Can dynamic vegetation information improve the accuracy of Budyko's hydrological model?

This study examines whether the incorporation of dynamic, remotely sensed vegetation information into Budyko's hydrological model can improve that model's accuracy, particularly at finer spatial and temporal scales. Using numerous variables derived from three primary time-series variables (precipitation, potential evaporation and remotely sensed estimates of vegetation cover), linear one-variable models are developed to explain the variability in stream flow predictions not already captured by the Budyko model (the 'scatter'). Analyses are applied to 221 catchments across Australia and cover the period 1981-2006. At the annual average temporal scale, results show that vegetation-related variables are not the primary determinants of scatter at coarse spatial scales (i.e., continental), and that little improvement could be made on the original Budyko model. However, at medium spatial scales vegetation-related variables dominate models - most notably variables relating to intra-annual dynamics in vegetation cover - reducing prediction error to 30% of average annual stream flow from the 42% error achieved using the original Budyko model. By contrast, at the annual time-scale the most important variables for explaining scatter relate to precipitation dynamics regardless of spatial scale. This is interpreted as the influence of non-steady-state conditions at this finer time-scale. Overall, vegetation information is shown to improve the accuracy of long-term annual average stream flow predictions as spatial scale decreases and that, at annual scales, the presence of non-steady-state conditions prohibits the exploration of the hydrological role of vegetation dynamics regardless of spatial scales of analysis. Crown Copyright (C) 2010 Published by Elsevier B.V. All rights reserved.