Regarding the general temperature dependency, outside of the specified environmental range, you could get the PicoScope calibrated for the additional environmental range (as long as you don't exceed the operating range limits).
You are correct regarding the type of averaging used. For downsampling ratio_mode_average calculates the average as the sum of all the samples in each block, divided by the size of the block, so it is block averaging, while Enhanced Resolution Mode implements a type of moving average filter on the data. So, downsampling is not a good way of cleaning up the signal. Better methods would be:
Implementing a moving average filter in your application. It is a fairly simple implementation in code and there are numerous examples on the Internet.
If you have a repetitive waveform, segmenting the sample buffer to collect multiple waveforms and then averaging the samples across the waveforms.
If you are retrieving samples in Block mode the downsampling is done in the software if the downsample ratio is 9 or less. If the ratio is 10 or greater the downsampling is done in the hardware. So, if you are software downsampling you will be retrieving all the sample data, which will increase the CPU overhead. If you are using streaming mode then the downsampled data is sent along with the streamed data which can also increase cpu overhead.