aren 
form 
Nave 
one- 
ique 
]ress 
  
ESTIMATING SNOW DEPTH IN THE UK WITH SSM/I IMAGERY 
Richard E.J. Kelly 
Peter M. Atkinson 
University of Southampton 
Dept. of Geography 
Highfield 
Southampton, S09 5NH 
United Kingdom 
ISPRS Commission VII / Working Group 8 
ABSTRACT 
Daily information on the spatial distribution of snow cover properties in the UK is routinely required by the 
National Rivers Authority and the regional water authorities for general monitoring and flood prediction 
purposes. However, such information is presently limited to point measurements of snow depth made at 
sparsely located meteorological and volunteer stations. The Defense Meteorological Satellite Program 
(DMSP) of the USA now provides synoptic information related to snow cover properties from its Special 
Sensor Microwave Imagery (SSM/I). The SSM/I imagery is provide twice per day with a pixel size of 
approximately 8,5 km x 8,5 km. Snow cover properties for the whole of the UK may be estimated by 
applying algorithms to the SSM/I imagery. However, at least initially, such estimates must be calibrated with 
available point measurements of snow properties made on the ground. A problem is that snow properties 
defined on a point support are likely to bear little relation to the true values of snow depth defined on a 
support of 8,5 km x 8,5 km that the SSM/I pixels estimate. Therefore, given point measurements of snow 
depth for several dates, it was necessary to produce maps of snow depth for each date, co-registered with the 
SSM/I imagery and define on the same support. Since collocated digital elevation data for the entire UK 
were available, it was possible to map snow depth from a combination of point snow depth data and 
elevation data. We used kriging to interpolate snow depth measurements to the same set of locations as the 
imagery, and regression and cokriging to estimate snow depth while assimilating information on elevation. 
The relative precisions of the techniques for estimation were evaluated by cross validation. Generally, the 
most precise technique was kriging, but for areas of high elevation, regression increased precision over that 
attainable by kriging. This combination of kriging in lowland areas and regression in upland areas was used 
to map snow depth on the same support as the SSM/I imagery, and these maps were used to calibrate 
estimates of snow depth made from coincident SSM/I image data. 
115