An Object-Conditional Land Cover Classification System (LCCS) Wetland Probability Detection Method for West African Savannas Using 250-Meter MODIS Observations

Kurzfassung

Within the GLOWA Volta river basin hydrology project in West Africa, a unique decision hierarchy based on FAO LCCS definitions was derived for two exemplary wetland sites in Burkina Faso. Primary aim is to infer wetland vegetation structural spaces for further biodiversity assessments and to simultaneously align these spaces to the FAO LCCS standard legend definitions on semi-natural regular flooded woody and herbaceous areas. We utilized 250-meter resolution 16-day Normalized Differential Vegetation Index (NDVI) as well as near infrared (NIR) time-series composite metrics from the moderate resolution MODIS sensor, to calculate harmonic vibrations/frequencies and NDVI dynamics or first derivatives between the 16-day NDVI MODIS composites. The MODIS metrics were corrected for cloud, shadow and other noise by an in house developed data segmentation tool (TiSeg). The MODIS time-series data were processed for the years 2000 to 2005, and used in conjunction with digital elevation data variables from the SRTM topographic mission. The MODIS wetland features were verified from wetland types mapped using contemporary high resolution 30-meter Landsat data and 4-meter Quickbird data. We found the first (January) and last (December) time measures of the NDVI harmonics in any particular year from 2000 to 2005 to be most useful in separating wetland spaces that are characterized by increased chlorophyll production. The wet season (July to October) near-infrared (NIR) MODIS time-series harmonics were well suited to discriminate near to permanent flooded herbaceous wetland spaces. Using the dynamics that is first derivative NDVI observations we further mapped highly dynamic wetland spaces, mostly changes of great magnitude from flooded herbaceous to bare soil. The location of topographic pits and streamline areas from the 90-meter radar SRTM data allowed further discrimination of wetland regimes. Herbaceous and woody regularly flooded areas as per LCCS were accurately mapped; regular flooded woody were mapped with an error of omissions score of between 9 percent and 5 percent, and herbaceous regular flooded spaces were mapped with an accuracy of 11 percent omission. Well corrected time series of 250-meter MODIS NDVI phonologies and NIR band observations, NDVI derivatives and topography variables are most powerful in mapping and characterizing wetland spaces in West African savannas that are related to biodiversity significances.