This summer, I have gotten a number of complaints about the poor quality of our local forecasts.
Several of you have suggested that the weather is often much better (generally warmer) than predicted by the National Weather Service and several media outlets.
I have to admit, I had noticed the same thing. My calibration as a long-term forecaster seemed to be off, with temperatures surging far higher than than in the past for the same large scale conditions. In the gym, I mused about this with Nick Bond, State Climatologist. He agreed, and we talked about some reasons this could be true.
But is it?
Have forecasts been unusually bad this summer?
To answer the question, I made use of a wonderful web page created by Luke Madaus, a Ph.D. student in my department. Let re show the the results.
Here are the high (max) temperature errors for the past 200 days for Seattle from the National Weather Service. Forecast minus observed is shown, so negative temperature errors indicate forecasts being too low. From Feb to May they were up and down, but during June trough early August the Weather Service was too low almost every day (by roughly 3-4F).
Now the central tools of National Weather Service forecasters (and others) is something called MOS or Model Output Statistics. MOS takes numerical model output and corrects it based on past statistics, generally using at least 2 years of history. This is called statistical postprocessing. MOS generally improves on the numerical forecasts by lessening systematic biases.
Here are the errors from MOS based on the NWS NAM model. Same problem: too cold during the summer....even worse than the NWS forecasters (who at least improved things a bit).
Another way to check out this issue is to go to the weather forecast verification site forecastadviser.com. Below is their accuracy statistics (% of time the forecasts were within 3F of the truth)...in this case for Seattle. For the last month, the high temperatures forecast by the National Weather Service was accurate about 36% of the time--a bit worse than persistence, which is a very low bar. Persistence is simply forecasting for tomorrow what happened today. Not good. The Weather Channel did far better.
I suspect there are several reasons, both connected with the warm temperatures. First, soil moisture is less than normal because of the dry/warm conditions. The figure below shows the percent of normal of soil moisture (from the UW Hydrological Monitoring Site). Parts of western Washington are at record lows, many with less than 10% of normal soil moisture. Dry soil tends to warm up more, everything else being equal (less water to evaporate and evaporation causes cooling).
Second, we have extremely warm water offshore (the Blob), which causes the air moving into our region to be warmer at low levels.
The models know about all this, but the Model Output Statistics, trained on years of normal conditions, make the wrong adjustments because they use long-statistics. This is a hypothesis, but a reasonable one.
Why is the Weather Channel doing better? Because they don't use MOS. They have their own very sophisticated model statistic postprocessing system. One that is more adaptive and able to compensate for short-term unusual conditions.
Tomorrow will be very warm by the way...enjoy.