Northern Prairie Wildlife Research Center
Figure 1. Study area used in model construction and validation. Shaded areas are parkland; unshaded areas are grassland. Horizontal lines are aerial transect locations. Numbers are stratum designations.
Table 1. Geographic areas within the Prairie Pothole Region used in model development
|Model Region|| n|
|Maximum Basins1||Area (km2)||Basin Density2|
|United States Grassland||20||8,019||169,515||2319||3.46|
1 Maximum number of wet basins counted during aerial transect surveys for the time period covered by the model: 1968-90 for parkland and Canadian grassland, 1973-87 for United States grassland.
2 Basins/km2 within sampled area.
Figure 2. Average annual maximum temperature and total annual precipitation for study areas in Canada (red line) and the United States (blue line) during the years used in model development (1973-1987 for United States; 1968-1990 for Canada).
I used multiple linear regression to examine the relationship between climate variables and percentage of wet basins for the three geographic regions: parkland, Canadian grassland, and United States grassland. Explanatory variables included minimum, maximum, and mean monthly, seasonal, and annual temperature; total monthly, seasonal, and annual precipitation; monthly snowfall (in Canada); and the number of wet basins counted the previous May. The effects of one- and two-year lags were examined for the seasonal and annual data. Two moisture indices were also examined, the Thornthwaite moisture index (McCabe and Wolock, 1991) and a conserved soil moisture index (Boyd, 1981). These indices are essentially linear combinations of seasonally weighted temperature and precipitation values. In addition, I examined effects of monthly temperature ranges (mean maximum - mean minimum) for spring and fall. All calculations were based on a year defined to be May-to-April. Because the same transects were sampled each year, I treated transects as blocks. The MAXR selection method in SAS PROC REG (SAS Institute, Inc., 1988) was used to determine which climate variables best explained the variation in the wet basin data. Only variables resulting in a substantial increase in R2 were included in the final models. When two or more variables produced similar increases in R2, variables not resulting in multicollinearity were favored. Finally, models were inspected for presence of interactions and quadratic terms.
I tested the models in three ways. First, I randomly withheld 10% of the data, re-ran the models on the restricted data set, and compared observed and predicted values for percentage of basins holding water (hereafter, percent wet basins). Next, I withheld two randomly selected transects from each stratum, one transect at a time, re-ran the models on the restricted data sets, and again compared observed and predicted values for percent wet basins. Finally, I used the parkland model to predict percent wet basins on transects in Strata 34 and 35, which were not used in model development, and compared observed and predicted values.
To explore the potential effects of increasing temperatures on number of wet basins in the three regions, I increased temperature variables in the original climate data set by 3 °C and 6 °C but held precipitation at observed levels. Likewise, I increased and decreased precipitation by 10% but held temperatures at observed levels. Finally, I combined the 3 °C temperature increase and 10% precipitation increase and decrease to examine the potential for interactive effects in the models.