Scholarlink

<p>Accurate estimation of root-zone soil moisture (SM) is critical for agricultural water management and sustainable crop production. This study develops and evaluates a methodology to estimate SM in sorghum root zones by high-resolution unmanned aerial vehicle (UAV) multispectral and thermal imagery with machine learning (ML). A two-season field experiment (2020–2021) with four irrigation regimes provided UAV data and concurrent ground-based SM measurements. This study conducted a comparative analysis of four ML algorithms: Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGB), and K-Nearest Neighbors (KNN). The models were evaluated both as standalone predictors and as components of ensemble structures. Among single models, RF achieved the highest test performance (R² = 0.84, RMSE = 11.22&#xa0;mm/90&#xa0;cm, MAE = 9.32&#xa0;mm/90&#xa0;cm). An ensemble combining XGBoost, LGB, and KNN yielded a slight improvement (R² = 0.85, RMSE = 11.124&#xa0;mm/90&#xa0;cm, MAE = 8.775&#xa0;mm/90&#xa0;cm), indicating that ensemble learning can modestly enhance model performance. The proposed workflow offers a practical approach for field-scale SM monitoring, demonstrating potential applications in irrigation scheduling and agricultural water management.</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

Integration of UAV images and ensemble learning for root zone soil moisture estimation in sorghum

  • 期刊论文

  • 发表日期：

    2025年11月28日

查看全文
  • Emre Tunca,
  • Eyüp Selim Köksal,
  • Sakine Çetin Taner

摘要

Accurate estimation of root-zone soil moisture (SM) is critical for agricultural water management and sustainable crop production. This study develops and evaluates a methodology to estimate SM in sorghum root zones by high-resolution unmanned aerial vehicle (UAV) multispectral and thermal imagery with machine learning (ML). A two-season field experiment (2020–2021) with four irrigation regimes provided UAV data and concurrent ground-based SM measurements. This study conducted a comparative analysis of four ML algorithms: Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGB), and K-Nearest Neighbors (KNN). The models were evaluated both as standalone predictors and as components of ensemble structures. Among single models, RF achieved the highest test performance (R² = 0.84, RMSE = 11.22 mm/90 cm, MAE = 9.32 mm/90 cm). An ensemble combining XGBoost, LGB, and KNN yielded a slight improvement (R² = 0.85, RMSE = 11.124 mm/90 cm, MAE = 8.775 mm/90 cm), indicating that ensemble learning can modestly enhance model performance. The proposed workflow offers a practical approach for field-scale SM monitoring, demonstrating potential applications in irrigation scheduling and agricultural water management.