A sensitivity analysis of Rillstats for soil erosion estimates from UAV derived digital surface models.

Fertile topsoil is being eroded ten times faster than it is created which can result in lowered crop yields, increased river pollution, and heightened flood risk (WWF 2018). Traditional methods of soil erosion monitoring are labour-intensive and provide low resolution, sparse point data not representative of overall erosion rates (Báčová et al., 2019). However, technological advances using Uncrewed Aerial Vehicles (UAVs) obtain high-resolution, near-contactless data capture with complete surface coverage (Hugenholtz et al., 2015).  

 

Typically, analysing UAV-Structure-from-Motion (SfM) derived soil erosion data requires a survey prior to the erosion event with repeat monitoring for change over time to be quantified. However, in recent years the ability of soil erosion estimations without the pre-erosion data has emerged. Rillstats, which is specifically designed to quantify volume loss in rills/gullies, has been developed by Báčová et al., (2019) using the algorithm and Python implementation in ArcGIS to perform automatic calculations of rills. Although this technique has been developed, it is not yet tested. 

This research evaluates the sensitivity of Rillstats to estimate soil erosion volumes from Digital Surface Models (DSM) obtained using a DJI Phantom 4 RTK UAV. The aims of the research were to test i) the influence of UAV-SfM surveys with varying flight settings and environmental conditions and ii) the effect of the size and shape of the boundary polygon. Results will be presented that analyse the sensitivity of estimations of soil erosion to changes in DSM resolution, image angle, lighting conditions, soil colour and texture to develop recommendations for a best practice to optimize results.