nwt climate notes precipitation (ctw, v1 saddle precip)

braindump workflow for precipitation QA and infill

aims: similar to temp data, read in raw-ish annual ppt (probably in jan of the following year?.. may be problematic if adding by water year).. scan for problems (summary report), flag and NA (there aren't too many potential issues that can be flagged since ppt is so variable), backfill any qdays > 1, infill the rest, reality check, write out.

saddle will be tricky with overcatch, but metadata search shows not too much to be done. it is near impossible to say what could have happened on any given day in the winter up there. just have good documentation of everything we know. note possible undercatch at d1 too

0. prep data

• get in format to run functions

1. scan ahead: summarize possible issues

• missing days (with available backfill vs. no data)
• high precip days (esp. when other stations have no snow)

2. backfill qdays > 1

• when 0 it's easy, otherwise regressions

3. infill all else (data truly missing)

4. reality check

• do infill values look realistic?

5. final report and dat out

• summarize trends (try to match what ML used to make, e.g. # ppt days, monthlies and annual trends)
• final data to publish on EDI

options

• 'snotelr' is an R package to dynamically read in snotel data.. however, it does not download data with flagging. it does download data with site info, which is nice.
• 'amerifluxr' similarly dynamically reads ameriflux data.. but process seems not too much better than requesting static files if only infilling once per year (multi-step code process to get from downloaded zip to actual data frame in environment via amerifluxr.. similarly flagging may not be there).

GHCNd data to download to follow methods in Kittel et al. 2015 paper (+ other sites that are nearby and possibly useful)

WEST SLOPE (west of CD) N -> S

• Grand Lake 1 NW (montane, Station ID: 053496, Longitude: -105.83222 Latitude: 40.26694, Elevation: 8720 ft.)
• Grand Lake 6 SSW (montane, Station ID: 053500, Longitude: -105.86667 Latitude: 40.185, Elevation: 8288 ft.)
• Fraser 3 (montane, Station ID: 053113)
• Fraser 6 (montane, Station ID: 053113)
• Berthoud Pass (alpine tundra, Station ID: 050674, Longitude: -105.78333 Latitude: 39.8, Elevation: 11313 ft.)

EAST SLOPE (east of CD, same side at D1, SDL and C1) N -> S

• Estes Park (montane, Station ID: 052759, Longitude: -105.48583 Latitude: 40.37666, Elevation: 7480 ft.)
• Loch Vale RAWS (alpine forest)
• Allenspark Lodge (aka Allenspark 2 NNW aka Allenspark 2SE..) (montane, *used to assess possible inhomogeneity in C1 ppt in 1960s, ID: 050183, Longitude: -105.50194 Latitude: 40.18806, Elevation: 8215 ft.)
• Caribou Ranch (montane, Station ID: 051342, Longitude: -105.51667 Latitude: 40 Elevation: 8366 ft.)
• Nederland 2 NNE (055878, montane, Longitude: -105.5 Latitude: 39.98333, Elevation: 8240 ft.)
• Coal Creek Canyon (051681, montane, Longitude: -105.38472 Latitude: 39.89584, Elevation: 8950 ft.)

OTHER MAYBE HELPFUL SITES

• Boulder 14 (US Climate Ref Netowrk site)
• Allenspark 3NW (Station ID: 050185, Longitude: -105.51833 Latitude: 40.22889, Elevation: 8500 ft.)
• Nederland 7.2 NW
• Sawtooth (05J45S)
• there are several stations in the Nederland area and a few in Rollinsville
• Snotel station at Berthoud pass (still in operation as of 2022, GHCNd at Berthoud stopped)

TO KNOW ABOUT:

• Silver Lake (057648, runs 1910-1950s .. I think includes one of the biggest snowfalls in state history [in the 1920s]).

These sites' data can be found on the CSU Climate Center page (https://climate.colostate.edu/data_access.html), but grabbing data from there is annoying because you can only query 1 site at a time and the data displays in html as a table on the page (one must copy/paste the table into a text file or spreadsheet -- risk for human error). So use NOAA's GHCNd data site to the extent possible (which is also annoying but less annoying and less error prone than individually copy/pasting data tables).

NOAA GHCNd query tool (on version 3 at the time of writing, July 2022): https://www.ncei.noaa.gov/access/search/data-search/daily-summaries. The least headache-inducing way I've found to search for sites is via using the map tool and making sure the box to update the search results as you move the map is checked. This is also another option (and the one I ended up using to get data -- most of these are COOP stations, but Boulder 14 is US CRN): https://www.ncei.noaa.gov/maps/daily/ There are other .gov sites to access these data and their search function both returns too few results at times (e.g. when using a town or geographic place point location) and too many (e.g. searching a county).

CTW recommends a bulk download. Be sure only the variables you want are selected in the data request (by default they will give you all variables possible.. which is a lot), and that the units reported are the ones you want (e.g., inches vs. mm, F vs. C).

• to watch out for: if site has data pre 1965, take care using its precip data. Rain gauges were systematically changed per US Weather Bureau guidelines from unshielded to shielded across the US in the mid 20th century (and more broadly across the world throughout the 20th century). This change in instrumentation creates an artificial step change in precipitation time series data that go back to those decades. Instruments were not changed at the same time/same year, so one must look at the metadata for the site and hopefully the change date will be documented. If not, use statistical methods to screen for changepoints. Yang et al. (xxxx) is the reference paper for seasonal adjustments to make if you know (from site history) or detect a changepoint.

• similarly (and this is for temp, but also a US Weather Bureau matter so may occur at this precip data sites) -- if using COOP data, pay attention to any changes in time of observation per day. Morning readings bias towards colder temperature, afternoon readings bias towards warmer. If the bias is consistent throughout the dataset, it may not matter as much for using in a regression to infill a target site, but if observation times changed in a record that is something to deal with before using as a reference series (or don't use it as a reference series if you have other options). There is another network of weather stations (USCRN = US Climate Reference Network, run by NOAA) set up later with more precise electronic instrumentation to monitor climate change (there is a reference paper I will dig up). Those would be better data to use if not scrutinizing the metadata closely. The reference paper compares paired COOP-USCRN data and finds a systematic difference in readings between the two, but it depends on where in the country and what type of system (e.g. mountain with variable weather and steep climatic gradients vs. temperate plain with more moderated climate across the region)

• At NWT, unshielded US Weather Bureau standard totalizing gauge used at C1 through 1964, and at D1 through 1969. A Belfort weighing-bucket with Alter shield, encircled by Wyo snow fence, was installed at C1 in 1961 and at D1 1965. TK and Co used the overlapping years of instrumentation at D1 (1965-1969) and C1 (1961-1964) along with regression equation provided in Yang et al (USWB global comparison project) to convert unshielded data to Belfort comparable data. Seasonal adjustments were needed for D1 but not for C1 (issue: winter undercatch for unshielded data at D1). When installed new instrumentation at C1, moved location from forest clearing to within forest for natural shielding. TK et al. 2015 notes they define 'warm season' as May-Sep, and 'cold seasonal' as Oct-Apr, and they found seasonal effects. Include May in warm because it is a shoulder season and historically month when snowmelt initiates.

• Also, notes in NWT data reports that climate data for end of 1970s is questionable.

Because of the above, and climate trends are in 30yr normals anyway, for modern-record infilling maybe just use 1980 onwards (but can compare to infill values when use full record).