ST 590G -- Computation for Data Analysis

  Second Assignment -- due Tuesday, 25 September 2012

    In the 'ushcn' directory are a few files of interest taken
    from the US Historical Climatology Network page on the CDIAC/ORNL
    site.  The file 'data_format.txt' has information on how the data
    files are organized.  I have downloaded and uncompressed files for
    several states to the 'ushcn' directory.  See the file 'h2f12who'
    for which file/state you should analyze.

    1) From the stations file 'ushcn-stations.txt', read the station id 
    number, latitude, longitude, and elevation.  Note any missing values 
    -- you may not want to deal with those stations.

    2) From the data file, read the TMAX values and the flags for
    each day and month.  Notice that each record holds the information  
    for 31 days.  Note that missing values are coded as '-9999'.)
    If a flag is thrown -- either the MFLAG or QFLAG, then record that 
    observation as missing.  Read only the element of interest (TMAX, 
    maximum daily temperature) and delete the others.  

    3) Since we have daily values, there's a lot of data -- monthly 
    would be easier.  So construct a dataset for each month and year
    from the daily values, using the mean of TMAX.  If too many days
    are missing, denote that month as missing.

    4) For each station in the state, run the ANACOVA model with monthly 
    effects and year as the covariate:
       proc glm data=whatever ;
         class month ;
         model precip = year month / solution ;

    5) Assemble these results in a SAS dataset (use ODS) and keep
    only the slopes (and maybe t-statistics) of the covariate effects
    (year coefficient).

    6) Merge these back with the station information to get lat/lon.
   
    7) Construct some coding of the slopes and/or t-statistics so we
    can discern whether the slopes are big or small, positive or 
    negative.

    8) Using your code from (7) as a plotting symbol, plot these
    effects by latitude and longitude.  (The shape of the plot should
    resemble the shape of the state.)

    The information on the US Historical Climatology Network can be
    found at their site:
         http://cdiac.ornl.gov/epubs/ndp/ushcn/ushcn.html

    *) If you want to do a fancier statistical model with the daily
    data, then fit a time trend with at least three harmonics of a 
    yearly (365.25 day) cycle instead of the ANACOVA model.  If you 
    don't understand what I'm talking about, then ignore this point.