A Simple Method for Predicting Snowpack Water Equivalent in the Northeastern United States

Daniel Samelson

NRCC Research Publication RR 92-1

Increasing the spatial density of Northeastern United States locations for which daily snowpack water equivalent (SWE) information is available would be useful for flood forecasts and agricultural, urban, and recreational water supply prediction. To that end statistical regression models were developed using as predictors meteorological variables available at National Co-operative Observer Program (Co-Op) sites. The 15 National Weather Service Offices (NWSOs) in New York and New England were the source of 28- to 35-year daily summary data sets which included SWE. These sets were used to develop the models. Monthly NWSO models, all predicting the square root of SWE, were created for the climatological winter months of December, January, and February, as well as for November, March, and April. Predictors included the square root of snow depth, the number of consecutive days with the maximum temperature below freezing, the snowfall during the 24 hours preceding the day of prediction, and the water equivalent of precipitation which fell in that period. Values of R2 ranged from 43.3% to 87.5% for the winter months, with 67% of all predictions still falling within +-15% of observed when calculated for a 10 inch (25 cm) SWE.

The stations were then grouped, first monthly by geographic regions, and then into one "Total" group encompassing all 15 NWSOs for each month. Winter groups combining the data for the three winter months, and also a "Total" winter group, combining all 15 NWSOs and 3 months, were created. Models developed for these groups provided a separate intercept term for each station (and month in the case of the "Total" winter group), but included one parameter estimate per predictor for all sites. All of these models also included a correction factor for rain on snow events. In the case of the winter "Total" model, 45 monthly station models were reduced to a single equation. This model still exhibited an R2 of 72.0%.

Independent verification studies were performed to determine the degree to which the NWSO-derived models could predict SWE for Co-Op stations. Periodic Co-Op snow survey data was used to determine both the efficacy of the models as well as the NWSO model best suited for use at a Co-Op station. Results showed that SWE at each Co-Op station was well predicted by the models from at least one NWSO, with results at least as good as those from developmental sites. Model-described variation ranged from 60.0% to 81.0% for most January data, and from 54.0% to 90.0% for most February data.

A method was developed to assist potential users in determining the appropriate model for a given location. Predictions for stations in western New York which receive Lake Effect snow may be made using Buffalo or Rochester models which receive Lake Effect snow, while Binghamton models should be used at those sites not affected by these snows. In the mountains and valleys of the Adirondacks and Catskills, two criteria may be used. Caribou, ME, models should be used for prediction if a station has single digit (°F) long-term monthly average minimum temperatures, or, if these temperatures are double digit, the site is above 1200 feet in elevation. If these criteria are not met, Albany models yield best results.

Further calibration studies are needed to determine if these models are equally useful in the remainder of the Northeastern United States and whether they have applications in adjacent Canadian provinces.

151 pp.

Click here to download/view this publication: Click to download/view pdf