My initial data is weekly cloud to ground lightning flashes, binned per week of the year, in a heatmap display. Some week to week changes in lightning activity really present themselves in the heatmap, but I'm trying to determine if some of these week-to-week changes are statistically significant (indicative of some underlying weather/climate process) or just an artifact of binning the data by week-of-year...or just noise.
So in the above image, I noted a sharp drop in lightning during the week of Jul 30-Aug 6, but I want to find out if this is statistically significant. So I download the daily lightning data and compute a 7-day moving average for the data in excel, and usually I'm just showing data for 4-6 hours of the day that show the significant change. Now that I've done some smoothing, I want to put some upper and lower control limits on a graph and see if any ups or downs cross the limits for statistical significance. I found some examples of this online, but they were for static control limits. I want the control limits to adjust according to the seasonality of the data. So I created 7-day stdev and lower/upper control limits based on avg+/(2*stdev). Does this seem appropriate? The chart shows very little crossing the limits, and the change noted in the heatmap doesn't even come close. Do you have any other suggestions for doing this kind of test?
I can add more detail if you need. Thanks for any help! Here's a link to the spreadsheet: lightning_spreadsheet