CNN and LSTM models were also efficient in predicting drought duration and severity better than the traditional CNN and LSTM models. This study recommends a wavelet-based LSTM model for the long lead prediction of agricultural drought in the study area. The findings of the study will facilitate informed decision making and effective resource allocation, while also aiding in the development of robust early warning systems for drought forecasting.</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

Comparative study of different wavelet-machine learning models for agricultural drought prediction

  • 期刊论文

  • 发表日期：

    2025年7月24日

查看全文
  • Ayush Singh,
  • K. Saranya Das,
  • Athira Vijay,
  • Akhil R. Nath,
  • N. R. Chithra

摘要

Early prediction of agricultural drought helps in mitigating the impacts of drought, by developing reliable warning systems and proper resource allocation. This study investigates the strength of wavelet transform as a signal denoising tool to build a robust agricultural drought prediction model up to 6 months of lead time. Wavelet-coupled support vector machine (WSVM), convolutional neural network (WCNN) and long short-term memory (WLSTM) models were used for agricultural drought prediction in the Palakkad district of Kerala state, India. All the wavelet-coupled models showed reliable prediction performance during evaluation. The WLSTM performed exceptionally well for longer lead predictions making it a suitable choice for future drought forecasting. In comparison, WLSTM outperforms WCNN and WSVM, especially at higher leads of 5 and 6 months at all the grid points in the study area. The hybrid wavelet SVM model shows superior performance than simple SVM in predicting agricultural drought, exhibiting a 9% increase in the average R2 value at a 4-month lead time. This underscores the efficacy of wavelet denoising techniques. The wavelet CNN and LSTM models were also efficient in predicting drought duration and severity better than the traditional CNN and LSTM models. This study recommends a wavelet-based LSTM model for the long lead prediction of agricultural drought in the study area. The findings of the study will facilitate informed decision making and effective resource allocation, while also aiding in the development of robust early warning systems for drought forecasting.