I have GPS tracking info on 107 rides of a century or more, including BMB and nearly all of PBP, plus four SR series. The tracking info is broken down into (typically) five mile intervals, over which I know speed, moving time, stoppage time, ascending and descending. And I know what time it is, so I know if it's dark out. Plus from weather records I know the temperature through the day. My forecasting model for moving speed is:

MovSpeed=f(Intercept,Bike & gear weight, ascent rate (feet per mile), descent rate, dist into ride, miles of riding in previous 30 days, drafting, nighttime, sickness, tandem, and extreme temperature).

Running this model only for rides that were brevets or permanents, I get an adjusted R-squared of 57%, i.e. the model explains 57% of the variability in moving speed. All variables have significant T statistics. Parameter values are as follows:

Intercept=21.2
Bike & gear weight= -0.074
Ascent rate= -0.053
Descent rate= -0.010
Distance = -0.00657
PrevRide= 0.000626
Drafting= 0.621
Night= -0.465
Sick= -0.281
Tandem= 0.673
Extreme= -0.631

So the interpretation is: average speed would be 21.2 mph with a zero-weight bike, on flat terrain, at the start of the ride (dist=0), being generally fit from commuting but not having done any centuries in the prior month, riding solo in the daytime, healthy, and in temperate conditions (i.e. temperatures between 32F and 90F).

For every ten pounds of extra bike weight, I get slowed down 0.74 mph. I typically have 40 pounds of gear so that makes a base initial speed of 18mph on the flat ...

A 1000 foot climb over 5 miles would be an ascent rate of 200 feet per mile and would slow me down by 10.6 mph (200*-.053). So if my base speed is 18 then I'm down to 6-1/2mph, which sounds about right for an extended 4 percent grade.

Descending apparently slows me down, too ... I'm guessing that this is because if you're in terrain with ascents then you also have descents, thereby violating the maintained assumption of independent explanatory variables, so the descent variable is picking up some of the variability from the ascent variable.

For every hundred miles into the ride, I slow down by .657 miles per hour. So if my load-adjusted base speed is 18mph at the start, then by 750 miles into a grand randonee, then I have slowed down by 4.93 mph to 13mph. Sounds about right.

If I rode an extra 100 miles within the prior 30 days then I am somewhat fitter, and that makes me faster by .0626 mph on this ride.

Drafting makes me 0.621mph faster. Nighttime makes me 0.465mph slower. Being sick makes me 0.281mph slower. Riding the tandem makes me 0.674mph faster (about the same as drafting). Really cold or hot makes me 0.631mph slower.

Your mileage may vary.

Nick

PS, In case you're wondering how I know about drafting ... it's a sufficiently rare occurence that I'm drafting anyone for any length of time, so for those times when I have been drafting during an interval, then I can add that to the info for the given leg.