Using MODIS and Landsat Data to Produce Land Use Land Cover Maps for the Lower Mekong Basin: Implications for Regional Water and Disaster Management

The people, ecosystems, wildlife, and fisheries of the Lower Mekong Basin (LMB) are subjected to many threats, including severe storms, flooding, drought, hydrologic modification (e.g., river damming), pollution, and Land Use Land Cover (LULC) change. Such threats are being addressed in part by water and disaster management in the LMB that includes use of the Soil & Water Assessment Tool (SWAT) hydrologic model. The latter employs LULC maps as an input, given that LULC types can affect surface runoff and streamflow. Such LULC types include agricultural, forest, urban areas, water, and industrial forest plantation classes.

There are multiple data sources and methods to derive LULC maps. While it may be advantageous to derive LULC maps using a single method and data source, such a "one size fits all" approach may not be adequate for mapping the LULC types needed for aiding hydrologic models, such as SWAT. In addition, many LULC types have distinct vegetation greenness phenology and are difficult to classify with single season dates of satellite data. Cloud free Landsat data may only be available during the LMB's dry season. On the other hand, MODIS data cannot typically map LULC types that are spatially fine scaled, such as urban areas and industrial forest plantations. In response, a multi-sensor approach was employed to derive 2010 LULC maps in which monthly MODIS NDVI and Landsat data from the dry season were used to classify LULC types within Sub-Basins (SBs) 1 through 8 of the LMB, as defined by the Mekong River Commission (MRC). MODIS data was used with unsupervised classification techniques to map regionally evident LULC types with distinct vegetation phenology (e.g., forest and agricultural types). Landsat data was used with unsupervised and supervised methods to map locally common yet regionally scarce LULC types used in the MRC SWAT model. LULC map accuracy was assessed for SBs 4 and 7 via comparison of stratified random sample locations on test LULC maps and available reference data. The overall agreement for the two tested LULC maps both exceeded 80% at full scheme specificity (16 classes for SB 4 and 14 classes for SB 7). The 2010 LULC maps from the project were used to update the circa 1997 LULC maps used previously in MRC SWAT models. The updated SWAT models are being used to aid regional water and flood disaster management.