NBM Forecast Accuracy Analysis

Some Generalities:  In the panels below I compare forecasts predicted by the National Blend of Models, taken for a 10-day forecast period. First of all, the daily 'amounts' should only be believed to about 1 significant figure. My measurements at Sweetwater are not rigorously constrained to be exactly 24-hour periods and cannot account for any snow settling or snow that is affected by the wind. I believe that the models have roughly the same accuracy although I haven't been able to verify this assumption.

Each panel below presents two graphical representations: one is the predicted snowfall versus date for 24-hour periods, taken directly from the NBM website graphs. That is shown on the left in each panel. The other is the accumulated snowfall during the forecast period taken from the NBM graph for 6-hour snowfall predictions. In both cases, the models are shown in orange and the observations in blue. For the 24-hour snowfall graph, on the left, the length of the orange bar shows the range of the predicted snowfall. The blue squares are the observed amount. The error bars on the blue points are estimated by the standard deviation between the observed amount and the mean of the predicted range. For the total accumulated snowfall, the comparison is shown as smoothed lines through the data points.

Storm Data Summaries:

I have summarized the three storms illustrated in the panels below plus three others where I had adequate notes in my ski log AND there were archived data on the NBM website. On average, the NBM forecasts predict about 2±2 inches more snow than we actually receive. Part of this is to be expected since my measurements do not correct for any loss of snow due to settling. In general, you would expect that amount to be less than the actual snowfall amount, in accordance with the data in the table below. Further, the onset of the storm is about day later than forecast.

Recall that my method uses the 10-day forecast from the NBM website, starting 1 day before the predicted onset of the storm. I will add to this table as additional data become available.

Storm Dates Predicted
Storm Delay
2-11 Jan 2022 11 10 +1 +1
22-31 Dec 2021 30 30 0 0
10-19 Dec 2021 20 15 +5 +1
8-17 Nov 2021 2 4 -2 0
26 Jan - 3 Feb 2021 22 15 +7 0
11-20 Nov 2020 9 9 0 0
2022 Jan 2-11 Snowfall:  The snowfall distribution agrees with the model but is offset by about 1 day. The accumulated snow from the storm is about 90% of the forecast amount. This is the last detailed comparison I'm going to make. The panel above will summarize the forecast versus the observed amounts.

2021 Dec 22-31 Snowfall:  The model distribution is close to the observed 24-hour snowfall. The accumulated amount predicted by the model agrees almost exactly with the observed amount. It is worth noting that, when I spoke to Michael R on Jan 3, he reported that the snowfall accumulation at his house (about a mile from ours) was 26 inches. So, that's a range of 4 inches. That gives some estimate of the errors involved in trying to measure the depth. Still, the model prediction is remarkably good, probably better than I might have a priori expected.

2021 Dec 10-19 Snowfall:  The model forecast 24-hour snowfall distribution is similar to what was observed. The observed accumulated amount is about 75% of the predicted amount. The onset of observed snowfall was about 1 day later than the forecast.