Conceptual and computational challenges in continental-domain hydrologic model parameter estimation: VIC applications over the contiguous USA

Abstract:

Estimation of spatially distributed parameters is one of the biggest challenges in hydrologic modeling, especially over a large spatial domain. This problem arises both from methodological challenges such as the transfer of calibrated parameters to ungauged locations as well as computational demand. Consequently, many current large scale hydrologic assessments rely on spatially inconsistent parameter fields, e.g., showing patchwork patterns resulting from individual basin calibration or spatially constant parameters resulting from the adoption of default or a-priori estimates.

In this study we apply the Multi-scale Parameter Regionalization (MPR) framework to generate spatially continuous and optimized parameter fields for the Variable Infiltration Capacity (VIC) model over the contiguous United States (CONUS). The MPR method uses transfer functions that relate geophysical attributes (e.g., soil) to model parameters (e.g., parameters that describe the storage and transmission of water) at the native resolution of the geophysical attribute data and then scale to the spatial resolution of the model simulation with several scaling functions, e.g. arithmetic mean, harmonic mean, and geometric mean. Model parameter adjustments are made by calibrating the parameters of the transfer function rather than the model parameters themselves.

This presentation discusses conceptual and computational challenges in a “model agnostic” continental-domain application of the MPR approach. First, we describe development of new transfer functions for the VIC model, and discuss challenges associated with extending MPR to multiple models. Second, we describe the “computational shortcut” of headwater basin calibration where we estimate the parameters for only 600 headwater basins rather than conducting simulations for every grid box across the entire domain. We calibrated the transfer function parameters using different spatial groups; a single CONUS-wide group and several climatic-based clusters. In addition, we performed individual basin calibration, which provides a benchmark of the maximum achievable performance in each basin. Lastly, we discuss how estimated parameters for each spatial group can be used in ungauged locations to generate spatially consistent parameter fields over the CONUS.