Determination of the best geo-statistical model for field data interpolation using Remote Sensing

Remotely sensed data are able to monitor some characteristics of the environment and also their spatial structure. The latter one is the first and main step in field data interpolations. Therefore, spatial analysis of some field data is possible by employing of related satellite data bands. In this study, as an example, Landsat ETM+ thermal band (band 6) was acquired to determine the spatial structure of surface temperature distribution. For the purpose of evaluation and selection of the best interpolation model, the band 6 data was crossed with a regular sampling grid and therefore, a dataset was constructed. Using geo-statistical analysis, empirical semi-variogram was calculated and various mathematical models were fitted to its points (e.g. gaussian, exponential, circular and spherical). Afterwards, the fitted models were applied to generate different temperature distribution maps using kriging interpolation approach. Finally, the optimum model which could better predict temperature changes and distribution was recognized. Result of the study shows that the exponential model would be better to predict and estimate surface temperature in un-sampled locations in the area of studied. So, the model can be used for interpolation of field temperature data with a high confidence level. The represented method can be developed for all the other environmental factors which could better characterized by remotely sensed data, like minimum and maximum temperature, evapotranspiration and so on.