Short-Term Phenological Predictions of Vegetation Abundance Using Multivariate Adaptive Regression Splines in the Upper Colorado River Basin

Abstract

Abstract The accurate prediction of plant phenology is of significant importance for more sustainable and effective land management. This research develops a framework of phenological modeling to estimate vegetation abundance [indicated by the normalized difference vegetation index (NDVI)] 7 days into the future in the geographically diverse Upper Colorado River basin (UCRB). This framework uses phenological regions (phenoregions) as the basic units of modeling to account for the spatially variant environment–vegetation relationships. The temporal variation of the relationships is accounted for via the identification of phenological phases. The modeling technique of Multivariate Adaptive Regression Splines (MARS) is employed and tested as an approach to construct enhanced predictive phenological models in each phenoregion using a comprehensive set of environmental drivers and factors. MARS has the ability to deal with a large number of independent variables and to approximate complex relationships. The R2 values of the models range from 91.62% to 97.22%. The root-mean-square error values of all models are close to their respective standard errors ranging from 0.016 to 0.035, as indicated by the results of cross and field validations. These demonstrate that the modeling framework ensures the accurate prediction of short-term vegetation abundance in regions with various environmental conditions.