University of Twente Student Theses


Automating surface water detection for rivers : the estimation of the geometry of rivers based on optical earth observation sensors

Thissen, J.J.M. (2019) Automating surface water detection for rivers : the estimation of the geometry of rivers based on optical earth observation sensors.

[img] PDF
Abstract:This thesis studies the extent to which the geometry of rivers around the globe can be determined in an automated manner, based on openly available optical Earth Observational (EO) satellite sensors. Knowledge of the course of a river allows for the bathymetry to be estimated. Currently, the course of a river, at any given point in time, is derived by means of visual inspection based on geographical maps, which may be time-consuming. Automating this process can therefore be highly beneficial. The utilization of remote sensing technology is investigated to observe the dynamics of rivers at frequent time intervals. Google Earth Engine is adopted for the analysis, interpretation and manipulation of multispectral satellite data. In order to detect surface water within a multispectral satellite image, water indices are utilized, followed by a HAND-map and an image thresholding approach (Otsu’s method). The purpose of a water index is to enhance water features, the HAND-map to avoid potential errors due to shadows, and Otsu’s method to reduce the resulting grayscale image to a binary surface water mask, separating foreground (water) from background features. To estimate the bounds of a river under cloudy circumstances, multiple historical images are sampled to generate a composite, representing the water occurrence of a river over time in the form of a binary image. Historical images are weighted based on the date they were sampled, in order to reduce the impact of varying surface water widths over time. The composite is used to estimate segments of a river that were initially unknown due to the presence of clouds. The resulting surface water mask is converted to a smooth river polygon. River widths are subsequently derived based on a Euclidian distance map in combination with a centerline, which is obtained by extracting a skeleton from a Voronoi diagram, followed by a pruning procedure. The approach has been validated based on two river polygons provided by Rijkswaterstaat, representing a segment of the Meuse and the Rhine. The estimated geometries based on EO satellite imagery were found to be highly similar to that of the two river polygons. Throughout 2016, for a segment of the Meuse, deviations between +2 and -2 meters (+1.53% and -1.53%) as well as +1 and -8 meters (+0.76% and -6.11%) were found compared to data provided by Rijkswaterstaat, for cloudless and cloudy images respectively. For the segment that comprises of the Rhine and the Waal, deviations between +29 and -15 meters (+7.36% and -3.81%) as well as +17 and -13 meters (+4.31% and -3.30%) were found for cloudless and cloudy images respectively, for the same year. Furthermore, a comparison towards a recently build database consisting of global river widths from Landsat imagery (GRWL; Allen and Pavelsky, 2018) revealed clear similarities. Although the estimation of the geometry of rivers in a more or less automated manner was found to be achievable, its global applicability remains limited to a local scale. Currently, the biggest limitation is the fact that the amount of usable memory is capped within GEE. Computations were found to be too resource-demanding, limiting the maximum size of a river polygon that can be generated. Furthermore, the estimation of the geometry of a river is found to be limited to rivers that are at least three to four times wider than the corresponding satellite’s spatial resolution in order to obtain usable results. Rivers that are relatively relative narrow (<100 meters) were found to be difficult to identify using either Sentinel-2 or Landsat 8 multispectral satellite data.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:56 civil engineering
Programme:Civil Engineering and Management MSc (60026)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page