ANALYSIS

Investigating Vegetation in the Phoenix Basin

The rather large figure below illustrates the outcome of a linear regression that I ran on the above data layers, with average biomass as the dependent variable. The multiple R-squared value for the equation is a meagre 10%. The primary reason is that the data layers do not account for the extremely high values of biomass in the riparian vegetation of the Gila and Salt River beds.

When I ran a regression that included a layer based on distance from the river bed, the R-squared value increased to 30%, but the resulting predicted biomass map looked nothing like the actual biomass, except for in the river beds. The high biomass in riparian areas swamps the lower values in other parts of the desert. I decided to use the original equation in the figure above, with the lower R-squared value for several reasons. First, biomass in the riparian area showed virtually no change in the three years examined, with the exception of some vegetation scouring in the river channels in 1993 as a result of the floods that year. Second, I am interested in the effects of local precipitation on biomass at this stage, and riparian vegetation is more closely related to rainfall and snowpack in the mounains of Arizona and New Mexico. Finally, riparian biomass can be modelled separately using a streamflow model, most likely much more simply than the model here.

The equation shown in the figure works fairly well for predicting biomass across much of the landscape, but not for the riparian areas. A two- step modeling process will be used to deal with this problem. A majority of the 40,000 cells used in the regression model are predicted within 5% of the actual biomass value.