|
|
| Methods for Mapping, Characterization, and classification of the wetlands of Limpopo River basin using using Landsat ETM+ and SRTM Data |
Kulawardana, W., Thenkabail, P. S., Musiyadima, M., Biradar., C.,
Vithanage., J., Islam., A., Finlayson., M., Gunasinghe, S., Alankara., R.
Abstract
The International Water Management Institute (IWMI) has embarked on Global wetland Mapping project (GWMP) using remote sensing and secondary data. The overarching goal of the GWMP is to map, characterize, and classify the wetlands of the world at various scales or pixel resolutions through wide range of partnerships including the Ramsar Convention. Given the complexity and costs involved in mapping vast areas, the need to develop methods using freely available data was a top priority. Within this context, the goal of this paper was to develop a methodology for mapping wetlands using freely available high-resolution Landsat ETM+ 30-m Geocover and SRTM 90-m data. The Landsat Geocover for nominal year 2000 and the SRTM data, both having global coverage, were selected so the methods developed at one location can be applied elsewhere. The methodology was developed for the Limpopo river basin in Southern Africa. The Limpopo River basin is one of the major trans-boundary basins within the Southern African Region and occurs in four countries: South Africa, Mozambique, Zimbabwe and Botswana. |
Delineating the wetland boundary is the overwhelming component of wetland mapping and is time consuming. Automated and semi-automated methods were investigated for delineating the wetlands boundaries. The automated methods consisted of: (a) slope derived from SRTM, (b) tassel cap wetness index (TCVIs), (c) normalized difference water index (NDVIs), (d) multi-band vegetation indices (MBVIs), (e) two band vegetation indices (TBVIs), and the (f) normalized difference vegetation index (NDVI). Various threshold values of these indices were used to separate wetlands from non-wetlands. However, the best of these indices provided only an accuracy of less than 30 percent. The errors of omissions or errors of commissions were very high. The results clearly indicated inability of automate methods to accurately delineate the wetland boundaries.
Semi-automated techniques involved image enhancement, display, and digitizing techniques. The most useful displays of ETM+ image enhancements (e.g., ratios) and band combinations, displayed as RGB, were: (a) ETM+4/ETM+7, ETM+4/ETM+3, ETM+4/ETM+2, (b) ETM+4, ETM+3, ETM+5, and (c) ETM+3, ETM+2, ETM+1. The SRTM slope threshold of < 1 percent was very useful in delineating higher-order wetland boundaries. Once the image is enhanced and displayed, it is “zoomed in” and the wetland boundaries are screen digitized. The accuracy of such delineation, based on ground-truth data, was 86.4 percent; with an additional 7.7 percent of ground-truth points falling within 1 pixel (30-m) of wetland area. The total wetland area in the limpopo river basin was 5.2 million hectares (or 12.5 percent of the total total basin area which is 41.5 million hectares). However, the distribution of wetlands varied significantly: with lows of 4.2 percent in Botswana (of 8 million hectares of basin area) and 3.8 percent in Zimbabwe (6.1 Mha)- both in the upper reaches of the basin. In south Africa 8.9 percent (of 8.6 Mha), where the wetlands are in the middle and upper reaches of the basin. In the lower reach flood plains of Mozambique the wetlands were as high as 24.7 percent (of 8.8 Mha) which is comparable to Ruhuna basin in Sri Lanka which had a 24 percent.
The wetlands were classified into a 4-class and 8-class map and is dominated by (a) grasslands (33.8 percent), (b) riparian vegetation (35.9 percent), (c) farmlands and natural vegetation mosaic (25.3 percent), and (d) water body and marshland wetlands (5 percent). The overall accuracy was high (82 percent), errors of omissions reasonable (20 percent), and errors of commissions low (12 percent). The irrigated areas in limpopo was negligible and most of the wetlands were unutilized for agriculture. The study provides time-series characteristics of each wetland class over space and time.
|
|
|
Click on image to access/download the High-Res version (83.0 MB)
|
|
|
|
