Adapting the QC to account for the June 2021 North American heatwave (part 1)

As we noted on the previous post, the extreme temperatures during June 2021 in western Canada and USA were being erroneously flagged by some of the HadISD QC tests.  This is the first in likely a number of posts as we delve deeper into the causes and resolutions.

Climatological Outlier Check

We go back to the station which we showed in the previous post, 711130-99999 (Agassiz, BC).  The plot we showed was one of the raw diagnostic outputs from HadISD which we ran to see what was going on.

Firstly, while implementing changes for this test, we noticed that the plot was incomplete. It does correctly show the distribution from which the flagging thresholds were calculated as well as the highlighted flags in red.  

However, for this test, the thresholds are determined from the data themselves, using the distribution of the anomalised observations.  So that the addition of each month during the year does not affect the thresholds set by any test, the thresholds are calculated from a distribution using the complete years only. In this case, that's all data from all Junes up to the end of 2020.  What was shown on that plot was the distribution (black) of all the observations from complete years only, and the fitted Gaussian (blue).  The red values were the observations that were flagged, which were from data in 2021 only, but missing were data from June 2021 which were not flagged.  The updated plot is below (Fig. 1), where the grey histogram includes all data from June 2021.  This shows that there are other observations in June 2021 which are warmer than the average of previous years.  Some are even warmer than all previous years, but not sufficiently so to be flagged by this test.

[As a reminder, this test fits a Gaussian to the histogram, and then uses this to determine a threshold (from where the fit crosses y=0.1).  Observations which are further from the peak than the threshold are treated in two ways.  If they are separated by an empty bin from the main distribution, then they are flagged. However, if they are "attached" (no empty bin) then they are tentatively flagged (and could be re-instated by the buddy check).]

Fig. 1: the distribution of scaled anomalies for June from Agassiz (711130-99999), with the flagged ones highlighted in red.  Distribution from all years before 2021 is in black, and from all years including 2021 in grey. Note the logarithmic y-axis.

This updated plot highlighted something we hadn't realised when adapting this test to work for the monthly updates.  The thresholds are set on the complete-year data (up to 2020), and because these data all fall into a single distribution, this test identifies any observations further than this value as bad and flags them (highlighted in red).  However, when including the 2021 data in the monthly update, there isn't an empty bin in the distribution.  We note that had some observations in June in earlier (complete) years been very hot or very cold, they would have been correctly flagged by the test.

So, the first thing we have done is to rectifiy this, and ensure for cases like this, rather than flags being set, only tentative flags are set (Fig. 2), as these values are part of a contiguous distribution rather than being separate from it.  In the case of monthly updates, the threshold for flagging (requiring that empty bin) is re-estimated, and shown by the purple line in Fig. 2. As in the case of complete years, any observation in the final year further from the peak of the distribution is only tentatively flagged as there is no empty bin (pink).

Fig. 2: Same as Fig. 1, but with the data from June 2021 now being correctly marked as tentatively flagged.  Red and Orange vertical lines are derived from the distribution from complete years only (black) and fit (blue).  The Purple and Pink are derived when including the final month.

The other thing to address was to allow for a skew in the Gaussian fit as it is clear that a symmetric function is not the best fit to these data, which is shown in Fig. 3.  This now reduces the number of observations on which a tentative flag is set to single figures.  However, at the low temperature end, the threshold for the tentative flag has reduced, so should Agassiz get a very cold June, then it's possible some values may get tentative flags set instead

Fig. 3: Same as Fig. 2, but now using a skewed-Gaussian ditribution for the fit. 

The automated quality control works on data from meteorogical stations from around the world.  For the case of Agassiz in Canada, we could be reasonably confident that all the data from June 2021 has been included in this update to HadISD, and therefore we could not bother with the "complete" versus "in progress" year distinction.  However, for other locations, we do get data coming through in earlier months than the most recent one (e.g. data filling in during January through to May for the release that included June).  In that case it is possible (though maybe unlikely) that thresholds for this test in earlier months could change from monthly release to monthly release, resulting in values being flagged or unflagged in different releases.  Our approach is more stable during the monthly updates, and so we keep this distinction.

At the end of the calendar year, we run the QC on the data for the final data release of that HadISD version.  For this release, it is on a complete year, so for that release only (the ones processed in January each year), then this distinction isn't made.  Therefore all the June data will go into the distribution from which the thresholds are determined.  The original form of the test would have received a contiguous distribution for this station, and so only set tentative flags.  However, but updating it to use the skew-Gaussian, in fact no observations are flagged (Fig 4.).

Fig 4. Same as Fig 3. but for the version of the test as would be run for the update at the end of a calendar year.

We will continue checking other QC tests as well as run further diagnostics before these changes are implemented in the HadISD QC suite, with a version number increment to reflect the changes.  It is likely that these will not be available in time for the release in August 2021 (including data up to the end of July).

The June 2021 North American Heatwave and v3.1.2.202106p

We have just run the automated quality control (QC) for the latest monthly update, which includes data from June 2021.  I'm sure you are all aware of the severe heat wave which affected the western part of North America in the last part of that month.  British Columbia (Canada), Oregon and Washington (USA) experienced a number of days of exceptionally high temperatures, well over 40C in some cases.

Records measured by stations did not only fall, they were smashed, with new values set that were up to 5C higher than the previous records. A report by the World Weather Attribution project indicates that this event was virtually impossible without human induced climate change. Given that this event was so much warmer than anything experienced in this region in the past, we thought to check how the automated QC handled these exceptional values.

Any QC procedure will always result in retaining some bad values (false negatives), and also erroneously removing some good ones (false positives).  It is impotant, however, to minimise these as best possible, and do "least harm".  To this effect, observations that have been flagged by the QC are removed from the main data stream, but are available in a separate data field in the netCDF files, should any user wish to re-insert them into the time series.

HadISD QC

The town of Lytton (BC) recorded the highest temperatures during this event, but that station does not form part of the HadISD.  We had a look to find nearby stations, and show these in Figure 1 for Agassiz, which is south of Lytton and further down the Fraser River, towards Vancouver.

Figure 1. Temperature timeseries for Agassiz (BC, 711130-99999) for (a) all of 2021 and (b) the latter half of June 2021.  Observations are in black with any flagged by the climatological QC test in red.

As you can clearly see, the highest values at the peak of the heatwave on the 27-29th June have all been removed, in this case by the Climatological Outlier check.  At some level, this is unsurprising, given that the temperatures experienced surpassed anything in the previous record.  Climatologically speaking they are exceptional values, and so without any other information to go on, could be dubious.  

The Current QC

Of course we know that these are likely to be valid observations, and as HadISD has been designed to retain true extremes, some adjustments to the QC algorithms are necessary.   Firstly, let's have a look at how the current test is identifying and flagging these values. For full details see the HadISD paper (Dunn et al, 2012).

The Climatological Outlier check works on a monthly basis, and calculates climatological values for each hour of the day for each month using the winsorized observations (Winsorizing is a process where all values exceeding a certain threshold [5% & 95% in this case] are replaced by these threshold values).  Using these 24 climatological values, anomalies are calculated, and then scaled using their inter-quartile range.

We have included a way to account for some of the effects of a shifting climate using a low-pass filter.  However, this is only applied to complete years of data, and so on our monthly updates has so far not been included. The resulting distribution is fitted with a Gaussian, and we use where this fitted Gaussian crosses the y=0.1 line to set our threshold, rounded up to the next whole degree.

Figure 2: the distribution of scaled anomalies for June from Agassiz (711130-99999), with the flagged ones highlighted in red.  Note the logarithmic y-axis.

The test operates two levels of flagging, depending whether there is an empty bin between those further from the centre than the threshold values.  If there is a gap, then these are flagged, as shown in Figure 2.  If there isn't an empty bin and the are bins part of a contiguous distribution but are further from the mean than the threshold, then these are "tentatively" flagged (see Figure 10 in the HadISD paper).  When running the neighbour checks, these tentatively flags can be removed if sufficient neighbours indate these are reasonable.

In the case of Agassiz, the observations were so extreme, that this test has flagged them without the option of the neighbour check undoing this (Figure 2).

Amending the QC

There are a number of options as to what we could do to improve the actions of this automated QC.  However, the important thing is to make sure that whatever we implement, there are as few knock-on effects in other regions and flags as possible.  The intention being that we improve this test in a robust, responsible way.

A number of options have so far come to mind, including:

• Amend the low-pass filter to include data from the year in progress.

• Amend the fitting function from a pure Gaussian, to one which allows skew or even kurtosis.  As seen in Figure 2, the distribution has a high tail above the fitted Gaussian, and accounting for this will affect the threshold used.  This approach is already used in a different check in the HadISD QC.
  

• Use a rolling range to determine the years contributing to the climatologies used when creating anomalies, so values from 1931 are not contributing to 2021.

• Amend the neighbour check so that spatially coherent anomalies result in flags being unset from a greater subset of QC tests.

All of these approaches will need to be tested with care to ensure that any updates do not result in detrimental performance of the QC suite elsewhere in time or space.

We will release this version of HadISD (v3.1.2.202106p) with a note that observations from this event have been erroneously flagged.  As this is a preliminary version of HadISD, this is reasonable, and gives us time to implement a solution in "slow time".  Watch this space for an update.

References:

Dunn, R. J. H., Willett, K. M., Thorne, P. W., Woolley, E. V., Durre, I., Dai, A., Parker, D. E., and Vose, R. S.: HadISD: a quality-controlled global synoptic report database for selected variables at long-term stations from 1973–2011, Clim. Past, 8, 1649–1679, https://doi.org/10.5194/cp-8-1649-2012, 2012  

[Edited 9-Jul-2021 10.50BST to add option of amending the neighbour check]