ForecastWatch September 2010 Customer Newsletter
I wanted to let you know that the August 2010 aggregations are complete. There was a little delay as the Environment Canada website kept having availability issues ("503-Service Unavailable" errors) while getting the Canadian actuals.
I also wanted to let you know of two re-aggregations that will be occurring over the next week or two. The first is highmos aggregations from October 2009 through January 2010 for the U.S. and from June 2009 through January 2010 for Canada. It was discovered that code introduced in October incorrectly aggregated the highmos data on the lowmos days out, causing some odd numbers. The error was inadvertently fixed when the aggregation code was greatly improved in late February (one of my first projects after going full-time on ForecastWatch). Specifically, the overall average counts are incorrect, which affects the percentages. Overall, the changes are not large, but wanted to let you know.
Second, precip (not POP) forecasts for both U.S. and Canada from May 2010 through August 2010 (change was not in by these last aggregations) will be reaggregated. About 8000 forecasts that contained "T-Storm" (such as "T-Storms Early") were not being correctly counted as precipitation forecasts. This will result in a largely negligible change for the precip pages (there are approximately 10 million forecasts for the period), but want to be complete, especially for the thunderstorm aggregations.
Finally, starting with these August 2010 aggregations, a small improvement was made to low temperature aggregations. Providers who forecast overnight lows didn't have the last low temperature forecast of month scored in either the month the forecast was made nor the month the forecast was for. It ended up in a gray area and never was aggregated. Now, low temperature forecasts for overnight on the last day of the month count towards that month.
I have put code in place to ensure neither issue occurs again, and improved the audit checks to look specifically for these cases. Over the past several months I have improved the internal checks in the ForecastWatch system to check and recheck as much of the data as possible, so that the quality of the aggregated data remains of the highest quality. Thanks so much to you guys for helping spot issues when you see them and offering suggestions on how to make ForecastWatch even better. Those comments and criticisms are always welcome (as is praise ;-).