Abstract
The specification of summer season precipitation in the contiguous United States from summer season fields of 700 mb height, sea level pressure (SLP) and Pacific sea surface temperature (SST) was carried out using stepwise multiple linear regression. The specifier fields were characterized by their first five Empirical Orthogonal Functions (EOF's). The objectives were to assess the overall skill in specifying summer season precipitation, examine the differences among predictands with regard to both spatial averaging and type of statistic, compare the usefulness of the specifier fields, and to look at spatial variations in specification skill.
Overall, the strongest relationships between actual summer season precipitation and the predictors were found for 700 mb heights (R2 ∼ 0.24) followed by Pacific SST’s (R2 ∼ 0.21) and SLP (R2 &sim 0.12). The use of large area averages (∼ 105 km2) for the predictand produced slightly greater R2 values than for individual climatic division averages (∼ 1O4 km2).
The use of transformed summer season precipitation statistics to account for precipitation skewness, did not improve upon the use of actual summer season precipitation as the predictand. However, frequency of precipitation greater than 0.1 inch resulted in an almost doubling of explained variances over actual precipitation (0.47 versus 0.24) when 700 mb heights were used as the specifier field.
The areas of weakest relationship (west of the Rockies and southern states) between predictor and summer precipitation statistic generally had R2 values less than 0.3, even for the best models. Elsewhere, the R2 values generally ranged from 0.5 to 0.7 for the best model (700 mb heights and precipitation frequency). After accounting for artificial predictability which results from imperfect estimates of the statistics, skill values (explained variances) cast of the Rockies ranged from 0.01 to 0.44.
