Scholarlink

<p>Climate change increases the variability of drought-induced crop yield and leads to serious food security threats, particularly in wheat-harvesting arid lands such as South Punjab in Pakistan. Despite existing research on long-term and multi-index analysis of remote sensing data of drought monitoring, the use of machine learning (ML) concerning monthly yield losses during crop growth stages is not very well studied, especially involving data from Pakistan. This study evaluates the performance of seven ML models, including Random Forest (RF), Extreme Gradient Boost (XGBoost), Decision Tree (DT), Support Vector Machine (SVM), Na ve Bayes (NB), and a hybrid model, Convolutional Neural Network-Recurrent Neural Network (CNN-RNN) in predicting wheat crop yield using drought indices such as Normalized Difference Vegetation Index (NDVI), Temperature Condition Index (TCI), Normalised Difference Water Index (NDWI), Vegetation Health Index (VHI), Palmer’s Drought Severity Index (PDSI), and Vegetation Condition Index (VCI) for the period of 2001 to 2022. The drought indices are calculated on monthly stages of wheat crop production. Three Moderate Resolution Imaging Spectroradiometer (MODIS) products (MOD13Q1V6, MOD13A2, MOD11A2) and a Terra Climate product are used to calculate the drought indices data. XGBoost and DT models demonstrated the highest classification accuracy, followed by RF, CNN-RNN, and SVM, achieving accuracy of 98%, 96%, and 94%, respectively. Analysis indicates an increasing risk of wheat yield loss over the time period, mainly influenced by climate change and resulting drought conditions. These findings highlight the capability of ML models to capture latent spatial and temporal patterns in drought-related datasets. Analysis revealed variations in drought indices during the crop-growing season, highlighting their utility in monitoring wheat growth stages. The yield loss risk of wheat indicated an increasing trend from 2001 to 2022. In addition, Pearson’s correlation analyses demonstrated significant relationships between drought indices and wheat yield. The high classification accuracy of these models can help policymakers and farmers build early warning systems to support decision-making for irrigation planning and enhance drought resilience strategies in vulnerable agricultural regions.</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

Enhancing drought monitoring through multi-stage crop yield and machine learning

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Fiaz Majeed,
  • Hira Ahmad,
  • Ansar Siddique,
  • Mahrukh Iftikhar,
  • Tahir Khurshaid,
  • Nagwan Abdel Samee,
  • Imran Ashraf

摘要

Climate change increases the variability of drought-induced crop yield and leads to serious food security threats, particularly in wheat-harvesting arid lands such as South Punjab in Pakistan. Despite existing research on long-term and multi-index analysis of remote sensing data of drought monitoring, the use of machine learning (ML) concerning monthly yield losses during crop growth stages is not very well studied, especially involving data from Pakistan. This study evaluates the performance of seven ML models, including Random Forest (RF), Extreme Gradient Boost (XGBoost), Decision Tree (DT), Support Vector Machine (SVM), Na ve Bayes (NB), and a hybrid model, Convolutional Neural Network-Recurrent Neural Network (CNN-RNN) in predicting wheat crop yield using drought indices such as Normalized Difference Vegetation Index (NDVI), Temperature Condition Index (TCI), Normalised Difference Water Index (NDWI), Vegetation Health Index (VHI), Palmer’s Drought Severity Index (PDSI), and Vegetation Condition Index (VCI) for the period of 2001 to 2022. The drought indices are calculated on monthly stages of wheat crop production. Three Moderate Resolution Imaging Spectroradiometer (MODIS) products (MOD13Q1V6, MOD13A2, MOD11A2) and a Terra Climate product are used to calculate the drought indices data. XGBoost and DT models demonstrated the highest classification accuracy, followed by RF, CNN-RNN, and SVM, achieving accuracy of 98%, 96%, and 94%, respectively. Analysis indicates an increasing risk of wheat yield loss over the time period, mainly influenced by climate change and resulting drought conditions. These findings highlight the capability of ML models to capture latent spatial and temporal patterns in drought-related datasets. Analysis revealed variations in drought indices during the crop-growing season, highlighting their utility in monitoring wheat growth stages. The yield loss risk of wheat indicated an increasing trend from 2001 to 2022. In addition, Pearson’s correlation analyses demonstrated significant relationships between drought indices and wheat yield. The high classification accuracy of these models can help policymakers and farmers build early warning systems to support decision-making for irrigation planning and enhance drought resilience strategies in vulnerable agricultural regions.