Climate change, hydrology, and ecological models: intercomparison and validation

This dissertation consisted of three related studies. The first study used historical streamflow records from the Hydro-Climatic Data Network (HCDN) representing 13 unimpaired watersheds to validate runoff simulated by six terrestrial ecosystem models (Biome-BGC, Century, GTEC, LPJ, MC1, and TEM) comprising VEMAP. The validation exercise tracked the historical period of VEMAP from 1895-1993. This study demonstrated that the models performed better in wet watersheds, where runoff was generally underestimated, than dry ones, where runoff was overestimated. The static vegetation models outperformed the dynamic vegetation models. TEM, an independently calibrated model, was the best performer. GTEC, a model missing a snow routine and an evaporation function, was the lowest performer. This study suggested that improvements to these types of models may come from more realistic descriptions of hydrologic processes in arid environments. The second study examined water balance parameters of four of these models (Biome-BGC, Century, LPJ, and MC1) from the inception of the VEMAP climatological data base in 1895 through 2100. The 21st century simulations were based on the Hadley Centre and Canadian Centre for Climate Modeling general circulation models (GCM). Runoff and actual evapotranspiration trends were positive in the majority of cases. Percentage changes in runoff exceeded underlying changes in precipitation and this amplification increased over time. Results varied by model and GCM. The third study sought to address an underlying assumption of the validation work. Streamflow gauging stations included in the HCDN have been carefully screened to filter out factors that could confound their use in the assessment of long-term climate signals. One criterion for station inclusion was that land uses in the watershed had not changed over the period of record so as to significantly alter natural streamflow. This assumption was explicitly tested in a watershed selected for the validation study by interpreting remotely-sensed imagery from the North American Landscape Characterization (NALC) project, and land cover was found to be relatively unchanged over the period examined, 1972-1992. Runoff modeling of land-cover changes was accomplished using the Long-Term Hydrologic Impacts Assessment model and runoff changes were less than 5% over the twenty year period.