Testing climate WNA modeled climate data against independent weather station records

Background/Question/Methods

Weather data is necessary for answering many ecological questions, but is often unavailable for locations of interest. Modeled climate data is a potential solution to this problem. Parameter-elevation Regressions on Independent Slopes Model (PRISM) and ClimateWNA are two popular sources for modeled climate data. PRISM produces ~4km² and 800m² grids of temperature and precipitation variables based on National Weather Service Cooperative Observer (COOP) daily observations from the National Climatic Data Center, Digital Elevation Maps, and a general elevation regression function. ClimateWNA is a program that translates ~4km² PRISM data into a continuous surface, then adjusts for the precise elevation at each point, creating essentially scale-free climate data. We tested ClimateWNA predictions in semi-arid and topographically variable northern Arizona using five weather stations along a ~45km long 1087m elevation gradient in Northern Arizona’s San Francisco Peaks. These stations are not included in the COOP network that informs PRISM (and thus ClimateWNA), and so are a good test of ClimateWNA’s predictions. We compared data from these stations [Great Basin Desert (1556m), Grassland (1760m), Pinyon-Juniper (2020m), Ponderosa (2344m), and Mixed Conifer (2620m)] to the ClimateWNA predictions for yearly precipitation, summer precipitation, winter precipitation, average monthly temperatures, average minimum monthly temperatures, and average maximum monthly temperatures between 2002 and 2006 at each site.

Results/Conclusions

A perfect match between ClimateWNA modeled data and measured weather station data would result in a linear relationship with a slope and an R² of 1. ClimateWNA predicted temperature very well: temperature variable slopes were between 0.70 and 1.05 and R² values were all 0.85 or higher (except minimum temperatures at the Ponderosa site, which only had an R² of .52). Average temperatures were underestimated at high elevations and overestimated at low elevations, likely because minimum temperatures were usually overestimated and overestimation increased as elevation decreased. Maximum temperatures were almost always underestimated at all sites. Precipitation was not predicted very well. While ClimateWNA’s authors acknowledge that their method improves temperature predictions more than precipitation predictions, they obtained R² values of ~0.65 with independent precipitation data. Our summer precipitation R² values ranged from 0.04 to 0.86 and decreased with increasing elevation. Winter precipitation predictions were also better at lower elevations (R² values from 0.08 to 0.40). Precipitation was almost always overestimated. So, ClimateWNA temperature predictions are reasonably good and can be used in the semi-arid southwest. Unfortunately, the correlation between measurements and precipitation predictions is highly variable and usually bad.