Geostatistical spatio-temporal modeling of landscape spatial heterogeneity

This study provides a new approach to characterize the spatial structures within high spatial resolution (similar to 20m) remote sensing imagery as the weighted linear combination of two stochastic models: a Poisson line mosaic model and a multi-Gaussian model. We first apply this approach to describe the nature of the processes structuring distinct types of landscape. We show that the mosaic model is an indicator of strong NDVI discontinuities within the image, mainly generated by anthropogenic processes such as the mosaic pattern of agricultural site. The multi-Gaussian model shows evidence of diffuse and continuous variation of NDVI over natural vegetation and forest sites, generally engendered by ecological and environmental processes. We implement a method based on the simultaneous use of the first- and second-order variograms to distinguish between the multi-Gaussian and the mosaic model, and to retrieve the fraction of the image variance explained by each model. The second part of this paper consists in applying the previous stochastic models to a series of remote sensing images taken at a single site for modeling the temporal variations in surface spatial heterogeneity observed over an agricultural site. We build a model describing the temporal course of the image second-order variogram as a function of crop seasonality. Once calibrated from a temporal sampling of few high spatial resolution scenes, this model proves to be powerful to predict the second-order variogram at a date at which the high spatial resolution scene is not available, and thus to retrieve the spatial heterogeneity within an area of similar to 1km through the seasonal cycle with a mean relative accuracy of 20%.