westonslaughter / data_processing

Scripts for the retrieval and harmonization of federally funded river and stream data (USA)
MIT License
0 stars 0 forks source link

DOI

data_acquisition directory structure

new network/domain acquisition SOP

this is a work in progress. update as we go, so these steps generalize

note: we have three data levels within macrosheds. these are used in naming processing kernels.

begin SOP:

  1. for a new domain or new network, research it's naming conventions. fill out name_variants.csv with a new column
  2. you might already have a retrieve.R script for this network/domain. if not, borrow one from some already handled network/domain. use this as a template
  3. step through ms superstructure until processing can't proceed successfully without modification
    1. if error/incompatibility occurs in retrieve.R modify retrieve.R to accommodate new idiosyncrasies
      • mind existing retrieve.R scripts as you develop. attempt to harmonize accross them
      • add, remove, modify helper functions and/or the body of retrieve.R in order to accomplish this
      • ideally, one day, retrieve.R will be flexible enough to handle all network-domains. then it can be a single function in a macrosheds package
    2. if error/incompatibility occurs in the outer retrieval function (e.g. get_neon_data, get_lter_data), do as above.
      • e.g. modify get_networkx_data to accommodate new idiosyncrasies. try to harmonize it with get_neon_data and get_lter_data as much as possible
    3. borrow a retrieval kernel (the innermost level of retrieval processing, where the data are actually downloaded) to use as a template. if it doesn't work as-is, modify it
  4. you might already have a munge.R script for this network/domain. if not, borrow one from some already handled network/domain. use this as a template
    1. repeat above instructions for munge.R
    2. repeat as above for munge function, e.g. munge_neon_site()
    3. error/incompatibility should occur in the munge kernel, the innermost level of munge processing
      • here's where you need to accommodate somewhat arbitrary data structures in whatever form they arrive. within domains, we can expect good consistency. within networks, some. between networks, anything goes.
      • there will be errors in the external database. build to accommodate unforseen issues:
        • naming inconsistencies
        • missing data
        • incorrectly specified data
        • unexpected columns, etc.
      • make sure datetime is converted to UTC (lubridate::with_tz and lubridate::force_tz are great for this)
      • make sure variable units are converted to those specified in data/general/variables.csv
      • make sure column names are converted to the ones we use in ms
        • "site_code"
        • "datetime"
        • e.g. "spCond" (again, consult variables.csv for variable names)
      • if necessary, separate one site into multiple sites. we have to do this for some neon "sites", where there's secretly an upstream sensor array and a downstream sensor array.
  5. we haven't started deriving data yet, but the process will be similar to the above. build out this section when we get there