Using a low-cost drone to assess herbaceous biomass and quality in the Sahelian Rangeland ecosystems

Abstract

Existing ways of assessing rangeland plant biomass and nutritional quality mostly rely on field surveys, which are difficult to generalise across plots, along with laboratory-based techniques that entail lengthy pre-processing procedures. As a solution, drones have emerged as a promising tool capable of collecting low-altitude images over expansive areas with minimal effort and cost. Here, we explore the potential of low-cost drone images to estimate the rangeland biomass, and quality of the Sahelian rangeland ecosystem. Model calibration and validation were conducted using a random forest machine learning algorithm, where the response variables were field vegetation samples, and the explanatory variables were derived from drone image outputs. A principal component analysis (PCA) was conducted to explore the multivariate relationships between drone-derived vegetation indices and field-measured biomass and quality attributes. In the validation datasets, the random forest model exhibited relative root mean squared errors (RRMSE) of 31% for fresh mass and 37% for dry mass. The random forest model demonstrated a relatively high prediction accuracy, yielding RRMSE values of 32% for crude protein, 9% for neutral detergent fibre, 8% for acid detergent fibre, and 17% for organic matter digestibility contents. The PCA revealed that the first two components explained 53.3% of the total variance. Overall, these results showed that red, green and blue (RGB) images acquired from low-cost drones can be used to estimate rangeland biomass and quality.