During a recent tour I had a chance to cycle alongside some mates (of similiar bike/body weight) with power metres on their bicycles and it showed that the run.gps power estimate was much too high. Much as I would like to be producing c. 400 watts up a 8% gradient with an average of >500 watts over 4 hours, I'm afraid it is just not possible :(
What algorithm are you using? It consistently overestimates instant power and almost double/triple average power.
Remote measurement of power uisng proxies is notoriously difficult on a bicycle, especially when on the flat/downhil/drafting/windy conditions, but for slow/steady (<10 kph), uphill (>5%), windless climbs it is possible to get a reasonably accurate figure because many variables are minimised.
Apart from improving the algorithm itself, a few suggestions which may help to improve the accuracy:
1. Enter bicycle weight in user setup
2. Subdivide average power in to laps, e.g. mountains climbs
3. Enter a estimated 1 to 5 drag coefficient in user setup, e.g. 5 = sit up straight bicycle with panniers and baggy/flapping cloths, 1= full time trial aerodynamic racing bicycle with body suit