We use soil samples data for interpolating soil fertility parameters using ArcGIS geostatistical Kriging. While doing kriging, the resultant map values ranges are not matching with our input samples range. Suppose available phosphorus of analysed soil samples data vary from 30 to 75 ppm (for 500 samples). After prediction, map shows range between 35 to 60ppm. Therefore we can't classify upper concentration, above 60 - 75 ppm.
Is it because of prediction modelling by calculating statistical relationship of input points?
What is the actual principle behind this?