How Much Climate Information Does a (Temperature Index) Snow Model Need?

Changes in snowpack climatology are taking place because of changes in climate. Information is needed on how future changes in climate may affect snowpack regimes.

This information is usually generated by running time stepping models for which time series of forcing data must be supplied. In this study we explore whether it is possible to make reliable estimates of snowpack regime without explicit knowledge of the temporal sequence of forcing data. This could support development of a hydrological theory of seasonal snowpacks. We will test whether it could be enough to know some statistics of the forcing data, rather than the complete time series. In this presentation, we begin by trying to identify where it is necessary to maintain the correlation between temperature and precipitation amount in a temperature index model. Our interest is in correlations at the timescale of a precipitation event; seasonal-scale correlations will be captured separately.

We investigate these questions at 4736 locations in the northern hemisphere, using the NH-SWE database combined with precipitation (P) and temperature (T) data from GHCN. We run the temperature index model once with the original forcing data, and then again with the temperature data displaced by a few days in time from the precipitation data, to reduce their cross-correlation. We calculate statistics of the modelled snowpack for the two model runs (for each station and each year: the start date, peak date and end date for the snowpack, and the peak SWE – snow water equivalent). If the cross-correlation is not important, then the statistics of modelled snowpack should not change much between the two model runs. Since our interest is in snow accumulation and melt, we expect that the most important P-T correlations are at times of year when both rainfall and snowfall are likely to occur.

Initial results show that for the 60% of sites with a positive correlation between P and T-anomaly, neglecting the correlation generally leads to an overestimation of peak SWE (by an average 12%). The overestimation presumably occurs because when the correlation is removed, days below freezing are more likely to be paired with the higher precipitation amounts which tend to occur on days above freezing, and thus the amount of snowfall is increased by neglecting the correlation.   For the remaining sites with a negative correlation between P and T-anomaly, neglecting the negative correlation generally leads to a slight underestimation of peak SWE (by an average -4%).

We will also carry out several other similar model experiments with degraded forcing to identify key features of the climate data. The intended endpoint of the work is an improved theory of snowpack hydrology, i.e., a stochastic version of the deterministic theory in Woods 2009 (https://doi.org/10.1016/j.advwatres.2009.06.011)