Estimation of aboveground biomass production using an unmanned aerial vehicle (UAV) and VENμS satellite imagery in Mediterranean and semiarid rangelands

Abstract

Rangeland management requires frequent and accurate estimation of aboveground vegetation biomass (AGB) as a proxy of forage production. However, traditional methods for AGB measurement are time-consuming and only provide a low number of spatiotemporal measurements. Newly developed remote sensing platforms such as UAVs and new generation satellites allow unprecedented large-scale and frequent monitoring of rangeland vegetation that is typically produced in marginal lands characterized by high surface heterogeneity. This study used high-resolution UAV data and the new multi-spectral VENμS satellite to monitor AGB production for two consecutive years in a Mediterranean and semiarid grassland rangeland in Israel. We then studied the effects of grazing intensity and precipitation along the growing season on AGB. Data were collected from two long-term ecological research sites with plots under controlled grazing pressures (i.e., no grazing, moderate, and intensive grazing). We used high-spatial-resolution UAV imagery for land cover classification in order to mask different levels of mixed pixels in the VENμS satellite images along the growing season. A support vector regression model (SVM) for satellite-based AGB estimation was developed using more than 600 ground-truth AGB measurements collected during 2019–2020. The effect of the percentage cover for mixed pixel removal was examined. The results showed an improvement in the prediction model by 35% and reduced the RMSE by half for the support vector machine regression model's best prediction. The best AGB prediction results were achieved by including satellite pixels with over 50% vegetation cover. For the first time in Eastern Mediterranean rangelands, our study illustrates the benefits of high-integrated remote sensing data (i.e., satellite and UAV) for generating more accurate AGB estimations even in highly heterogeneous ecosystems in terms of surface cover. Such an approach can provide crucial and better information for sustainable rangeland management under varying spatiotemporal and climatic conditions.