Mapping Precipitation in Switzerland with Ordinary and Indicator Kriging
1 May 1998
Peter M. Atkinson , Chris D. Lloyd
This paper uses the geostatistical methods of ordinary kriging (OK) and indicator kriging (IK) to address the problem of estimating values of precipitation at locations from which measurements have not been taken. Several problems or issues were raised including: (i) lognormality of the data, (ii) non-stationarity of the data and (iii) anisotropy of the spatial continuity. Given that the aim of SIC'97 was to compare a variety of different approaches to estimation. IK (informed using directional indicator variogram models) was selected because it is a means to account for lognormality and it was a method that was unlikely to be used widely within the competition. Accuracy of estimates made using IK were compared with OK estimates. It was observed that the OK algorithm, as implemented here, provided more accurate estimates than IK. This was considered to be due, at least in part, to the method used for tail extrapolation and also the small number of data used in estimation (100 data locations). OK was recommended over IK in this instance as OK provided more accurate estimates and was also more easy to implement.
Journal of Geographic Information and Decision Analysis, Vol. 2., No. 2, pp. 65-76, 1998.
KEYWORDS: indicator kriging, ordinary kriging, precipitation.