Empirical Modeling of Leaf Chlorophyll Content in Coffee (Coffea Arabica) Plantations With Sentinel-2 MSI Data: Effects of Spectral Settings, Spatial Resolution, and Crop Canopy Cover

Abstract

Approaches to monitor coffee leaf Chl are necessary for field, landscape, regional, and national scale decision making to improve coffee stand health and productivity and to reduce yield losses. In this study, we investigated the influence of spectral settings, spatial resolution, and age-related crop canopy cover on leaf Chl modeling performance with Sentinel-2 MSI data. We ran the random forest algorithm to predict coffee leaf Chl using all nine Sentinel-2 MSI bands at 20 m, all nine bands at 10 m, five bands at 10 m, and four bands at 20 m spatial resolutions for all coffee stands (3–8 years, $N = 72$ ), and then for mature stands only (5–8 years, $N = 60$ ). Results showed that the best modeling results $(R^{{2}}= \text{0.69},\text{RMSE}= \text{6.8})$ were achieved when all the bands at 10 m spatial resolution were used in modeling coffee leaf Chl for all coffee stands. The prediction accuracy improved $(R^{{2}}= \text{0.77},\text{RMSE}= \text{5.9})$ when only mature coffee stands were used. The 20-m bands (red-edge and shortwave infrared) performed similarly in coffee leaf Chl estimation as using all bands at 20 m and using only 10-m bands for all coffee stands and for mature coffee stands. Accuracy metrics were highest and error values lowest when coffee leaf Chl was estimated for mature stands only compared to when all coffee stands were modeled. We concluded that Sentinel-2 MSI is a valuable dataset for predicting coffee Chl; however, based on our findings, we suggest that finer spatial resolutions of 10 m applied on mature coffee stands should be adopted in modeling coffee Chl.