Estimating leaf chlorophyll content of crops via optimal unmanned aerial vehicle hyperspectral data at multi-scales

Abstract Leaf chlorophyll content (LCC) is a crucial indicator of nutrition in crop plants and can be applied to assess the adequacy of nitrogen (N) fertilizer for crops while reducing N losses to farmland. This study estimated the LCC of maize and wheat, and comprehensively examined the effects of the spectral information and spatial scale of unmanned aerial vehicle (UAV) imagery, and the effects of phenotype and phenology on LCC estimation. A Cubert S185 hyperspectral camera onboard a DJI M600 Pro was used to conduct six flight missions over a long-term experimental field with five N applications (0, 70, 140, 210, and 280 kg N ha−1) and two irrigation levels (60% and 80% field water capacity) during the growing seasons of wheat and maize in 2019. Four regression algorithms, that is, multi-variable linear regression, random forest, backpropagation neural network, and support vector machine, were used for modeling. Leaf, canopy, and hybrid scale hyperspectral variables (H-variables) were used as inputs for the statistical LCC models. Optimal H-variables for modeling were determined by Pearson correlation filtering followed by a recursive feature elimination procedure. The results showed that (1) H-variables at the canopy- and leaf-scales were appropriate for wheat and maize LCC estimation, respectively; (2) the robustness of LCC estimation was in the order of the flowering stage > heading stage > grain filling stage for wheat and early grain filling stage > flowering stage > jointing stage for maize; (3) the reflectance of the red edge, green, and blue bands were the most important inputs for LCC modeling, and the optimal vegetation indices differed for the various growth stages and crops; and (4) all four algorithms maintained an acceptable accuracy with respect to LCC estimation, although random forest and support vector machine were slightly better. This study is valuable for the design of appropriate schemes for the spectral and scale issues of UAV sensors for LCC estimation regarding specific crop phenotype and phenology periods, and further boosts the applications of UAVs in precision agriculture.