Fig. 1.

Frequency distributions showing biomass yields reported for (a) both upland and lowland ecotypes, (b) upland ecotypes, and (c) lowland ecotypes. The number of observations analyzed for each category is shown in parentheses.


Fig. 2.

Box plot of biomass yields for upland and lowland cultivars of switchgrass. Bold horizontal lines indicate the median for each cultivar. Boxes represent the interquartile range (IQR, or middle 50%) of the yield values for each cultivar. Vertical whiskers extend to the furthest values up to 1.5 IQR. Additional points are shown as outliers. Box widths are proportional to the number of observations for each cultivar. Cultivars are grouped by ecotype, with gray horizontal lines indicating the median yield across cultivars within each ecotype.


Fig. 3.

Biomass yield in (a) lowland and (b) upland ecotypes as a function of total N applied during the growing season.


Fig. 4.

Relationship between biomass yield and plot size for lowland and upland ecotypes of switchgrass. Plot sizes were binned according to a given range and associated with biomass yield estimates contained within the database. Standard deviations for both plot size and biomass yield are shown. Plot size is graphed on a log scale.


Fig. 5.

Biomass yield in (a) lowland and (b) upland ecotypes plotted as a function of annual average air temperature. Latitude and longitude for each site were used to extract temperature records in PRISM.


Fig. 6.

Biomass yield in (a) lowland and (b) upland ecotypes plotted against growing season (April–September) precipitation. Latitude and longitude for each site were used to extract temperature records in PRISM if not provided by the authors.


Fig. 7.

GAM smoothing functions (solid curves) and partial residuals (points and box plots) for biomass yields. Dashed curves and box plot notches indicate standard errors for each GAM function. A logarithmic transformation was applied to yield to avoid negative predicted values and to better comply with the assumption of normally distributed errors.


Fig. 8.

Measured vs. predicted yield from the parametric model. Gray points represent upland cultivars; black points represent lowland cultivars. Size of points is proportional to experimental plot size. Solid line represents a 1:1 fit, and dashed lines represent the limits of the 80% predictive interval. Five plots had measured yields greater than 30 Mg ha−1 and therefore fall outside the range of the plot.


Fig. 9.

Map of switchgrass biomass yields across the United States predicted by the empirical model. Simulations assume annual application of 100 kg N ha−1 Thin circles represent observed yield as reported in the literature, black for upland and red for lowland ecotypes. Diameter of circle is proportional to yield, from 1 to 39 Mg ha−1 Apparent thick circles are the result of multiple yield observations too similar to be resolved at this scale. Regions of the United States where extrapolations of switchgrass production fall outside the precipitation and temperature conditions used to parameterize our model are outlined in black. Small enclosed areas may appear as solid colors.