More and more environmental and agricultural data are now acquired with a high precision and temporal frequency. These data are often represented in the form of rasters and are useful for agricultural activities or climate change analyses. In this paper, we propose a new method to process very large raster. We present a new technique to improve the execution time of the selection and calculation of data summaries (e.g., the average temperature for a region) on a temporal sequence of rasters. We illustrate the use of our approach on the case of temperature data, which is important information both for agriculture and for climate change analyses. We have generated several data sets in order to ana-lyze the influence of the different value properties on the process performance. One of our final goals is to provide information about the value conditions in which the proposed processing should be used.