I've recently finished a basic comparison of the air temperature values produced by the Fetch Climate service with a values provided by the PRISM climate group. PRISM climate group provides the most accurate climate data grids for USA at this moment. Thus a comparison with PRISM is a crucial step for the sanity check of the result produced by the Fetch Climate.
The comparisons were made for the values produced by FetchCimate with GHCN datasource. As it involves more complicated computations than any other datasources supported by the Fetch Climate now. And therefore this should be checked first of all.
A set of points was formed for the test. The points were randomly placed across Wisconsin and Michigan area. For the each of the point a number of timeseries for 2010 year was calculated. Timeseries included maximum, minimum and calculated mean temperature of PRISM dataset. These timeseries were compared with mean temperature timeseries produced by Fetch Climate (red curve on video).
The video below enumerates timeseries for different points. It shows that the values of the Fetch Climate don't exceed the [prism_min;prism_max] interval almost everywhere (except for November temperatures in some of the points), although the prism_mean timeseries looks more smooth.
The numeric results will be published later. But these visual comparisons give a basic confidence for the correctness of the implemented algorithms.
 |
| click on the image to view video |
 |
| FetchClimate values for June 2010 (GHCNv2 datasource) |
 |
| (PRISM_MAX+PRSIM_MIN)/2 for June 2010 |
