Figure 1- Historical wet-day fractions from 4 statistical downscaling tecniques (plots a-d), all of which have different fractions compared to observations (plot e). The low resolution data they are downscaled from is shown for reference (plot f). Figure from Gutmann et al. (2014)
Figure 2 - Impacts of 4 statistical downscaling techniques on hydrologic simulations
In general, statistical downscaling methods respond to the reality that GCMs simulate some features of weather and climate well, and others poorly, or not at all given their scales. Hydrological applications require their weather and climate inputs to exhibit certain characteristics that lie across the spectrum of these GCM strengths and weaknesses (for instance, wet day frequency affects the rainfall-runoff relationship, but is not well simulated by GCMs). As our findings indicate, existing downscaling methods have mixed success in leveraging GCM strengths and avoiding their weaknesses so as to providing the required hydrologic inputs. This prompts us to seek improvement of statistical downscaling through a more explicit recognition of what information is needed for hydrology, and what can be reliably extracted from GCMs versus what must be parameterized or generated through analysis of local watershed hydroclimatology.
Figure 3 - Impacts of spatial resolutions on WRF simulations of winter total preciptation (Oct-May)
Key findings found from evaluations of impacts of downscaling technique on hydrologic assessments include:
- Many statistical downscaling methods that are popular in the water management community produce hydroclimate representations with too much drizzle (Figure 1), too small extreme events, and improper representation of spatial scaling characteristics that are relevant to hydrology. These deficiencies vary by method, significantly impacting results of hydrologic simulation (Figure 2).
- The resolution used in dynamical downscaling matters. The 12-km and 36-km WRF simulations have poor correspondence to observations (Figure 3), and also very different change signals compared to the 4-km WRF simulations. The impact of WRF resolution on hydrology is primarily due to differences in precipitation among WRF simulations, although differences in the spatial resolution of the hydrology model are still important.
Gutmann et al. (2014) An Intercomparison of Statistical Downscaling Methods Used for Water Resource Assessments in the USA. Water Resources Research. In Revision