Herbaceous biomass predication from environmental and remote sensing indicators

Abstract

Feeding patterns and distribution of herbivores animals are known to be influenced by quality and quantity of forage such as grass. Modelling indicators of grass quality and biomass are critical in understanding such patterns and for decision makers such as park managers and farmers to efficiently plan and manage their rangelands. This study focused on predicting grass biomass using remote sensing and environmental variables. Since some of these variables were highly correlated, multivariate techniques such as partial least squares (PLS) and ridge regression were used to predict grass biomass in the Kruger National Park and the surrounding areas. The results indicated that both the environmental and remote sensing indicators had potential to predict grass biomass. Ridge regression showed better results since it explained about 41% of variation in the grass biomass, compared to the PLS model which explained approximately 33% variation.