Time series analysis of riparian vegetation impacted by irrigation-influenced return flow in a semi-arid climate

Time series analysis of riparian vegetation in a heavily irrigated river valley can describe trends in water use and how it varies with changing connectivity to return flow. Temporal trend analysis is commonly used to examine natural-ecosystem vegetation, but little is known about how trends in riparian vegetation greenness can vary depending on the temporal discretization of the remotely-sensed data (i.e., monthly or annual data). This work investigates the spatio-temporal trends in annual-peak biomass using the Normalized Difference Vegetation Index (NDVI). The study area is a heterogeneous riparian corridor that buffers the Arkansas River and stretches over 160-kilometers from Pueblo, Colorado to the Colorado-Kansas border. The annual-peak biomass is calculated using Landsat 5, 7 and 8 surface-reflectance imagery from 1984-2019. Each annual series of NDVI is analyzed for trend using autoregressive-integrated-moving-average (ARIMA) models within a Bayesian statistical framework. Each model is fit to the series and a goodness-of-fit value is estimated for the model. The models are evaluated using a posterior predictive loss score, similar to a deviance information criterion. To increase the predictive ability of the best model, descriptive covariates are added as regressors to the ARIMA model. Initial results for an autoregressive model of order one indicate neutral trends in peak annual growth over 36 years. This suggests that the vegetation in our study area has a mean-reverting nature and the forces that control the behavior are near-constant over the study period. We expect to find that an ARIMA model with regressors will show which hydrologic and climatologic variables are most predictive of a phenological time series. Autoregressive and moving-average models can identify which covariates, and temporal lags of covariates, best describe the variability in riparian greenness which can be used to forecast riparian growth or to model natural-ecosystem water use in groundwater dependent regions.