Biomass estimation of a high Andean plant community with multispectral images acquired using UAV remote sensing and Multiple Linear Regression, Support Vector Machine and Random Forests models

Remote sensing with large-scale satellite images for precision studies in grasslands has spatial and spectral resolution limitations. Against this, using spectral signs and vegetation indices obtained with microsensors transported by unmanned aerial vehicles (UAV) constitutes a more accurate alternative for biomass estimation. In the fieldwork, images were acquired with microsensors, and fixed transects of 100 m were used where vegetation samples were collected. The photographs acquired with the UAV were processed in Pix 4D, Arc Gis, and algorithms elaborated in R programming language. The biomass estimation was carried out with Multiple Linear Regression, Vector Support Machine, and Random (Forest Random) models. The Random model showed a Kappa coefficient of 0.94 in the training set and 0.901 in the test set (R2 = 0.482). The Random Forest model predicted 3 g/pixel of MV for Puna grass in the rainy season and 2 g/pixel for the dry season; the predicted biomass for the Tola bush was 15 g/pixel of MV for both seasons of the year. The estimation of biomass/hectare for the tolar plant community with its tola shrub and Puna grass components was 6,535.88 kg/ha for the rainy season and 6,588.81 kg/ha for the dry season. The difference between the biomass estimated in the field and the biomass estimated with Random Forest was 5.48% for the rainy season and 9.63% for the dry season.