Met Office Hadley Centre observations datasets
Home > HadSST4

Diagrams from the paper

The following diagrams are updates from the paper (where appropriate).

Figure 1

Figure 1: numbers of observations passing QC (a, c, e) and super-observations (b, d, f, see Section 3 for the definition) per month for the globe (black), Southern Hemisphere (orange) and Northern Hemisphere (blue) for (a,b) 1850-1880, (c, d) 1880-2000 and (e, f) 2000-2018. Note the very different scales for the y-axes.

Figure 2

Figure 2: Fractional contribution of different SST observation methods, 1915-2018, to (a) the Global average, (b) the Southern Hemisphere average and (c) the Northern Hemisphere average. The brown/orange/tan areas indicate ship observations as labeled in panel (a) and the blue areas indicate buoy observations. The pale lilac area represents unknown measurement method (assumed to be from ships). These are the initial assignments (Section 2.2.1) and are not the assignments finally used to calculate the adjustments.

Figure 3

Figure 3: Example fields from the gridding procedure for the 5° by 5° by pentad grid boxes: (a) SST anomalies (°C) for June 2003 relative to the 1961-1990 average; (b) number of observations contributing to each grid-box average; (c) number of super observations in each grid box; (d) fractional contribution to the grid-box average by ERI measurements; (e) fractional contribution to the grid-box average by drifting buoys (drifters); (f) fractional contribution to the grid-box average by moored buoys (moorings); (g) estimated uncertainty associated with uncorrelated errors; (h) estimated uncertainty (°C) associated with under sampling; and (i) estimated uncertainty (°C) associated with micro-bias errors.

Figure 5

Figure 5: Annual global averages of the 200 realizations of the bucket corrections (a) R06-style corrections (b) SR02-style corrections, (c) combined R06 and SR02 corrections. Blue lines in (a)-(c) show ensemble members with a linear transition from wooden to canvas buckets and orange lines show ensemble members generated assuming a step change in the fraction of wooden and canvas buckets. (d) Estimated seasonal cycle of insulated bucket biases.

Figure 6

Figure 6: (left column) Estimated seasonal-average ERI biases (orange) and bucket biases (blue) (°C) 1940-2018 for (a) the globe, (c) the southern hemisphere and (e) the northern hemisphere. (right column) Smoothed estimated monthly ERI biases (°C) for (b) the globe, (d) the southern hemisphere and (f) the northern hemisphere. The smoothed estimate is shown in orange with the seasonal-averages from the left column shown in grey.

Figure 7

Figure 7: Smoothed time series of the estimated fraction, fcorrect, of measurements labeled as buckets that were correctly identified as buckets for (a) rejected start and end dates and (b) accepted start and end dates. (c) accepted (blue) and rejected (red) start and end date combinations. (d) The black line is the annual average inferred fraction of correct bucket assignments which is the average of the unsmoothed blue lines in (b). The red lines indicate the mean and ranges used to draw samples. The mean values are held constant at 0.5 between 1945 and 1952 and at 0.95 after 1978. The blue lines show 10 samples of the full 200 member ensemble from 1945 on. Before 1945 the mean is set to 0.5, but the uncertainty is larger.

Figure 8

Figure 8: (a) Estimated monthly bias (°C, orange) in global average SST 1850-2018 for the full gridded dataset, including ships and buoys and (b) estimated bias in global average SST anomaly (°C, blue) relative to a 1961-1990 period, again for all data. (c and d) as for (a and b) except for the Southern Hemisphere. (e and f) as for (a and b) except for the Northern Hemisphere. The black line is the median bias and the shaded area represents the 95% range of the estimated biases.

Figure 9

Figure 9: Monthly global average SST anomalies (°C) 1850-2018 (a) relative to the unadjusted 1961-1990 climatology and (b) relative to the bias-adjusted 1961-1990 climatology. (c and d) as for (a and b) except for the Southern Hemisphere. (e and f) as for (a and b) except for the Northern Hemisphere. The grey line shows the unadjusted data, the black line is the median of the adjusted data. The blue and orange shading represents the 95% range of the ensemble.

Figure 10

Figure 10: (a) Annual global average unadjusted SST anomalies 1940-2014 (°C relative to unadjusted 1961-1990 climatology) for collocated bucket (blue) and ERI (orange) measurements. The solid line is the best estimate and the shaded area is the 95% uncertainty range (accounting for measurement and sampling errors). (b) Adjusted anomalies with expanded uncertainty range including bias adjustment uncertainty. The dotted line indicates the best estimate of the unadjusted series from (a). (c) and (d) as for (a) and (b) but for the Southern Hemisphere. (e) and (f) as for (a) and (b) but for the Northern Hemisphere.

Figure 11

Figure 11: Comparisons between sea-surface temperature data sets from different sources 1990- July 2018, different comparison data sets cover different periods. (a) Global average SST anomalies (°C), relative to the 1961-1990 HadSST2 climatology, from Argo floats (purple) and HadSST.4.0.0.0 (black solid line and grey shading) each reduced to their common coverage. The unadjusted SST data are shown as a black dotted line. (b) Indicator of the number of 5° grid boxes in HadSST.4.0.0.0 (pale grey) and in the Argo data set (purple) Data are plotted only for the overlap. (c) as for (a), but with ARC (in red) substituted for Argo. (d) as for (b) but with ARC substituted for Argo. (e) as for (a) but with buoys (in blue) substituted for Argo. (f) as for (b) but with buoys substituted for Argo.

Figure 12

Figure 12: Average SST difference (°C) between HadSST.4.0.0.0 and the three instrumentally homogeneous data sets (a) buoys 1995-2018, (b) ARC 1995-2012, and (c) Argo 2000-2017. (d) Shows the difference between HadSST.4.0.0.0 and the unadjusted gridded SSTs, 1995-2018.

Figure 13

Figure 13: (a) Collocated global annual average NMAT anomalies (°C) 1900-2010 offset by 0.15°C (blue, relative to 1961-1990) and global annual average SST anomalies from HadSST.4.0.0.0 (black is central estimate and grey shading indicates 95% uncertainty range). (b) Offset NMAT anomalies minus SST anomalies with combined 95% uncertainty range (taking into account the bias errors from the HadSST.4.0.0.0 ensemble, and measurement and sampling errors in the SST). The dashed line indicates zero difference. (c) Collocated global annual average offset NMAT anomalies (blue), global annual average near-surface water temperature from HadIOD excluding Argo (orange) and SST (black is central estimate and grey shading again indicates 95% uncertainty range). (d) Difference between HadIOD and HadSST.4.0.0.0 (orange) and NMAT and HadSST.4.0.0.0 (blue). The shaded area indicates the 95% uncertainty range.

Figure 14

Figure 14: (a and b) Global, (c and d) Southern Hemisphere and (e and f) Northern Hemisphere annual average SST anomalies (°C) 1850-2018 relative to 1961-1990 for HadSST.4.0.0.0 (black line is the median in the left column and the grey shading in the right column represents the 95% uncertainty range) and HadSST.3.1.1.0 (blue line is the median and the blue shading represents the 95% uncertainty range). Uncertainty estimates combine the bias-adjustment uncertainties from the ensemble with measurement and sampling uncertainties.

Figure 15

Figure 15: (a) Global average SST anomaly 1850-2012 (°C relative to 1961-1990) series from HadSST.4.0.0.0 (black), ERSSTv5 (blue, thick line is operational version and thin-thick dashed lines are ensemble range from the 1000-member ERSSTv4 ensemble), COBE-SST-2 (orange), HadSST.3.1.1.0 (green) and unadjusted SSTs (red). All data sets are averaged on to a 5° grid and reduced to HadSST.4.0.0.0 coverage before comparisons. (b) Global-average difference for each data set from HadSST.4.0.0.0. The grey shading shows the 95% uncertainty range including effects from measurement, sampling and bias-adjustment errors. The bias-adjustment uncertainty range is shown in darker grey. (c) and (d) as for (a) and (b) but for the Southern Hemisphere. (e) and (f) as for (a) and (b) but for the Northern Hemisphere.

Figure 16

Figure 16: Global and regional average SST anomaly trends to 2012. Median trends from HadSST.4.0.0.0 are indicated by a black horizontal line and the grey shading indicates median and 95% uncertainty range including effects from measurement, sampling and bias-adjustment errors. The bias-adjustment uncertainty range is shown in darker grey; ERSSTv5 (blue), the lozenge is the operational version and the vertical line is the 95% ensemble range from the 1000-member ERSSTv4 ensemble); COBE-SST-2 (orange); HadSST.3.1.1.0 (green); and unadjusted SSTs (red).

Figure 17

Figure 17: (a) Standard deviation of SST anomalies (°C), (b) Zonal length scale Lx (km) and (c) meridional length scale Ly (km) used in the interpolation scheme

Figure 18

Figure 18: Example of the inputs and outputs of the interpolation for July 2003. (a) gridded SST anomalies (°C) from ships. (b) gridded SST anomalies from buoys. (c) interpolated SST anomalies. (d) estimated ERI biases (°C). Note that the ERI data are a subset of the ship data so the coverage is not identical.

Figure 19

Figure 19: Tests of effectiveness of the reconstruction method from 2000-2014. Blue lines show results for the 19th century coverage tests, red lines show the results of the missing-at-random tests and the purple lines show the test where data were removed at selected locations of Argo observations. (a) probability distribution for chi-squared statistics calculated from the withheld data. (b) average of histograms from all scaled, withheld observations in the 19th century coverage test (blue line) compared to the samples drawn from the posterior of the reconstruction (black line and grey shading). (c) as for (b) but for the missing-at-random test (red). (d) as for (b) but for the test where data were removed at the locations of Argo observations (purple). In (d) the dashed purple line shows the offset that occurs when doing the reconstruction based on unadjusted data and the thin blue line shows the scaled residuals of the adjusted SSTs at the locations of the Argo observations.

Figure 20

Figure 20: Density map of estimated biases for ships from the HadSST.4.0.0.0 analysis (where the estimated uncertainty on the bias was less than 0.25°C) and from the IQUAM analysis between 2000 and 2012. The blue lines show the regression of the IQUAM estimate on HadSST.4.0.0.0 and vice versa. The red line is the y=x line offset by 0.19°C. A small white cross marks [0,0].

Figure 21

Figure 21: Tests of bias estimation using synthetic data. (a) histograms of differences between prescribed and estimated characteristic biases for each month for each measurement type in the synthetic data set including buckets (blue), ERI (orange) and hull sensors (hot pink). (b) time series of difference between assigned and estimated biases for bucket measurements (blue) and the 95% uncertainty range (grey lines). (c) as for (b) but for ERI measurements in orange. (d) as for (b) but for hull sensors in hot pink.

Figure 22

Figure 22: (a) Density map of prescribed vs estimated micro-biases for all ships in the synthetic data set. Darker colours indicate higher densities. (b) as for (a) except it shows only those ships for which the uncertainty in the estimated micro-bias is less than 0.5°C. (c) as for (b) but with an uncertainty less than 0.15°C. (d) as for (b) but for uncertainties less than 0.05°C. The black diagonal line is y=x.

Figure 23

Figure 23: Histograms for trend estimates in global average temperatures (°C/decade) for different periods (indicated in the individual plot titles) and different data sets. HadSST.4.0.0.0 is shown in grey (generated from the 200-member ensemble of bias adjustments only) and black (generated from the 200-member ensemble of bias adjustments combined with samples from the measurement and sampling uncertainties). The ERSSTv4 ensemble is shown in blue, with the operational ERSSTv5 point shown as a blue lozenge. COBE-SST-2 is shown in orange. The unadjusted data are shown in red. In addition, the trends from the IPCC AR5 SST data sets are shown as green lozenges numbered as follows: 1 is HadSST3, 2 is HadSST2, 3 is HadISST1.1, 4 is COBE SST, and 5 is ERSSTv3. They do not appear in panels (e) and (f) as these periods were not considered in Chapter 2.