I contribute CPU time to this project using Windows via BOINC, and I only occasionally take a peek at the screen saver. I am running the application entitled "HadCM3 Coupled Model Experiment Optimised File I/O" and the Work Unit Name is "hadcm3ivolc_l2g0_2000_80_66001744_2" (currently 5.519% complete). When I checked the screen saver last night, I noticed that climateprediction.net was processing Feb 29, 2005 at 07:30. I realize that we are dealing with many years of data, but shouldn't the date algorithm at least properly account for leap years only in actual leap years? If every year is processed as a leap year (i.e., 366 days/year), then it is adding 60 unnecessary days of processing for every 80-year block of data it is processing. A net two-month date shift in climate after 80 years is, clearly, statistically significant. It makes winter appear to start two months earlier after 80 years of processing in this model, rendering it useless. I hope I am wrong, because this, to me, is a very important project.

---

Climate models use a standard system of 12 months of 30 days each.

This is noted in the 2nd README file, linked from my sig below. Look in the 3rd section (INFORMATION about the model): Feb 30th (360 day model calendar)

---

Backups: Here