emmamendelsohn
commented
9 months ago That said, because we're calculating the anomalies relative to a fixed present day value, it probably wouldn't affect performance to use our current approach. It's just slightly more interpretable and explainable to do it as above?
Currently we calculate anomalies using the mean of the lagged x days (0-30, 31-60, 61-90) and the historical mean for the present day of year. I realized we should probably take the mean of the historical means for the lagged time period. For example, if we're calculating an anomaly for a 61-90 day lag, we'd want the historical mean from the days of the year 61-90 days previous to the present date. Because we already have the historical means for each DOY pre-calculated, I think we can get away with using the mean of these means (as well as the means of the SDs when normalizing).