Detection of inland water bodies with high temporal resolution - assessing dynamic threshold approaches

摘要

Information on the spatio-temporal dynamics of inland water bodies is of high value for many applications, for example in the context of water and land management or for ecosystem service assessments. In this study, different approaches to delineate inland water bodies from MODIS 250 m time series were compared. Here, the performance of different input bands and indices, of trainings pixel selection methods, and of dynamic threshold definition approaches were assessed with the goal to find an optimized approach applicable for global inland water body detection based on moderate spatial and high temporal resolution MODIS data. The results of the tested approaches were cross validated with high resolution Landsat-8 classifications. The results show amongst others that a combination of near infrared band (NIR) and difference index (NIR - red band) performed best for most of the globally distributed test regions and that single band approaches revealed higher commission errors.