Temperature is a major factor for completing workouts, marathons have an ideal starting temperature, etc.. If I put my trainer in an unheated solarium during winter the ambient temperature will be quite different compared to +35C, sunny, with 75% humidity.
My garmin records temperatures (and GPS can be used to correlate with wind/weather data) in the FIT file, perhaps this could be used as an input to the ML model.
Interesting suggestion… Definitely a factor in performance but I think similarly hard to take into account to elevation/altitude. For example for your predictions you’d then also have to input the elevation/expected temperature of your event + measurement errors (e.g. watch on wrist would overread temp). Will think about it.
Even if the idea is interesting, the temperature given by Garmin products is linked to the wearing conditions or enveloppe product.
Per example, temperature on the GPS is not well placed and gives always 5/10° higher than the real temperature in colder conditions. Otherwise, 5/10° below than real conditions during summer conditions (here, closed to 40°C).
For the watch, same trouble especially during the winter, the watch is below suits or jacket.
It will be interesting to take into account watch temperature and an accuweather (or like this) real conditions.
For the stryd per example, a wind sensor is active during the training. More clear.
I found my Garmin if I wear it outside will drop from 25C or so down to 5-10 even if it is sub-zero, because of heat from my body. In this case the GPS-based outdoor estimation would be much better.
Indoors, such as on the treadmill in a gym, it is higher than ambient temperature. I have not done an analysis to see how reliable it is (long-sleeves vs. short sleeves, being near or not near a vent, etc.). And ambient is different from airflow from say a fan.
Edit: I forgot, the CORE temperature sensor. I don’t know how popular they are, other than it appears on GCN frequently.