Markov random field model-based soil moisture content segmentation from MODIS satellite data

The soil moisture (SM) content plays an important role in hydrology, agronomy, and meteorology. We propose to estimate the type of soil moisture content. This estimation is modeled as a Markov random field over which a regression of NDVI and LST MODIS data is constructed into Gaussian distributions. Under this model, the estimation of SM types is achieved by the maximum a posteriori (MAP) segmentation of MODIS data. Experimental results show that our ICM based on regression of MODIS NDVI and LST data can successfully segment the wooded grassland region under studying Our method also has the advantage that it can successfully distinguish "dryness" and "wetness. " This distinguishing can not be achieved by the linear two-source model, which is much more complex. This type information can be used for further applications in hydrology or drought management.