Findings: annual failure rates seem to not depend on temperature at all between about 30°C and 45°C, with failure rates increasing both above and below that temperature range.
Findings: annual failure rates seem to scale linearly with average drive temperature, roughly doubling from 5% at 32°C to 10% at 43°C. Variation in drive temperature seems to have a small, but negligible, negative effect.
Findings: Temperature seems to have no effect on annual failure rates, but average temperatures of drives in study were between 21°C and 31°C, so their conclusions are not that helpful.
The 2013 paper unfortunately is stuck behind a paywall, but message me if you would like to read it.
My suggestion given the work in the field would be to keep temperatures around 30°C if you can because at higher temperatures, you could be reducing the lifespan of your drives.