Scholarlink

<p>The rapid spread of the competing weed <i>L. arvensis</i> poses a major threat to wheat production; therefore, modern risk assessment methods are necessary for its management. This study developed and compared machine learning models (Random Forest [RF], Boosted Regression Trees [BRT], and Maximum Entropy [MaxEnt]) to evaluate habitat suitability for <i>L. arvensis</i>, a dominant weed in the wheat cropping systems of Pakistan’s semi-arid regions. For this purpose, weed data from 402 wheat fields, along with 20 environmental factors, including topography, climate, soil characteristics, anthropogenic factors, and proximity metrics, were analysed. Soil texture (silt and clay), soil chemistry (EC, OM, TDS), and rainfall patterns were identified through a partial least squares (PLS) algorithm as major factors affecting the species distribution. The ROC–AUC results showed that MaxEnt (AUC = 0.93) and RF (AUC = 0.92) performed slightly better than BRT (AUC = 0.86). All models identified the eastern and southeastern regions as the main areas of highly suitable habitat. Although these models are reliable, their predictions may be affected by changes in environmental factors in cropland. These results demonstrate that machine learning methods are effective for mapping weed distribution and provide a scientific foundation for sustainable weed management in these 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

Predicting habitat suitability of weed species Lysimachia arvensis (L.) U. Manns & Anderb. in wheat cropping systems using machine learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Qurat Ul Ain,
  • Emran Dastres,
  • Mohsen Edalat,
  • Shujaul Mulk Khan

摘要

The rapid spread of the competing weed L. arvensis poses a major threat to wheat production; therefore, modern risk assessment methods are necessary for its management. This study developed and compared machine learning models (Random Forest [RF], Boosted Regression Trees [BRT], and Maximum Entropy [MaxEnt]) to evaluate habitat suitability for L. arvensis, a dominant weed in the wheat cropping systems of Pakistan’s semi-arid regions. For this purpose, weed data from 402 wheat fields, along with 20 environmental factors, including topography, climate, soil characteristics, anthropogenic factors, and proximity metrics, were analysed. Soil texture (silt and clay), soil chemistry (EC, OM, TDS), and rainfall patterns were identified through a partial least squares (PLS) algorithm as major factors affecting the species distribution. The ROC–AUC results showed that MaxEnt (AUC = 0.93) and RF (AUC = 0.92) performed slightly better than BRT (AUC = 0.86). All models identified the eastern and southeastern regions as the main areas of highly suitable habitat. Although these models are reliable, their predictions may be affected by changes in environmental factors in cropland. These results demonstrate that machine learning methods are effective for mapping weed distribution and provide a scientific foundation for sustainable weed management in these regions.