BirdForum is the net's largest birding community dedicated to wild birds and birding, and is absolutely FREE!

Register for an account to take part in lively discussions in the forum, post your pictures in the gallery and more.

Reply to thread

Message: <blockquote data-quote="johnraven" data-source="post: 1581443" data-attributes="member: 14963">The giant peak was Saturday 29th September 2007.  This was a period of brisk N to NE winds, with NE to E winds in Copenhagen, with falling pressure and rain.The variables of the model are as follows:(1) Day of month - the number of records tends to increase over the month.  I tried a non-linear function (polynomial), but only the simple upward trend passed the 95% confidence test.  The number of records increases by 0.127 each day.  (2) Wind direction today - only N, NE and E wind directions matter.  I thought SE might matter, but it doesn't.  N are NE are best, both boosting records by 3.(3) Wind direction yesterday - again N, NE and E.  NE is the best, boosting records by 4.3.  East is next best, by 2.(4) Wind direction day before yesterday - only NE proved to be statistically significant.(Overall NE wind is therefore best)(5) Weekend - Sat and Sun boost records by 1.6, unless there is an easterly wind (NE or E or SE) then records are boosted by 3.5.  I suspect easterlies attract more birders.(6) Pressure - the pressure in isobars (e.g. 1020).(7) Rain - actually, this drops out of the equation when I include Pressure, but is significant if I do not.  The two are correlated.  Rain is good, btw.(8) Copenhagen wind yesterday - East wind is best, then NE wind.  Surprisingly, SE is not very good, although I note a lack of SE wind in my data.I also tried wind speed, but the relationship is too complex, for my 150 observations (5 years of 30 days) of data.  A strong SW is bad, but a weak NE is not as good as a moderate NE.  I tried a few possibilities, but nothing has worked.  I need to think about this one.The functional form of my equation is not right.  I use simple OLS, which means the prediction can be negative (see the chart).  Obviously the actual number of finds is limited to zero and discrete.  But I just cant be arsed to deal with the statistical complexity of a non-linear discrete function - I would rather test the OLS model.I'm very keen to try any suggestions you might have though.  I will need daily data back to 2004.John</blockquote>

Verification