'As simple as possible but not simpler': What is useful in a temperature-based snow-accounting routine? Part 2 - Sensitivity analysis of the Cemaneige snow accounting routine on 380 catchments

This paper investigates the degree of complexity required in a snow accounting routine to ultimately simulate flows at the catchment outlet. We present a simple, parsimonious and general snow accounting routine (SAR), called Cemaneige, that can be associated with any precipitation-runoff model to simulate discharge at the catchment scale. To get results of general applicability, this SAR was tested on a large set of 380 catchments from four countries (France, Switzerland, Sweden and Canada) and combined with four different hydrological models. Our results show that five basic features provide a good reliability and robustness to the SAR, namely considering: (I) a transition range of temperature for the determination of the solid fraction of precipitation; (2) five altitudinal bands of equal area for snow accumulation; (3) the cold-content of the snowpack (with a parameter controlling snowpack inertia); (4) a degree-day factor controlling snowmelt; (5) uneven snow distribution in each band. This general SAR includes two internal states (the snowpack and its cold-content). Results also indicate that only two free parameters (snowmelt factor and cold-content factor) are warranted in a SAR at the daily time step and that further complexity is not supported by improvements in flow simulation efficiency. To justify the reasons for considering the five features above, a sensitivity analysis comparing Cemaneige with other SAR versions is performed. It analyses the snow processes which should be selected or not to bring significant improvement in model performances. Compared with the six existing SARs presented in the companion article (Valery et al., 2014) on the 380 catchments set, Cemaneige shows better performance on average than five of these six SARs. It provides performance similar to the sixth SAR (MORD4) but with only half its number of free parameters. However, CemaNeige still appears perfectible on mountainous catchments (France and Switzerland) where the lumped SAR, MORD4, outperforms Cemaneige. Cemaneige can easily be adapted for simulation on ungauged catchments: fixing its two parameters to default values much less degrades performances than the other best performing SAR. This may partly due to the Cemaneige parsimony. (C) 2014 Elsevier B.V. All rights reserved.