Modeling sensitivity of topographic change with sUAS imagery

Abstract

Mapping the topographic surface and monitoring the change in such topographic surfaces has largely been a remote sensing-based solution for the last eighty years. The last few years has seen the dramatic rise in the use of small unmanned aerial systems (sUAS) for mapping both the topographic surface (in largely un-vegetated areas) and particularly, the overlying surface layer. How accurate are the sUAS derived elevation surfaces? How accurate are the change surfaces from a comparison of multi-date surfaces? How confident can the user be in the changes detected? Probing the expected accuracy for a topographic surface derived from low altitude sUAS imagery is a tad more problematic than many other types of remotely sensed imaging sensors. The precision and spatial resolution of the sUAS imagery, and subsequent digital elevation models (DEMs) is similar to the precision and spatial resolution of the very reference data sources used for accessing accuracy. In this research an approach was used to evaluate the performance of sUAS for creating digital elevation models on coastal sand dunes that did not change during ten repeat aerial collections at 40 m above ground level. A simple error budget model was used to empirically derive the intrinsic accuracy of the sUAS-derived topographic surface. The overall accuracy of the ten DEMs derived from independent aerial missions was 0.033 m root mean squared error (RMSE). The results indicate a confidence threshold of ~0.030 m can be typically used to separate 95% of the ‘false’ topographic changes mapped from two digital elevation models in this collection/processing context. By modeling and removing the reference data error (i.e. survey grade global navigation satellite systems (GNSS)-derived validation points) the average accuracy of the ten DEMs was 0.022 m (RMSE).