| This article has been：Browse 335Times Download 64Times |
scan it! |
|
|
| DOI：10.13522/j.cnki.ggps.2025289 |
|
| Estimating maize plant water content based on UAV multispectral-thermal infrared fusion and machine learning |
|
Zhang Hao, Wang Zuji, Li Caixia, Ma Yanchuan, James E. Kanneh, Zhong Daokuan, Li Shenglin
|
|
1. North China University of Water Resources and Electric Power, Zhengzhou 450046, China; 2. Institute of Farmland Irrigation, Chinese Academy of Agricultural Sciences, Xinxiang 453002, China; 3. Shandong Agricultural University, Tai’an 271018, China;
4. Henan Institute of Science and Technology, Xinxiang 453003, China
|
| Abstract: |
| 【Background & Objective】Crop water content is a critical indicator for characterizing crop water stress and physiological status; its accurate estimation is essential for agricultural water management, drought monitoring, and crop growth assessment. Traditional ground-based crop water monitoring methods are costly and rely on limited point-observations, which cannot support large-scale and real-time drought diagnosis. The advances in unmanned aerial vehicle (UAV) remote sensing provide a non-destructive, large-scale alternative to conventional methods. However, UAV-derived spectral features mainly reflect static crop structural properties and have a response delay to crop water stress. Canopy temperature directly characterizes crop physiological changes. In this context, this study proposed a method that integrates multi-source UAV remote sensing features to construct a high-precision, real-time crop water content estimation model.【Method】The method fused UAV-acquired multispectral and thermal infrared images to construct a multi-source data fusion model. Three machine learning algorithms, including Random Forest (RF), K-Nearest Neighbor regression (KNN), and Support Vector Regression (SVR), were used to estimate the variation in plant water content (PWC) of maize throughout its growing season.【Result】①The KNN algorithm combined with NDVI + GNDVI + RVI was most accurate for estimating PWC, with a coefficient of determination (R2) of 0.713 and a relative root mean square error (rRMSE) of 6.610%. ②Based on integrated multi-source spectral-thermal indices, the RF algorithm significantly outperformed SVR and KNN in PWC estimation, with an R2 of 0.887 and an rRMSE of 4.298%. ③The combination of spectral features (NDVI + GNDVI + RVI) and thermal feature (LST) coupled with the RF algorithm effectively captured the spatiotemporal dynamics of PWC in the study area.【Conclusion】The fusion of multi-source UAV remote sensing data combined with the RF algorithm can accurately estimate maize plant water content; it has good adaptability for crop water estimation under multidimensional feature conditions and offers a reliable technical support for crop water monitoring and drought assessment. |
| Key words: plant water content; multispectral; thermal infrared; random forest; vegetation indices |
|
|
