***ISSUE WITH 1.4.0.2022f MISSING OBSERVATIONS NOW CORRECTED - PLEASE SEE: HadISDH.marine.1.4.1.2022f Update***
1) How does the new version compare?
v1.4.0.2022f is virtually identical to v1.3.0.2021f over the 1973-2014 period for specific humidity (q) and relative humidity (RH). However, as can be seen clearly in Figs. 1 and 2 panels d to g the versions differ increasingly from 2015 onwards. The variability appears larger in the last few years, although this has not been tested. This divergence in the last few years likely explains the tiny consistent offset across the 1973-2014 because the years 2015 to 2020 will influence the 30 year (1991-2020) climatology period over which the anomalies are referenced.
A newer version of ICOADS has been used to provide data from 2015 onwards which includes 20% more observations overall. However, this has not resulted in any notable increase to the spatial coverage of HadISDH.marine.v1.4.0.2022f which is based only on ship observations of dewpoint temperature. In fact, in terms of gridboxes with sufficient presence over the climatology period, the new version v1.4.0.2022f actually results in 23 fewer gridboxes and only 1 additional gridbox (Figs. 1 and 2 panel a). However, these figures mask a larger loss of coverage in the last few years, where the regional average timeseries diverge. This is because the loss in coverage only affects the last 6 years of the climatology period. Comparisons of monthly gridded fields in HadISDH.marine v1.4.0.2022f compared to v1.3.0.2021f over the last two years show many fewer gridboxes over the north Pacific in particular, although there is an increase in observations over the north Atlantic and the southeastern Atlantic. It is likely this loss of coverage that is causing the divergence in regional timeseries.
ICOADS3.0.1 is no longer updated and suffered from a drop in drifting buoy observations as they transitioned over to being reported in BUFR format rather than TAC (Liu et al., 2022). ICOADS.3.0.1 did not use the BUFR reports. ICOADS.3.0.2 was designed to use both TAC and BUFR, generally prioritising BUFR reports when merging the many duplicate observations. ICOADS.3.0.2 brings a considerable increase in the number of surface drifting buoys. However, for ship obs, the ICOADS.3.0.2 Final files used for HadISDH.marine, after HadISDH.marine processing, provide ~100000 fewer observations per month than 3.0.1 and we do not yet understand why this is. HadISDH.marine only pulls through observations with a dew point temperature present so would miss any obs reporting only RH or wet bulb temperature. Further digging shows several observations that were present in HadISDH.marine from 3.0.1 are now missing in 3.0.2 - these can be traced using their Unique Identifier (UID). Presumably these are still available but ICOADS.3.0.2 uses the BUFR versions rather than the TAC versions.
The overall story of increasing specific humidity is concurrent in both versions but the long-term trends in v1.4.0.2022f that have an extra year of data and a change in coverage (see below) are 0.0 to 0.02 g kg -1 decade -1 smaller (Figure 1 panels d to g). Across the globe (Figure 1 panel a) the vast majority of gridboxes show trends in the same direction (97% of gridboxes - Figure 1 panel b), with 91.5% of these both showing increasing q (Figure 1 panel b). Trends appear to be slightly larger for v1.4.0.2022f compared to 1.3.0.2021f in the northern mid- to high latitudes. Trends around the west and east coast of South America are opposite in direction. In total, 3% of gridboxes have trends in opposite directions within the two versions.
The overall story of decreasing relative humidity (except over the tropics) is also still valid in both versions, with long-term negative RH trends in v1.4.0.2022f being -0.05 to -0.1 %rh decade -1 more negative than in v1.3.0.2021f (Figure 2 panels d to g). Across the globe (Figure 2 panel a) there is more of a mix of whether gridbox trends are of the same or different direction although 87% of gridboxes (Figure 1 panel b) have trends in the same direction. Gridboxes with differing directions appear more common over the extratropics and tropics. Overall, 57% of gridboxes agree on negative trends and 30% agree on positive RH trends. Note that the decrease in RH over oceans remains a very uncertain conclusion as it does not reconcile with models or theory. Conceivably, there may be regions where relatively warmer and dry air may be advected from the land which could locally lower RH, or regions where increased wind speed increases mixing which might have the effect of lowering RH. However, these are unlikely to occur on the scale to effect global trends. Ultimately, spatial coverage of HadISDH.marine is very limited, so the decreasing RH cannot really be taken as a global trend.
This growing divergence is of considerable concern for climate monitoring and requires further investigation, as does the loss of coverage. Clearly, HadISDH.marine can no longer be considered as globally representative.
2) What's New?
We use ICOADS3.0.0 (1973 to 2014) and ICOADS3.0.2 (2015 onwards) as the basis for HadISDH.marine.v1.4.0.2022f. In previous years we have used ICOADS3.0.1 (2015 onwards) but this is now superceded by ICOADS3.0.2.
Although this update presents a 20% increase in coverage (~1million extra observations) of essential climate variables (see https://icoads.noaa.gov/), for the surface humidity processed as part of HadISDH.marine there is a loss in coverage over the northern Pacific and a few other areas. The main gain in observations appears to be over the northeast Atlantic, where observations are already fairly dense. It is not clear why some observations are no longer passing through the processing chain.
This means that HadISDH.marine.v1.4.0.2022f deviates from HadISDH.marine.v1.3.0.2021f from 2015 onwards. This is particularly noticable from 2020 onwards. Coverage is now so limited that users should be aware that HadISDH.marine cannot be representative of the true global average.
A second change is the formula for calculating wet bulb temperature from dewpoint temperature and marine air temperature. Errors were found when the air temperature was very high but dewpoint temperature very low, resulting in spuriously high wet bulb temperatures. The Stull (2011) formula is now used as this has a far wider range of applicability. See the blogpost for details.
This mostly affects the wet bulb temperature fields although differences are very small, and far less than 1 degree for the most part. Differences are larger over warm, dry air conditions, which are less of a problem over ocean. The calculated wet bulb temperature is used to decide whether to calculate vapour pressure with respect to ice or water and so this new formula can lead to very small changes in vapour pressure and variables that use vapour pressure in their calculation (specific humidity and relative humidity). Stull (2011) tends to give higher wet bulb temperatures overall which will result in fewer uses of the calculations with respect to ice and therefore fractionally higher vapour pressures and related values. As it is consistently used across the time period it should not impact long-term trends in anomalies.
3) Summary of changes by level of technicality.
MAJOR CHANGES (X):
- none
MINOR CHANGES (Y):
- Change of source dataset from 2015 onwards from ICOADS3.0.1 to ICOADS3.0.2 which although increasing the number of initial observations by ~20% actually reduces coverage in some regions (especially north Pacific).
- Change of wet bulb temperature formula to Stull (2011) - see blogpost.
BUG FIXES AND HISTORICAL DATA UPDATES (Z):
- Change in spatial coverage and regional average time series from 2015 onwards caused by new data source ICOADS3.0.2.
No comments:
Post a Comment
Note: only a member of this blog may post a comment.