Comparing UKMO data with Jim Cumming's Argo data

Time profiles for 2010

Possible juld_qc values for 2010

ukmo_juld_qc	-999 (85,667) and 0 (30,906)
ukmo_jim_juld_qc	-999 (116,573)

Position profiles for 2010

Possible pos_qc values for 2010

ukmo_pos_qc	-999 (85,667), 1 (30,437) and 4 (469)
ukmo_jim_pos_qc	-999 (115,042), 1 (1,428) and 4 (103)

Relationships

• 84,691 times when ukmo_pos_qc = -999 and ukmo_jim_pos_qc = -999
  • When ukmo_pos_qc = -999, ukmo_jim_pos_qc is
    • 1 - 918 times
    • 4 - 58 times
  • When ukmo_jim_pos_qc = -999, ukmo_pos_qc is
    • 1 - 29,888 times
    • 4 - 463 times
  • 504 times when ukmo_pos_qc = 1 and ukmo_jim_pos_qc = 1
    • When ukmo_pos_qc = 1, ukmo_jim_pos_qc is
      • -999 - 29,888 times
      • 4 - 45 times
    • When ukmo_jim_pos_qc = 1, ukmo_pos_qc is
      • -999 - 918 times
      • 4 - 6 times
    • 0 times when ukmo_pos_qc = 4 and ukmo_jim_pos_qc = 4
      • When ukmo_pos_qc = 4, ukmo_jim_pos_qc is
        • -999 - 463 times
        • 1 - 6 times
      • When ukmo_jim_pos_qc = 4, ukmo_pos_qc is
        • -999 - 58 times
        • 1 - 45 times

      Temperature profiles for 2010

      Possible tmp_qc values.

      ukmo_tmp_qc value	-999 (85,667), 1 (30,419) and 4 (487)
      ukmo_jim_tmp_qc	-999 (115,042), 1 (1,428) and 4 (103)

      This data looks to be very similar to the position QC data.

      ETS score for temperature profiles for 2010

      Using raw data, taking 1 to be good and 4 to be bad

      		ukmo
      		Reject	Accept	Total
      d_mode	Reject	50 (0.173%)	195 (0.673%)	245 (0.845%)
      	Accept	381 (1.315%)	28354 (97.840%)	28735 (99.155%)
      	Total	431 (1.487%)	28549 (98.513%)	28980 (100%)

      ETS: 0.0744851167547365
      bias: 0.568445475638051

      Using Jim's data, taking 1 to be good and 4 to be bad

      		ukmo_jim
      		Reject	Accept	Total
      d_mode	Reject	2 (0.142%)	8 (0.567%)	10 (0.709%)
      	Accept	88 (6.237%)	1313 (93.055%)	1401 (99.291%)
      	Total	90 (6.378%)	1321 (93.622%)	1411 (100%)

      ETS: 0.0139905952918226
      bias: 0.111111111111111

      This is a much higher score when using the raw data - still not great though.