Previous - 3.3.3 Estimating Uncertainty in Bias Adjustments Index Next - 3.3.5 Assessing the Efficacy of Bias Adjustments
There are some factors that have not been explicitly considered in estimates of biases. Refinements to the models of pervasive systematic errors will address with factors that are implicitly included in uncorrelated and systematic measurement uncertainties. If it is possible to estimate the bias on a ship-by-ship, or observation-by-observation basis, taking account of the conditions peculiar to that observation, then it might be expected that uncertainties associated with uncorrelated and systematic observational error will decrease.
Both Kennedy et al. [2011c] and Hirahara et al. [2013] make simplifying assumptions about the systematic errors associated with modern insulated buckets. Various bucket designs have been used since the end of the Second World War, which are likely to have different bias characteristics. Physical models could be developed for each type of bucket similar to those used by Folland and Parker [1995], or statistical methods could be used to estimate the biases as was done in Kent and Kaplan [2006]. Carella et al. [2017b] studied heat loss from wooden and canvas buckets in a laboratory setting and suggested that heat loss is closely linked to the wet bulb depression, pointing the way to a simplified statistical approach to estimating bucket biases. Chan and Huybers [2019] showed significant biases between bucket measurements made by ships from different nations and from different ICOADS decks.
Other simplifying assumptions used in all analyses include such things as assuming that changes in the observing system happened linearly. Evidence suggests that changes in measurement method were not always monotonic and sometimes happened abruptly (see Figure 6). Improved metadata or more sophisticated statistical techniques could help assess these uncertainties. In particular, the changeover between canvas and wooden buckets and the speed of ships, are unlikely to be linear as was assumed in Folland and Parker [1995], Rayner et al. [2006], Kennedy et al. [2011c] and Smith and Reynolds [2002]. Huang et al. [2015, 2017] relaxed this assumption. Carella et al. [2017] suggests there was a step change in ship speeds in ICOADS close to the start of the 20th Century.
An uncertainty associated with pervasive systematic biases, which is not explicitly resolved by current analyses, arises when the conditions at the time of the measurement deviate from the climatological values assumed by the bias correction scheme. If, for instance, the air sea temperature difference is larger than that assumed by the Folland and Parker [1995] scheme, then there will be an additional locally-correlated error with a potential long-term component where differences persist for months or years. Likewise conditions vary during the day. Such discrepancies could be assessed by evaluating the systematic error using local conditions. Such information could be taken from reanalyses, or an appropriate bucket model could be explicitly included when SST observations are assimilated into ocean-only and coupled reanalyses.
Previous - 3.3.3 Estimating Uncertainty in Bias Adjustments Index Next - 3.3.5 Assessing the Efficacy of Bias Adjustments