ForecastWatch July 2007 Customer Newsletter
Welcome to the July 2007 Newsletter of ForecastWatch.com. You are receiving this email as a subscriber to a ForecastWatch.com product or by signing up for ForecastWatch news. Please see the bottom of this email for instructions on how to unsubscribe from this newsletter.
It has been very busy at Intellovations headquarters, but you'd never know it besides the recent lack of newsletters. We have spent the better part of the last nine months building the next-generation ForecastWatch and ForecastAdvisor. Unless you subscribe to ForecastWatch online, you might not realize how much has changed, yet.
Both ForecastWatch Online and ForecastAdvisor websites have moved to a more modern, easier to build, web platform. ForecastWatch Online had been built using Quixote, and ForecastAdvisor using Ruby on Rails. Both are now running on Django, and both are now written entirely in Python.
The back-end has substantially changed as well. We are doing three times the aggregations, taking several private feeds, and providing more weather forecast data than ever before. We can now calculate forecast temperature accuracy not just against the morning low and afternoon high, but against 24-hour high and low (for those forecasters that provide 24-hour forecasts), and MOS-defined high and low. For precipitation, we are now calculating Brier scores, and Brier skill scores against persistence forecasts. The online compare product now compares precipitation forecasts in addition to just temperature.
Our ability to calculate aggregations has been enhanced as well. We've introduced significant performance improvements that have taken monthly aggregations from a week to a day. The database is now running on a dual-Xeon server rather than a Pentium 4. So the next time you visit ForecastWatch or ForecastAdvisor, and it looks much the same as it did, realize that under the hood, it is doing a lot more and providing a lot more data to more people than it ever has before.
Because of our new infrastructure, we can now produce Brier scores for probability of precipitation (POP) forecasts. To celebrate that fact, we have released our first comparative study of short-term (one-day-out) probability of precipitation forecasts, covering the period from October 2006 through June 2007. Probability of precipitation forecast accuracy and resolution scores are more accurate and valuable the more forecasts are compared. This study includes nearly 750,000 (yes, that's three-quarters of a million) one-day-out forecasts, and is the largest POP forecast accuracy study ever released. You may view the study for free at:
http://www.forecastwatch.com/docs/Short_Term_POP_Accuracy_2007/
We've audited and loaded the June 2007 accuracy data into both ForecastWatch (http://www.forecastwatch.com) and ForecastAdvisor (http://www.forecastadvisor.com). Overall, temperature accuracy improved from June 2006. Overall one- to five-day-out temperature RMS error dropped from 4.65 degrees a year ago to 4.58 degrees this past June. High temperature RMS and standard deviation went from 5.01 and 5.00 degrees respectively, to 4.87 and 4.79 degrees. Low temperature RMS and standard deviation went from 4.29 and 4.28 degrees respectively, to 4.28 and 4.25 degrees.
Please let us know what you think at ideas@forecastwatch.com . We welcome your ideas and suggestions to make our services better and more useful.
- Copyright 2007 Intellovations, LLC.
