Analysis of optimal thresholds for identification of open water using MODIS-derived spectral indices for two coastal wetland systems in Mexico

Timely information on the extent of open water surfaces and wetland dynamics can be useful for decision makers to rapidly respond to flood or drought events and improve our understanding of large scale eco-hydrological variability. Episodic inundation is an important factor for biological and ecosystem processes and sustains a wealth of ecosystem services. This study tests 14 spectral indices and their appropriate thresholds for water mapping, using 500 m resolution MODIS surface reflectance data for two coastal sites in Mexico. The fraction of water within each MODIS pixel was estimated using 30 m Landsat-derived water masks. The error introduced by upscaling to coarser resolutions as well as the error associated with each index threshold was analyzed through an omission-commission error space. The lowest cost according to a hyperbolic function, and the lowest area difference relative to the reference water area served as criteria for definition of optimal fractions and identification of the most appropriate indices and their optimal thresholds. Our study shows that the normalized difference between the MODIS green (band 4) and short wave infrared (band 6) MNDWI6 consistently performed best at mapping of water. MNDWI6 showed little variations of the optimal index value (-0.14 +/- 0.05) and a low sensitivity to changes in the threshold value. Other useful indices for water extraction identified in this study are the MNDWI5, MSAVI2 and Tasseled Cap Wetness Indices.