Thursday, February 20, 2014

8-14 Day Forecast Visualization for Alaska

8-14 Day Visualization A few weeks ago I decided to download the GIS shapefiles than NCEP makes available on their website for the 6-10 Day and 8-14 Day extended forecasts. I am particularly interested in these because the push the boundary between a weather forecast and a climate forecast. By downloading the shapefiles and then generating temperature anomaly maps from reanalysis data, we can visualize how effective the forecast actually was (note: this is not an actual verification). The YouTube video below is a compilation of 30 consecutive 8-14 Day forecasts beginning with the December 31 to January 6th time period and ending with the the February 11th to February 17th time period (no forecasts are issued on the weekends). The GIS maps (left) are placed next to the reanalysis data (right) for comparison. There are a few days where the GIS data may not have been processed correctly (repeating polygons on consecutive days). Nevertheless, it makes for an interesting visualization. With the exception of a few days in mid-January, they 8-14 Day forecasts were actually pretty good.


  1. Very nice presentation Brian, especially for we visual learners...Like, what did she just say?

    I've wondered for some time why the NCEP excludes Canadian data from their forecast products?

    Is there a treaty in need of signing? It seems odd. It's available elsewhere:


    1. Thanks Gary. As for why the NCEP extended forecast excludes Canada but the Canadian extended forecasts include the U.S., I can only speculate. My first guess would be that someone, somewhere complained about their tax dollars providing forecasts for people (Canadians) who are not paying into the system. My second guess would be that the Canadians can better justify the expense within Environment Canada is there is no competing product coming from the U.S. Those are just guesses. It could also be that we plan on invading Canada someday and want to keep the element of surprise.

    2. Actual the invasion is already happening. There is a little bleed over the borders. And I guess Mexico and Cuba are next since they are not included in the forecast too.

      Really cool visual. I would say there was maybe about a 70% matching rate - which is perty good for the forecast being 2 weeks out. Forecasting has really improved in the last decade. Though expected with better computers, it puts a damper on the idea that the weatherman never gets it right.

      As far as verification, I always thought that it would be useful to see what a 7-day running daily correlation would look like. For every consecutive day in a certain period take the fitting value between the 7-day forecast and the actual 7-day temps. Do this over a couple months. You should get a multi-day graph of correlation coefficents showing how accurate the 7-day forecast is for any particular day. Then look at the weather maps and figure out what happend.

  2. Rick has pointed out the the CPC map categories represent the liklihood of being in the upper of lower tercile (+/- 0.43 standard deviations from normal). The reanalysis data shows raw departures from normal. Unfortunately the reanalysis data does not have daily temperature standard deviation values. I could probably calculate them from the climate normal files, make a grid of those data, and then compare them to the netCDF files from ESRL - but that would be a lot of work to say the least.

    CPC has a neat verification tool for the Lower 48: . Hopefully it will be extended to Alaska soon.