Crop monitoring using multispectral and hyperspectral remote sensing

Abstract

Nutrient stress is a significant threat to crop yield quality and can cause soil infertility. In satellite-based remote sensing, accurately monitoring nutrient stress with optical sensors is often challenging due to spatial and temporal resolution constraints and atmospheric effects. To overcome the limitation of satellite data, UAV (Unmanned Aerial Vehicle)-based imaging is preferred. Hyperspectral imaging is a preferred approach for monitoring nutrient stress due to its sensitivity to plant chemical composition and its high spatial and temporal resolution. In this study, we performed an experiment involving nitrogen stress at underdose (60 kg/ha), recommended dose (120 kg/ha), and overdose (180 kg/ha) on wheat plants in a precision farm setup. The reflectance obtained from hyperspectral data allows for distinguishing between underdose (0.08–0.11 in 400 nm–700 nm, 0.3–0.38 in the NIR spectrum range), overdose (0.04–0.05 in 400 nm–700 nm, 0.42–0.55 in the NIR spectrum range), and recommended-dose (control) wheat plants. A comparative analysis was performed on data obtained with a UAV-based hyperspectral sensor and a handheld spectroradiometer for overdosed and underdosed nitrogen treatments. The results indicate a high correlation between the spectra obtained from both sensors, with r = 0.9841 for the underdose treatment and r = 0.9913 for the overdose treatment. Thus, a spectroradiometer can fill data gaps when hyperspectral data are unavailable. Additionally, we analysed pure vegetation pixels after masking out soil pixels for nutrient-sensitive indices (NDVI, NDRE, ND705, CLred−edge, and DATT). Our results indicate that ND705 and NDVI are optimal indicators for early nitrogen stress detection due to their high separability. The results were validated using an ANOVA test, in which ND705 had the lowest p-value (p = 0.007) for data acquired on 28 February 2025.