Assessment of water-balance models for use with tree ring (proxy) records to reconstruct past streamflow in the upper Walker River basin

Reconstruction of past streamflow helps water managers and planners to better understand the past and assess future scenarios because such reconstructions augment limited observed water resources data. Tree ring data are useful for reconstructing past streamflow on seasonal or annual scales as these data are only available in annual or in seasonal scales at most. Such reconstructions can be done by traditional regression approaches or mechanistic modeling. An advantage of a mechanistic modeling approach over an empirical approach is that it can consider factors that alter streamflow independently of climate, such as changes in land use or land cover, as well as examine sources of model uncertainty. In this study seasonal water balance models were developed to examine the possibility of using a mechanistic approach to reconstruct streamflow with tree-ring inputs of precipitation and air temperature. We investigated the Thornthwaite model modified for use with seasonal input data, and modifications of a simple water balance model to reconstruct streamflow for the West Walker River Basin at Coleville, California. The models were developed using PRISM precipitation and temperature data for the West Walker River above Coleville due to the limitations of adequate measured temperature and precipitation data in the watershed and were investigated regarding their applicability for use with tree-ring data. Because we can only use one temperature input per season, use of average seasonal maximum and minimum temperatures, and average seasonal temperatures in each of the models were examined. Each water balance model with all temperature schemes was calibrated with the half of the available instrumental record (i.e. Water Year (WY) 1939-1977) and validated with the rest of the record (WY 1978-2010). Results suggest that all of the models with appropriate temperature schemes were able to simulate seasonal streamflow remarkably well with r-squared values greater than 0.9. The Thornthwaite seasonal model with average minimum seasonal temperature performed better in calibration and validation statistics than the other water balance models studied. Future work involves reconstructing flows for the West Walker River Basin with all of the models using reconstructed precipitation, and applying the approach to other watersheds in the Western U.S. with the best-performing model.