I spent the first 13 years of my USGS career as a field technician, the first 6 as a hydrologic technician and the next 7 as a professional hydrologist. All of those years were in what was called the basic data section or in a field office, first Eugene, Oregon, next Phoenix, Arizona and last Tacoma, Washington. The USGS requires all field people to have a thermometer to measure water temperature.
It is a requirement during every visit to any surface water (creek, stream, river, etc.) site or gage to measure the water temperature and record the date and time associated with the measurement. This goes back to the 1950's and the information is available on every field inspection summary (from 9-207) for every site or gage.
At some of the sites, the USGS operates a water temperature recorder. The technology of the water temperature instruments has obviously evolved from the 1950's thermograph, to digital sensors with paper tape recorders, to fully electronic sensors and recorders. The data is published in the Annual Data Reports for each state.
At the end of the water (also government fiscal) year (October 1 to September 30) the data is reviewed and published. The water temperature data was produced and reviewed under some rules which the USGS felt at the time were reasonable. It's safe to say now, it wasn't right, just reasonable to avoid conflicts in the data. Some of those rules still exist, which are based on some assumptions which aren't entirely accurate or correct.
Really? Yes, the USGS thought they were common sense and they weren't interested in getting embarrassed with the data. So during the production and review process there were checks for these rules and the data was "adjusted" to accommodate the rule. The adjustment wasn't great and really insignificant, but still an adjustment. And they were?
First, flowing water freezes at 0 degrees C (water temperature was always published to degrees Centigrade for international standards). This isn't exactly true and flowing water can be colder than 0.0 degrees, even as much as -1.0, but that's extreme. It's common to measure and record temperatures up to -0.3 degrees when there was sufficient flow to keep the water from freezing.
I know this to be true because I measured it one day at one of the gages on my field trip. Winberry Creek near Lowell, Oregon (12-150800). It was a cold January day, where everything was frozen. I got to the station about 8 am and after checking everything I went down to the bank to read the outside gage and take the water temperature. I measured -0.2 degrees.
The air temperature, something we also measured, was far colder (can't remember exactly, but very cold, in the ten's). Since I had my wading boots on, preparation for making a discharge measurement, I waded into the stream and up and down the stream taking more measurements to check.
And sure enough, while there was ice in the stream near the banks, the entire stream was at -0.2 degrees. The recorder also recorded this value. We later changed the value to 0.0 degrees to fit the rules for publication. So, when you see 0.0 degrees in the USGS data, it could be from -0.2 (or colder) to 0.05 degrees. Trust me. As they say, been there done that.
The other two rules I think have been either revoked or ignored in recent years because it doesn't make sense or match reality. It never did, the USGS only decided they didn't want to publish data that seened odd. And how so odd? The rules were simple.
The first is that the lowest (coldest) recorded temperature of any day couldn't be higher than the highest (warmest) temperature of any adjacent day. The second is the corollary, the highest temperature of any day couldn't be lower than the lowest of any adjacent day.
This rule breaks down under two situations.
First is when the weather changes dramatically across midnight from either a cold spell to a warm spell (storm or cold fronts) or the reverse. This happens in the data because you're recording data at 11 pm, the last measurement, for one day and midnight for the next, the first measurement for the next day. The water temperature for the last value of the day (11 pm) could be lower than or higher than the opposite extreme of the adjacent days.
The second is at gages below reservoirs where adjustments are made in the outflow from differents parts of the reservoir. Normally there is adequate vertical mixing in small to moderate reservoirs where the water temperature isn't so extreme, but for larger reservoirs there can be significant differences in the vertical profile from the different inlets to the dam and in the subsequent outlfow.
This means changes in those outflows can significantly change the water temperature of the river. This is seen in the data for large reservoirs where the water temperature will jump several degrees in a short time, usually hours. This, if timed right such as overnight, will cause the temperature anamoly in the data between adjacent days.
The USGS overcame this rule by making every day 25 hourly values and including both midnights in the dataset for the day. This avoids this conflict, but it will happen occasionally that the same value (midnight) could be recorded as the extreme for both days, which is another conflict, but one they'll live with the rarity of it.
Anyway, this is just my experience and thoughts on water temperature.