Exploring the Marine Data Literature to Understand Code Changes

Its time to start working on the marine humidity data. Marine humidity data comes from ships, moored buoys, and marine platforms. There may be sensors on some drifting buoys but its not mandatory - quality is likely to be an issue due to contamination of the humidity sensor with sea water (WMO No. 8, Part II, Ch. 4). Humidity measurement is more sensitive to contamination with sea water than dry bulb temperature. It is also more sensitive to moist biases in low wind speeds/poor instrument siting because the sensor requires adequate ventilation. However, humidity is less sensitive to heating errors as long as the heating effects both the dry bulb and temperature sensor simultaneously (Berry & Kent, 2005, IJC).

The majority of marine meteorological data collected since early 1900s has been recorded in a common format described by WMO FM codes. However, these codes have changed over time, with different ships adopting new codes at different times such that there were several codes in use simultaneously. As documented by the WMO's MMOP (History of SHIP code), between 1972 and 1982 there were five codes in use: FM 21.E SHIP (sea station report from manned or automatic), FM 22.E SHIP (sea station report from manned or automatic in abbreviated form), FM 23.E  SHRED (sea station report from manned or automatic in reduced form), FM 24.E SHIP (sea station permanently automatic) and possibly FM 26.B SPESH (special report). In January 1978 these are amended to FM 21-V SHIP, FM 22-V SHIP, FM 23-V SHRED and FM 24-V SHIP. In January 1982 a major change was made in that FM 21-V SHIP, FM 22-V SHIP, FM 23-V SHRED and FM 24-V SHIP were deleted and the FM 13-VII introduced.

It looks like dew point temperature was reported to 0.1 deg C precision (or at least there was the ability to do this within the forms) prior to 1972. The March 1966 recommendation Rec. 5, 9, 13, 24 and 25 CSM-IV (Mar. 1966) discusses FM 21.C, FM 22.C and FM 23.C and TdTdTd which suggests there are three characters allocated to dew point temperature. Negative values are identified by the preceding character s_n which can be a 0 for +ve or a 1 for -ve (see ANNEX XX (Annex to Rec.22) page 22) (was an 'X' over-punched over the first character or a preceding character of blank or 'M' for minus).

One of the previous explanations for the marine moist bias prior to 1982 was that reports for dew point temperature could only be stored as whole numbers until 1982. This does not appear to be the case from the documents listed above. However, there may still have been a common practice of only reporting dew point temperature to whole numbers until the changes brought in in 1982. Indeed, there is evidence across some ship decks that many more reports had decimal precision after 1982 than prior to it.

I have explored the effects of rounding/truncating/rounding up dew point temperature to a whole number or a half number and how the differences compare across the range of decimal places (.0 to .9) for positive and negative numbers to see how a bias may be introduced in a rounded period compared to a non-rounded period (e.g. pre-1982 compared to post-1982).  

In a normal rounding situation (.0 to .4 round down, .5 to .9 round up) compared to a non-rounded situation, on average, dew point temperature (and therefore q and RH) would be higher/lower for dew point temperatures above/below zero. Specifically, for ten dew point temperatures from 2.0, 2.1, 2.1...2.9 the average bias for rounded values would be 0.05 degrees C compared to the actual values. For the negative number equivalent this would be -0.05 degrees C. The maximum bias for any individual value would be +/- 0.5 degrees C. So if a distribution of dew point temperatures were symmetrical about zero, there would be no overall bias. Away from the poles, temperatures are generally above zero more than they are below zero (global average T is 14.5!)  so in a simple rounding case where some dew point temperatures prior to 1982 were rounded and post-1982 were not, we could expect a moist bias pre-1982 compared to post-1982. This shouldn't really be more than 0.5 deg C for any one individual observation. Table 1 explores this bias for a range of temperatures at 50 %rh and 80 %rh (at standard pressure). At 50%rh a 0.5 deg C increase in dewpoint results in 1.5 to 2 %rh increase and a 0.1 to 0.5 g/kg increase in q (~3.5% in percentage terms for both variables). At 80 %rh a 0.5 degree increase in dew point results in  a 2.5 to 3 %rh increase and a 0.1 to 0.65 g/kg increase in q (~3.5% in percentage terms for both variables). This shows that the effect is larger in higher humidity/cooler temperatures. Conceivably, if it is widespread and commonplace it could be noticeable in RH and q depending on the signal to noise ratio. Rounding to the nearest 0.5 has no overall bias when averaged across the decimal places - the maximum bias for an individual observation is +/- 0.2 degrees C.


Table 1 Changes in RH and q at a range of temperatures for a 0.5 degree C increase (away from zero - blue cells are negative dew points) in dew point temperature. All calculations are done at standard pressure (1013mb).

T	RH	q	Td	Td+0.5	new RH	new q	RH diff	q diff	RH diff (%)	q diff (5)
0.0	50	1.88	-9.18	-9.68	48.07	1.8	-1.93	-0.08	-3.86	-4.26
5.0	50	2.68	-4.55	-5.05	48.15	2.58	-1.85	-0.10	-3.70	-3.73
14.5	50	5.08	4.21	4.71	51.77	5.26	1.77	0.18	3.54	3.54
20.0	50	7.20	9.27	9.77	51.70	7.45	1.70	0.25	3.40	3.47
30.0	50	13.13	18.45	18.95	51.60	13.55	1.60	0.42	3.20	3.20
0.0	80	3.00	-3.03	-3.53	77.08	2.89	-2.92	-0.11	-3.65	-3.67
5.0	80	4.29	1.84	2.34	82.90	4.45	2.90	0.16	3.63	3.73
14.5	80	8.14	11.1	11.6	82.72	8.42	2.72	0.28	3.40	3.44
20.0	80	11.56	16.45	16.95	82.60	11.93	2.60	0.37	3.25	3.20
30.0	80	21.10	26.17	26.67	82.40	21.74	2.40	0.64	3.00	3.03

The effects of truncating or rounding up are much larger (maximum difference +/- 0.9 degrees C and average across the decimal places is +/- 0.5 degrees C). However, it is reasonably unlikely (although not impossible) that such practices were widespread or commonplace. 

If the dry bulb temperature has also been rounded then the effect on humidity could be both reduced or enhanced at dew point temperatures both above and below zero. This effect should average out over a large distribution because sometimes the dry bulb temperature will be rounded up and sometimes down. Therefore, if this practice was in place until 1982 I don't think that it would make much difference in addition to the dew point temperature rounding. Playing with this in excel suggests that having both the dry bulb and dew point temperature rounded results in no overall bias although any individual value could have a maximum bias of +/- 0.9 degrees C in terms of the dew point depression which would make a difference to the individual calculated humidity.

There is a possibility when both the dry bulb and dew point temperature are rounded that 100%rh occurs more frequently, although this really depends on the distribution of values. When only the dew point temperature is rounded, what should have been 100%rh could now be less than 100%rh because the dew point has been rounded down, or removed by QC because the rounded up dew point now exceeds the decimal dry bulb temperature. We could look at the proportion of 100% RH cases before and after 1982 by percentage and look at whether they have T and or Td in whole degrees more pre-1982. There would be times where the humidity is decreased by rounding (T rounded up and Td rounded down, both rounded up or down but from differences smaller than a degree). 

In summary, the effect of rounding values should be small on monthly, gridbox and regional averages (+/- ~0.05 degrees C?) but for any individual values it could be up to +/- 0.9 degrees C when both elements are rounded or +/- 0.5 degrees C when only the dew point temperature is rounded.

Another explanation was that there was a practice of reporting the rounded dew point temperature and then storing the decimal precision as supplementary info elsewhere in the report. When these reports came to be reformatted into a common format the decimal precision dew point temperature could be reconstructed from its two parts, assuming that if the decimal value was 0.5 or greater then the reported value had been rounded up (or down in the negative case). Errors can occur either because the incorrect method was used to make the initial report, or an incorrect method used in the reconstruction - both result in incorrect reconstructed values obviously. I cannot find any evidence of storing the decimal value in the FM codes or the IMMPC or IMMT formats.

An exploration of reconstruction of whole number reports, either using truncation or rounding up to round the initial value and normal rounding to recompute or vice versa, results in a maximum error for any one value of +/- 1 degree C which could expand to +/- 2 degrees if the practice is applied to both dry bulb and dew point temperature. An average across all decimals results in a maximum bias of +/- 0.45 degrees C.

Weather reports, once passed to WMO, were then reformatted into IMMPC
/IMMT format (now IMMA). This history of this is documented by the WMO's MMOP (History of IMMPC/IMMT). IMMPC format was in place prior to January 1982, it was then revised and IMMT format also introduced. The Annex to Rec. 23 document suggests that prior to 1982, humidity was recorded as wet bulb temperature as standard and the dew point could be found later on in the table. This is interesting as the FM formats suggest that dew point was the mandatory humidity variable to report. At that time, most humidity would have been observed as a wet bulb temperature so it would then have to be converted to dew point temperature for the FM report. It would have then been converted to wet bulb temperature for the IMMPC format. Lots of error could be introduced here should the initial dew point temperature reported have been rounded and reported in whole degrees. The first character of the IMMPC format is a single value to indicate what temperature was reported - fahrenheit (2) or celsius (1). The space for dry bulb, wet bulb and dew point temperature allows for decimal precision. If any number is negative then an 'X' is over-punched over the first character. If the wet bulb is in fact an ice bulb then and 'X' is over-punched of the third/final character.

The changes in January 1982 amended the IMMPC and brought in the IMMT which was then identical with the IMMPC for the first 80 characters. The changes are detailed in Rec. 8 CMM-VIII. Now the first character temperature indicator can be 0 to 5. All temperatures should now be in celsius. 0 to 2 indicate IMMPC format for tenths, halves and whole degrees respectively. 3 to 5 indicate IMMT format for tenths, halves and whole degrees respectively. It is not recorded here but in some cases the dry bulb may be in tenths or halves but the dew point or wet bulb in whole degrees - see ICOADS corrections page. This could be a major source of bias if, as it appears, use of decimal precision for dew point temperature was more commonplace post January 1982. There are no more over-punches so the three character dry bulb temperature is now preceded by a sign identifier. What was the wet bulb temperature now has an identifier: 0/1 infers positive/negative dew point temperature; 5/6 infers positive/negative wet bulb temperature; 7 infers an ice bulb. There are then three characters for the decimal precision wet bulb or dew point temperature. This could be a source of error if mistakes are made with the identifier. If the wet bulb temperature is mistaken for the dew point temperature then this would be a moist bias because the wet bulb temperature is higher than the dew point temperature until saturation is reached where they are equal. This first humidity column is intended for which ever value is observed. There is a later field (after 80 characters) for a computed humidity value (dew point or wet bulb) in the same format. I cannot see a field for the decimal place in cases where the value has been rounded.

By 1994 (Annex to Rec. 13) it looks like the first humidity field is constrained to dew point temperature only. The identifier now says whether it is measured (0 to 2 - positive/negative/ice) or computed (5 to 7 - positive/negative/ice).

Summary

Possible causes of error:

• Reading the wet bulb as the dew point would result in a moist bias.

• Reading the rounded dew point temperature when the dry bulb temperature is reported to decimal precision, relative to both not being rounded could lead to a small moist bias - +/- 0.05 on average maximising at +/- 0.5 degrees for any individual value. The number of 100%rh occurrences is likely to be fewer due to rounding down decreasing the humidity or rounding up leading to QC failures when the dew point exceeds the dry bulb temperature.

• Reading the rounded dew point and dry bulb temperature relative to non-rounded values is likely to average out and not create a bias but individual values could be as large as +/- 0.9 degrees different in terms of the dew point depression. There could be many erroneous cases of 100%rh, keeping in values that would otherwise have been removed by the QC for the dew point temperature exceeding the dry bulb temperature.
   
•  Deconstruction and reconstruction of the dew point temperature (and the dry bulb temperature) could lead to considerable errors but no evidence of this practice was found.

Useful websites and documents:

The ICOADS website not only has all digitised marine data available in a common format but also hosts many useful documents of formats, changes, data issues etc.
Main ICOADS webpage: http://icoads.noaa.gov/doc.html
Discussion on marine humidity issues: http://icoads.noaa.gov/corrections.html

A rather random website with a lengthy description of weather codes: http://www.ominous-valve.com/wx_codes.txt

WMO Manual on Codes Part 1:1 Alphanumeric Codes: http://www.wmo.int/pages/prog/www/WMOCodes/WMO306_vI1/VolumeI.1.html
See Section A FM 12-40 but whole document has interesting things about rounding and humidity.