Title:		Pixels, Inputs and Parameters: Assessing Uncertainty in 
			Snow Modeling
Dr. Andrew Slater
		Research Scientist
		CIRES & NSIDC
		University of Colorado, Boulder, CO 
		http://nsidc.org/research/bios/slater.html
		
			Abstract

In the Western US snow represents a major water resource so there is a large desire to improve our 
knowledge of snow  both in retrospect and in forecasting. However, there are many areas of 
uncertainty that must be accounted for in  order to increase our understanding; some of these shall 
be examined in this presentation. When solving the inverse problem of snow water equivalent (SWE) 
reconstruction potential errors can arise from many areas (i) the ability to infer the Final Date of 
the Seasonal Snow (FDSS) cover, particularly from remote sensing; (ii) errors in model forcing data 
(such as air temperature or radiation fluxes); and (iii) weaknesses in the snow model used for the 
reconstruction, associated with both the fidelity of the equations used to simulate snow processes 
(structural uncertainty) and the parameter values selected for use in the model equations. The 
trade-off between estimating the FDSS and model forcing errors is investigated using 10,000 
station-years worth of data from the western US SNOTEL network. Model structural and parameter 
uncertainty are eliminated by using a perfect model scenario i.e. comparing results to modelled 
control runs. We pose questions such as: "What is the equivalent error in model forcing for a 5 day 
error in the FDSS?". Following such an analysis, we ask what are the likely errors in model forcing 
data when undertaking large scale hydrologic studies in the US West - particular interest is placed 
upon solar radiation estimates because such data would be required for using more complex, 
physically based, hydrologic models. While there is a desire to move to such models, conceptual 
models of snowmelt (e.g. the NWS SNOW-17 model) are still used operationally due to questions about 
reliablilty of data needed to drive full energy balance models. One downside of conceptual models is 
the need for calibration as parameters either do not have a physical meaning or are not easily 
measured. With a hope to move to more distributed hydrologic models, the need to estimate parameters 
across a wide variety of climatic conditions increases. In this last piece of work, our aim is to 
produce an a priori parameter set for a given location based on climate information and long term 
satellite imagery. Using over 550 SNOTEL sites in the Western USA with 11 years (1998-2008) of 
continuous data we can demonstrate skill in parameter estimation, but there is room for improvement 
to an optimal solution.