Quick: I'd have a good look at GP ReCyko cells as a good starting point

Some assumptions:

• For temperature >> zero C a lower reporting rate may be OK. Presumably the station could transmit say once per hour as a confidence check BUT drop into "fast" mode whenever a temperature drop threatened.
• Assume total TX cycle at 500 mW is 1 second per 10 minutes.
• 1s/10mins x 500 mW = 5/6 mW. Assume 1 milliWatt means draw. Adjust as required.
• Assume ~~10V pack for convenience --> 0.1 mA mean draw.
• 6 months ~= 4000 hours or ~400 mAh delivered capacity required.
• So worst case 6 months of late Autumn - Winter - Spring a 3 times over provision of battery capacity to allow for temperature effects requires about 1200 mAh. 4x = 1600 mAh. 5 x = 2000 mAh.

So AA cells of 1200 mAh in many chemistries would probably work (see below).

Primary cells are not a terrible idea. AA Alkalines at 2500+ mAh and a degradation factor of say 3x average would last about 2 years.

Solar recharging looks highly attractive. 1 mW mean power = 24 mW.hour / day. ie 1 hour of charging at say 50 mW would suffice to keep the battery always topped up. That's about a 1 square inch mono or poly crystalline solar panel exposed to one "sunshine hour" of sunshine per day on average. As a guide, that would be more than adequate in NY NY in January (worst month) , and 3 square inches would work in Moscow in Russia in mid winter.

In all matters to do with batteries YMMV widely - experience is unfortunately the best guide as to how good claims are. Results can be very dependent on manufacturer. In many cases, if you don't have the volume to do your own investigations (and few do), choosing a reputable brand label which has been in the business for a substantial period and which is liable to have researched the product they sell and stand behind the results.

I would accept as likely to be approximately true, technical claims by Chinese makers BYD, BPI and GP (GoldPeak). Also mainstream labels such as Sanyo (make their own cells, usually very competent) etc. Most others I'd treat with far more care. Note that GP are so successful that there are Chinese clones of their products.

Note that for 6 months + lifetimes the self discharge rate of the battery used becomes relevant. NimH is very poor, NiCd is poor, lead acid are good and Lithium Ion and LiFePO4 are very good. Low Self Discharge (LSD) Nimh are very good. The latter are available as eg Sanyo Eneloop and GP ReCyko. Also now many more.

For general use the GP ReCyko are excellent. I have not yet found data on low temperature operation but I guesstimate that, based on other NimH data they'd be OK to say -20C at a sensible derating of their capacity - say 33% of nominal.

LiFePO4 are usually specified as operating to -20C. They will have substantial capacity loss at this temperature. Here's one manufacturers example. Reputable manufacturers are generally happy to provide detailed data to genuine inquirers.

Capacity loss That graph is from Hi Power group which seems to be a typical Chinese manufacturer. All such information should be regarded as a starting point and "due diligence" is definitely required for anything regarding batteries.

GP rate their NimH batteries as operating to -20C. I have no data on this but its probably available. Here's a sample GPn1500 mAh NimH datasheet I'd expect lower capacity batteries within a given size range to have somewhat improved low temperature operation all else being equal. eg 1500 mAh AA better that 2500 mAh AA. But the increased initial capacity may cancel this out. (Larger capacity batteries squeeze in all possible active material at the expense of electrolyte volume etc).

You can get NimH and NiCd in special low temperature versions. Here are some examples from Lionik battery Co another typical looking Chinese manufacturer.

You will be able to get US branded and sold low-temperature batteries. These will almost invariably be Chinese made. Choosing a reputable US label gives you some confidence (or hope) that they have done the due diligence required to ensure that claims meet reality.

Here's a somewhat informal comparison of LiFePO4 with 4 other Lithium battery chemistries. Note that in 3 cases Tmin is given as -20C. Proves little but worth noting.

It seems you are aware of the problem with the reference voltage changing and if you use a device like the TMP36 (fixed 10mV/degC) there is nothing you can do other than use a voltage reference from a chip to stabilize things.

However, if you are using an RTD or a thermistor then the problem won't arise. You ADC is making a ratiometric measurement - it compares the ADC input to its reference voltage BUT, if you power the RTD or thermistor (via a suitable resistor) from the same ref voltage it won't affect readings. If the ref goes up 10% then so does the voltage into the ADC.