Historical Ocean Surface Temperatures: Adjustment, Characterisation and Evaluation (HOSTACE)

陆地和海洋的表面温度是衡量“全球变暖”的主要指标。海洋表面温度（SST）的测量已经有200多年的历史了，最初是在帆船上进行的，现在是在船只和浮标（漂流和系泊）上进行的。在这段时间里，技术发生了巨大的变化，这引发了一个严重的问题：随着时间的推移，技术的变化是否会误导人们对温度如何变化的看法，从而导致气候如何变化。人们首先测量了装在木桶里的海水样本的温度。水桶现在是用绝缘橡胶做的。现在大多数直接的海温测量都是通过卫星从漂流浮标上发送的。许多其他的测量方法也被使用。不同的方法不会得到完全相同的海表温度值，而且由于全球变暖是一个渐进的变化，这些细微的差异（或“偏差”）可能会扭曲我们对全球变暖时间和幅度的认识。因此，我们必须确保我们了解用于测量海温的不同方法如何影响观测结果。海温的这些偏差多年来一直是一个众所周知的问题，那么为什么我们相信我们可以解决它呢？一个原因是，最近从历史资料中检索到的观察结果越来越多。许多船舶载有天气观测的航海日志已被数字化。这几乎是第二次世界大战前的两倍。另一个原因是地球轨道卫星上的传感器对海温进行了新的、稳定的观测。大多数卫星传感器提供海温模式的详细图像，并调整为漂流浮标海温，以提供合理的精度。但与全球变暖的微妙趋势相比，它们在年复一年和远距离上都不够稳定。通过对一系列特殊传感器的海温测量结果进行重新加工，获得的新的高质量海温测量结果足够准确和稳定。更好的是，它们不依赖于船舶或浮标的海温观测，因此我们可以将它们作为独立的参考点。一个主要的挑战是，不同的测量方法在船上产生的海温偏差是不同的，我们并不总是知道使用了什么方法。但我们确实知道每种方法的偏差是如何随着太阳加热量和风速等因素而变化的。我们将使用每艘船或浮标的这些偏差变化来分配观测的测量方法（或者，在不明确的情况下，方法是一种或另一种类型的可能性）。例如，我们可能有80%的把握认为某艘船使用帆布桶取样，但允许有20%的机会使用木桶。然后，我们可以根据方法调整预期偏差，并指出我们的调整可能有多大的不确定性。下一步将是把分散的观测结果结合成整个海洋的月平均海温图。我们还必须计算这些月度地图的不确定程度。19世纪的观测很少，因此全球海温图需要复杂的空白填补方法。最后一步是将我们的海温图与其他科学家绘制的海温图进行比较。通常情况下，当进行这种比较时，很难理解数据集之间差异的来源。是因为输入数据不同吗？或者不同的偏差调整？或者填补空白的方式？与其他数据集生产者合作，我们将分离这些不同的效果。例如，我们都将使用相同的输入，并隔离不同的空白填充方法的影响。这也将检验我们的不确定性估计——如果错过了影响海温偏差的重要因素，那么不确定性估计可能太小，无法解释不同群体产生的海温图之间的差异。这些问题会误导我们对气候变化的解释。我们将利用新的海温历史重新评估20 C期间气候变暖阶段的解释。

The surface temperature of the land and sea is the main measure of "global warming". Measurements of sea surface temperature (SST) have been made for more than 200 years, first on sailing ships, now on a mixture of ships and buoys (drifting and moored). Technology has changed dramatically over this period, raising serious questions about whether technology changes over time give a misleading impression of how the temperature has changed - and therefore how climate has changed. People first measured the temperature of a seawater sample hauled up in a wooden bucket. Buckets are now made of insulating rubber. Most direct SST measurements are now sent via satellites from drifting buoys. Many other measurement methods have also been used. Different methods don't yield precisely the same SST values, and because global warming is a gradual change, these subtle discrepancies (or "biases") could distort our picture about the timing and magnitude of global warming. So, we must be sure that we understand how the different methods used to measure SST have affected the observations.These biases in SST have been a known problem for years, so why do we believe we can solve it? One reason is that recently many more observations have been retrieved from historical sources. Many ships' logbooks containing weather observations have been digitised. This has nearly doubled the number of observations before World War 2. Another reason is new, stable observations of SST from sensors on satellites orbiting Earth. Most satellite sensors give a detailed picture of patterns in SST and are tuned to drifting buoy SSTs to give reasonable accuracy. But compared to the subtle trends of global warming, they are not stable enough from year to year and across large distances. New high-quality SST measurements from a reworking of the SST measurements of a particular series of sensors are accurate and stable enough. Even better, they do not rely on ship or buoy SST observations, so we can use them as an independent point of reference. A major challenge is that the biases in SST made on ships are different for different measurement methods and we don't always know what methods were used. But we do know how we expect the biases for each method to vary with factors like the amount of heating by the Sun and wind speed. We will use these variations of the biases for each ship or buoy to assign measurement methods to observations (or, where it is not clear cut, the likelihood that the method is one or another type). E.g., we might be 80% confident that a particular ship used a canvas bucket to sample the water, but allow a 20% chance that a wooden bucket was used. We can then adjust for the expected biases according to method, and indicate how uncertain our adjustment may be. The next step will be to combine the scattered observations into maps of monthly average SST over the whole ocean. We must also calculate our degree of uncertainty in these monthly maps. There are few observations in the 19thC, so a global SST map requires sophisticated gap-filling methods. The final step is to compare our maps of SST with those produced by other scientists. Normally when such comparisons are made it is hard to understand the source of differences between the datasets. Was it due to different input data? Or different bias adjustments? Or the way the gaps were filled? Collaborating with other dataset producers, we will separate these different effects. For example, we will all use identical inputs, and isolate the effects of different gap-filling methods. This will also test our the uncertainty estimates - if important factors affecting the SST biases have been missed, then estimates of uncertainty may be too small to explain the differences between the SST maps produced by different groups.Such problems can mislead us in interpreting climate changes. We will use the new SST history to reassess explanations of phases of climate warming during in the 20th C.