Optical flow algorithm as estimator of horizontal discrepancy between features derived from DEMs: rivers and creeks as case study

Abstract

DEMs (grid digital elevation models) are used for a broad spectrum of applications, some of which require deriving features such as drainage networks (rivers, creek, etc.). A precise positioning in the features taken from DEMs is necessary, but frequently DEMs are not homogeneous (e.g. the mapmaker sources vary, cell sizes differ), so that the expected precision fluctuates. Therefore, a measure to estimate the discrepancy between features built from different DEMs would be useful. In particular, we focus on the horizontal discrepancy (HD), which is the lesser studied discrepancy in the literature. Our approach is based in the optical flow (OF) algorithm, which has been used successfully in object movement detection in consecutive images or video records. We establish the analogy between an image and a DEM because both are composed of regular elements, pixels, and cells. The variation in the pixel value between two consecutive images is used by OF to compute movement. We use the variation in the DEM cell value (height) to apply the OF and to estimate the HD between rivers and creeks existing in our DEMs examples. In our study, OF proved to be a good estimator of HD when features were derived from hill and mountain terrain, but was not reliable when the terrain was almost flat. However, most studies in the research literature have indicated that nearly flat terrain poses the most difficulties in forecasting positioning errors. Therefore, we conclude that the OF is a good estimator of the HD between features derived from DEMs.