A statistical approach for predicting grassland degradation in disturbance-driven landscapes

TitleA statistical approach for predicting grassland degradation in disturbance-driven landscapes
Publication TypeConference Paper
Year of Publication2015
AuthorsJacquin, A, Goulard, M, Hutchinson, JMS, Hutchinson, SL
Conference Name2015 8th International Workshop on the Analysis of Multitemporal Remote Sensing Images (Multi-Temp)
Conference LocationAnnecy, France
Accession NumberKNZ001715

The relationship between fire and long-term trends in tallgrass prairie vegetation was assessed at Fort Riley and Konza Prairie Biological Station (KPBS) in Kansas. Linear trends of surface greenness were previously estimated using BFAST and MODIS MOD13Q1 NDVI composite images from 2001 to 2010. To explain trends, fire frequency and seasonality (fire regime) was determined and each site was divided into spatial strata using administrative or management units. Generalized linear models (GLM) were used to explain trends by fire regime and/or stratification. Spatialized versions of GLMs were also computed address unexplained spatial components. Non-spatial models for FRK showed fire regime explained only 4% of trends compared to strata (7-26%). At KPBS, fire regime and spatial stratification explained 14% and 39%, respectively. At both sites, improvements in performance were minimal using both fire and strata as explanatory variables. Model spatialization resulted in a 5% improvement at FRK, but with weak spatial structure in the residuals, and was not necessary at KPBS as the existing stratification most of the spatial structure in model residuals. All models at KPBS performed better for each explanatory variable and combination tested. Fire has only a marginal effect on vegetation trends at FRK despite its widespread use as a grassland management tool to improve vegetation health, and explains much more of the trends at KPBS. Analysis of predictors from spatial models with existing stratification yielded an approach with fewer strata but similar performance and may provide insight about additional explanatory variables omitted from this analysis.