|
Key words: Coregionalisation, Soil moisture content, SAR images, Factorial kriging analysis, Multitemporal analysis
Soil moisture content is an important and highly variable component of the hydrologic cycle. At basin and regional scale, however, the assessment of the soil moisture content remains a time- and labour-consuming practice: The more accurate the spatial interpolation should be, the more (spatially distributed) measurements are needed. One way to overcome this problem is to use a second variable that covaries with the soil moisture content and the value of which is known at more locations. Quantification of this covariation is carried out through coregionalisation and the interpolation is then performed using geostatistical tools. A time series of European Remote Sensing Synthetic Aperture Radar (ERS2 SAR) images was used for that purpose: the backscattered signal, known at every location in the investigated area, was used as the secondary variable. SAR images, however, are strongly contaminated by noise (spatially unstructured information). Therefore, coregionalisation was performed after spatial filtering using Factorial Kriging Analysis. In this way, the (for our purpose) redundant information could be eliminated, resulting in an improved covariation between soil moisture content and the backscatter signal, and thus an improved interpolation. This was done separately for all images. However, to further improve the predictive quality, a multitemporal analysis was carried out on all the filtered SAR images together, i.e., a Principal Component Analysis (PCA) was performed on all images in the time series. This resulted in a concentration of the soil moisture related information in one of the principal components and thus in a further elimination of noise and redundant information. This newly created image was then used for coregionalisation with the soil moisture content at all measurement dates.
The investigation was performed on a study area in Belgium. A winter sequence of SAR images was used to reduce the effect of seasonal change in land use.