Elevation Effect
The simple kriging model, so far, is pure interpolation.
The effect of differences in the elevation of the predicted point and the relevant stations
can be estimated with a jackknife procedure.
The figure below shows the jackknife residuals as a function of the difference in elevation
between the station being jackknifed and the effective elevation of the stations used in the prediction.
The single elevation for the relevant stations is obtained from the station elevations and the
same kriging weights used for the temperature prediction.
This is also done on a daily bases.
A linear regression (green line) has a negative slope (-0.01031)
and a near-zero intercept (0.00797) as expected.
This regression captures 10.1% of the remaining variance and helps correct for elevation differences
between the predicted point and the available air temperature stations.
In the graph below, the gray area is the minimal area to contain 80% of the points and the
black area is the minimal area to contain 50% of the points.
Seasonal Elevation Effect
The above analysis was repeated separately for the minimum temperature and for the maximum temperature
during each of the 3 seasons as before.
The table below gives the regression results where "%" gives the percent of the variation explained.
Sample sizes range from 184,263 to 308,042.
The results are all similar, but the elevation correction is more important for the maximum than
for the minimum and more important in the summer than in the other seasons.
| | Minimum | Maximum |
|---|
| | Slope | Intercept | % | Slope | Intercept | % |
|---|
| Spring | -0.00867 | 0.03933 | 5.6 |
-0.01084 | -0.05208 | 9.5 |
| Summer | -0.01000 | 0.08296 | 14.5 |
-0.01256 | -0.02243 | 17.8 |
| Autumn | -0.00874 | 0.03223 | 5.3 |
-0.00977 | -0.05865 | 8.2 |
27 AUG 2001, updated on 30 OCT 2001
dlg@rapid.msu.montana.edu