Previous - 6 Satellites Index Next - References
One of the chief difficulties in assessing the uncertainties in SST data sets is the impossibility of tracing individual observations back via an unbroken chain to international measurement standards. The creation of a global array of reference stations or buoys each making simultaneous redundant measurements (such as the triple measurements made at CRN stations, Diamond et al. 2013) of a variety of marine variables could solve some of the problems of SST analysis that have bedeviled the understanding of historical SST change and would provide a gold standard against which the future wider observing system, incorporating observations from ships, buoys, profiling floats and satellites can be assessed. Even without such traceability a climate record could be more easily maintained by stricter adherence to the Global Climate Observing System [GCOS 2003] climate monitoring principles.
In the absence of such a network the estimation of uncertainties has depended heavily on redundancies in measurement systems and in analysis techniques. Full use of the redundancies is now being made in the modern period via comparisons of the many available satellite sources with each other and with in situ sources [O'Carroll et al., 2008; Merchant et al., 2012, Hausfather et al. 2017] and sub-surface data [Gille, 2012, Huang et al. 2018]. Analyses that ingest a variety of data sources can produce bias statistics for each of the inputs [Brasnett, 2008; Xu and Ignatov, 2010, 2014]. Such information can be exploited to assess their relative quality and, as the analyses are pushed further back in time [Roberts-Jones et al., 2012], they will help assess uncertainties through a larger part of the record.
SSTs are physically related to other measurements including surface pressures and winds, salinity, air temperatures, sub-surface temperatures and ocean biology amongst others. Information from SST can be supplemented by analyses based on physical understanding of the climate system. It has already been shown that by combining information from night marine air temperatures with SST it was possible to greatly reduce uncertainties in early 20th and late 19th century SST. Yu et al. [2004] used a joint estimation method to minimize uncertainties in flux estimates based on a range of different variables mostly based on satellite data. Other studies [Tung and Zhou, 2010; Deser et al., 2010] have used physical reasoning based on a host of variables to explore uncertainties in the long-term trends of tropical Pacific SSTs first raised by Vecchi et al. [2008]. It has even been suggested that proxy records such as isotope ratios from corals and ice cores could be used, with appropriate care, to understand uncertainties in the longest-term changes in SST [Anderson et al., 2013]. The most advanced exemplars of physical and statistical synthesis are ocean and coupled reanalyses which will play an increasingly important role in understanding observational uncertainty and long-term climate change.
A key barrier to understanding SST uncertainty is a lack of appropriate metadata. Better information is needed concerning how measurements were made, which method was used to make a particular observation, calibration information, the depths at which observations were made, and even basic information such as the call sign or name of the ship that made a particular observation.
Some of this information can be inferred from data already contained in marine reports. Where reports in ICOADS cannot be associated with a particular ship, either because they have a missing ID, or a generic ID, there is much to be gained by grouping observations to give plausible ship tracks, or voyages (Carella et al. 2017a). By using data association techniques to infer such metadata from the location information and other clues such as how frequently observations were made and which variables were observed, it should be possible to assess systematic and uncorrelated errors on a ship-by-ship basis going back to the start of the record and even infer likely measurement methods based on characteristic variations of the measurements with the meteorological conditions (Carella et al. 2018).
A more systematic approach to the assessment of analysis techniques is needed to elucidate the reasons for the differences between analyses and to assess the verisimilitude of analysis uncertainty estimates. Approaches could include theoretical inter-comparisons of statistical methods, comparisons based on well-defined sets of common input observations, and benchmarks built from datasets (such as model output) where the truth is known a piori. Benchmark tests like those planned by the International Surface Temperature Initiative [Thorne et al. 2011b] provide an objective measure against which analysis techniques can be evaluated. Both analysis techniques and benchmarks will have to be tailored appropriately for the particular problems affecting SST measurements and the latest understanding of measurement uncertainties.
A key weakness of historical SST data sets is the lack of attention paid to evaluating the effects of data biases particularly in the post-1941 records. Further independent estimates of the biases produced need to be undertaken using as diverse a range of means as possible and the robust critique of existing methods must continue. Ideally, these would be complemented by carefully-designed field tests of buckets and other measurement methods.
A community review paper on SST biases (Kent et al. 2017) made a number of recommendations:
Combining new analysis techniques that have been appropriately benchmarked with novel approaches to assessing uncertainty arising from systematic errors, pervasive systematic errors and their adjustments will give new end-to-end analyses that will help to explore the uncertainties in historical SSTs in a more systematic manner.
For long-term historical analyses, there is no substitute for actual observations and relevant metadata. Efforts to identify archives of marine observations and digitize them are ongoing [Brohan et al., 2009; Wilkinson et al., 2011]. Such programs are labor intensive, first in identifying and cataloguing the holdings in archives around the world, then in creating and storing digital images of the paper books and finally in keying the observations. The difficulty of decoding hand written entries in a variety of languages, formats and scripts means that optical character recognition technologies are of limited use. A number of popular crowd-sourcing projects have been started to key information from ships logs that have historical as well meteorological interest. OldWeather.org has keyed data from Royal Navy logs from the First World War [Brohan et al., 2009] and is now working on logs from polar expeditions. Digitization of data also holds the possibility of extending instrumental records further back in time [Brohan et al., 2010]. New observations, with reliable metadata, can be used not only to reduce uncertainty in SST analyses, but also to test the reliability of existing interpolated products and their uncertainties.
The ultimate destination of newly digitized observations is the International Comprehensive Ocean Atmosphere Data Set (ICOADS) [Woodruff et al., 2011]. The ICOADS repository of marine meteorological data has long been the focus of advances in the understanding of marine climatology. It provides a consistent baseline for a wide range of studies, providing a solid basis for traceability and reproducibility. The continued existence, maintenance and improvement of ICOADS are essential to the future understanding of the global climate.
Finally, the work of identifying and quantifying uncertainties will be pointless, if those uncertainties are not used. Uncertainty estimates provided with data sets have sometimes been difficult to use or easy to use inappropriately. As pointed out by Rayner et al. [2009], "more reliable and user-friendly representations of uncertainty should be provided" in order to encourage their widespread and effective use.
Previous - 6 Satellites Index Next - References