Unmanned aerial systems accurately map rangeland condition indicators in a dryland savannah

Dry savannahs are highly sensitive to climate change and under intense anthropogenic pressure. Therefore, the methods for assessing their status should be easy and repeatable. Monitoring through satellite data and field measurements are limited in accurately assessing the spatiotemporal dynamics of ecosystems. Fortunately, emerging technologies like Unmanned Aerial Systems (UAS) allow to transcend these limitations. But their calibration with field data for application in rangelands is still relatively new and less common than for example in precision agriculture. In this study we developed a drone-based workflow for mapping the condition of rangelands in dryland savannah. We evaluated how accurately and efficiently the two common indicators (i.e., potential forage biomass and rangeland cover type) of rangeland condition can be estimated from drone imagery across a range of conditions (i.e., highly degraded to healthy rangelands). To develop the drone-based potential forage biomass model we tested the accuracy of four vegetation indices to predict field biomass, with the optimized soil adjusted vegetation index (OSAVI) showing the highest prediction accuracy (R2 = 0.89 and RMSE = 194.05). The OSAVI-based model yielded a significant strong relationship (R2 = 0.80, p < 0.001) between predicted and field observed potential forage biomass across the rangeland system. For land cover, we applied a decision tree classification based on thresholds determined using data mining, with a mean overall accuracy of 95.8%. The drone-based estimates of bare cover, herbaceous cover and woody cover showed strong agreements (R2 ranging between 0.86 and 0.97) with the two image-truthing methods (line-point intercept and visual estimations) tested. We show that the drone-based approach is more efficient, unbiased, and repeatable than the field methods. Based on these results, the drone-based workflow presented here offers a reproducible, accurate and efficient approach for near-real time monitoring of rangeland condition at a landscape level. This may assist with climate-adapted management to prevent further land degradation and associated threats to biodiversity and human livelihoods.