Our findings indicate that when modeling species diversity of birds and plants in agricultural environments, predictors derived from continuous information of crop productivity (NDVI) were consistently ranked higher than predictors derived from information based on a discrete classification of Landsat imagery. Furthermore, local measures of spatial autocorrelation, specifically the local Moran's I, are useful indicators of spectral heterogeneity, at least on par with existing measures such as simple image-based texture (CV). From a practical standpoint, the use of continuous information is preferable, as discrete land cover classifications involve an inherent level of error and generalization, and can be costly to produce and validate. While the overall amounts of variability explained by our taxon-specific models were low, they were generally commensurate with similar studies that relied on Landsat imagery.