The characterisation and quantification of the migration pattern of the Kapuas river, by determining erosion and deposition from remotely sensed raster images.
Observing river migration development from Landsat images: application to the Kapuas River
Supervisors: Bart Vermeulen, Muhammad Pramulya and Ton Hoitink
The Kapuas River is a tropical meandering river located in West-Kalimantan, Indonesia. Two serial objectives are set for this research. First, the objective is to quantify the error made when delineating a river bank from a raster image such as obtained from Landsat. After obtaining a generic insight into the accuracy of riverbank lines inferred from Landsat images, the objective is to reveal migration of the Kapuas River over time and to ground truth the results from satellite data.
In the theoretical analysis, a synthetic river planform is created using Kinoshita curves. For each synthetic river a grid is created with a predefined resolution. The error in the riverbank obtained from the raster image is then calculated by delineating the riverbanks from the raster image created, and establishing the difference with the input planform. Three conclusions can be drawn from the theoretical analysis. The theoretical riverbank error is on average 0.2 cell sizes over the length of the synthetic river. Secondly, the curvature is influencing the riverbank error: the higher the curvature, the larger the riverbank error. Finally, the filter method influences the riverbank error. A second order polynomial function like Sgolay or Loess with a distance between the neighbouring cells of 41 points yields the smallest riverbank error.
The Kapuas River is classified and delineated using Landsat images taken in six years: 1973, 1990, 1995, 2000, 2005 and 2013. For the downstream part of the study area, the riverbank error is estimated using SideScan sonar data. In an upstream part of the river, ground truth data was collected using a handheld GPS. The riverbank error between the classified riverbank from the most recent Landsat image and field observations is 0.3 times the cell size in both study areas, which exceeds the theoretical riverbank error due to noise (clouds and haze), water level variation and effects of riverbank erosion or accretion. The riverbank error established from the GPS measurements is higher for the aggrading riverbanks than for the eroding and stable riverbanks. The river bank delineations obtained for from the Landsat images from 1973, 1990 and 2013 are used to define the migration rates over time. As a result of the riverbank error amounting to 0.3 times the cell size, it is impossible to detect migration that has occurred between two images separated by 5 years of time, since the migration rate of the Kapuas is around 1 m/y. In the upstream area, migration is higher (1.8 m/y) than the downstream area (0.9 m/y). The upstream area may be more active due to the presence of peat, the buffer capacity of the lakes, side streams entering the Kapuas, or a combination of these factors. A stream reconnaissance for the upstream area around Putissibau has been performed, yielding 1200 photos with GPS locations classified into four bank categories: erosive, aggrading, stable and embayment. Over 90% of the observed points give the correct riverbank type as compared to the results from the Landsat images taken between 1990 and 2013. Both field data and analyses from the Landsat images show the same riverbank type for the upstream area. The downstream study area migration pattern is not referenced in the field due to high water levels during the field period, causing only a small part of the river bank to be exposed. Here, the change of the riverbank is too small to be visible in the Landsat image.