Systematically Optimized Degree-Index Models for Data-Scarce Regions: A Comparison of Structures Using Measured Versus Optimized Parameters

The reality of data-scarcity in many regions makes sophisticated, energy-mass balance hydrological models difficult to implement and calibrate. An alternative form of distributed model for determining runoff is the degree-index method, which tends to only require precipitation and mean temperature as meteorological inputs. Many groups have produced global datasets for these meteorological variables, yet especially for data-sparse regions, inter-group comparison of a single variable tends to reveal significant discrepancies. Many degree-day studies seemingly ignore these data peculiarities and select the majority of their model parameters (e.g. degree-day and melt factors) through physical experiments in their region of interest. In contrast, the current study advocates performing multi-variate optimization of modelled runoff on the majority of the model parameters. The current work applies a genetic algorithm optimization function, written by the authors, to a degree-index model with terms that are heterogeneous in space and time but parameterized through attributes such as latitude and day of the year. While this model formulation works with any precipitation and mean temperature time-series grids, the authors use a downscaled combination of WorldClim climatologies and Climate Research Unit monthly time-series, which are both freely available for the entire world. This allows monthly streamflow to be modelled for any region of the world using a single modelling package (from meteorological data downscaling through production of monthly streamflow grids), provided that stream gage data is entered at the optimization step. As a case study, the model is applied to the Gulf of Alaska and compared to a published degree-index model developed for that region.