Are Satellites More Accurate than Surface Thermometers?
Surface thermometers are old technology, and to be useful, they have to be extremely limited (more on this in a minute). Satellites, however, are the latest and greatest technology and are capable of measuring temperature (kind of) of different strata of the atmosphere. It can be tempting to think that we should trust satellite data over surface thermometers. But is that true? The tl;dr here is that surface thermometers are more accurate for Global Mean Surface Temperatures (GMST) while satellites make up for their lack of accuracy with some very helpful and useful information about how layers of the atmosphere are changing. In this post, I want to consider first how surface thermometers compare to satellites in general. In a second, follow up post, I will look more closely at individual satellite datasets and consider if one is more accurate than the others.
The Instrumental Record
Perhaps the biggest advantage of surface thermometers is that it's old technology. The instrumental record in Central England goes back to the 17th century, where we have monthly average temperatures going back to 1659.[1] Yet therein lies its limitation as well. To calculate a daily average temperature (Tavg), thermometers (max-min thermometers) were developed that would show a daily high (Tmax) and nightly low (Tmin). Tavg was calculated with a simple average, Tavg = (Tmax + Tmin)/2. This calculation makes intuitive sense for us, but in many cases, it's not going to give you a "true" average, since on any given day, it may be that much of the day was very hot and was only cold at night for a short time, or vice versa. There are other (arguably better) ways to calculate Tavg, and while there are challenges involved with these, scientists have largely adopted the simple average of Tmax and Tmin for one simple reason - consistency. If we change our definition of Tavg to, say, an average of every temperature reading at the top of every hour, that Tavg may not be comparable to historical Tavg calculations. What climate scientists are mostly concerned with, after all, is climate trends, and over time variability in Tavg is likely to cancel out, so over time consistent Tavg calculations provide an accurate assessment of temperature trends.
The thermometer record has not been completely consistent, however. The weather stations are typically monitored once every day at the same time every day. If a thermometer was read at 10 am, the max-min thermometer would show the Tmax from the previous day's high, and the Tmin from the morning low. If the thermometer was checked at 4 pm, the thermometer would return the same day's Tmax and Tmin. This variability by itself is just fine for assessing climate trends, but if the time of observation (TOB) changes over time, then a bias can be introduced in the temperature record, and this is precisely what happened in the U.S. There was a systematic shift of TOB from afternoon to the morning that introduced a bias into the temperature record.
At sea, many thermometer readings of sea surface temperatures (SSTs) were taken from ships, and here method of data collection had an impact on temperature. Before 1940, the data collection procedure consisted of dropping a bucket into the water, pulling it out, and measuring the temperature of the water in the bucket. But around 1940, this was changed to measuring temperature from the engine coolant water intake. This change in methodology also introduced bias into the temperature record. At a later time, I may go into these in more detail, but for the sake of brevity here, we can simply say that sources of bias like these can be quantified, and therefore, the bias can be essentially removed. There are numerous studies in the peer-reviewed literature that document these biases and the methodology to correct for them. Each of these are subject to correction or modification by the scientific method. But generally speaking, there is good evidence that these corrections are working well at removing non-climatic biases from the instrumental record. In the following graph, for instance, scientists were able to quantify the bias introduced into the temperature record through the change in collection method in 1940. Since this bias can be quantified, SSTs prior to 1940 can be increased to correct for this bias.[2]
 |
| Collection procedures biased SSTs cool prior to 1940 |
It should also be acknowledged that the number of thermometers and the global coverage of thermometers has not been consistent over time. While the instrumental record in Central England may go back to 1659, no dataset claims to have anything remotely similar to global coverage going back that far. Only one dataset goes back as far as 1750 (Berkeley Earth), and the error margins are so large in the early years that I wonder why they even bothered. But most datasets for GMST are reliable back to either 1850 or 1880, and that time frame is plenty good for climate purposes. IPCC reports are generally concerned with trends following the 1850-1900 mean, so having several datasets that extend back that far can be extremely useful. But all GMST datasets show less coverage in the late 19th century than we enjoy today. There are actually two problems with this. The first is having too little coverage in some places in the globe, and the other is having too much coverage in other places on the globe (like cities). This puts us in danger of calculating trends that are biased towards the trends in areas with the densest concentrations of thermometers.
To address this problem, scientists construct grids, and they calculate the average temperature of each grid cell based on whatever thermometers are located in that cell. That way if one cell has 100 thermometers and another cell only has 5 thermometers, the cell with 100 thermometers does not influence global temperatures more than the cell with only 5. The average temperature of each cell can be weighted by the surface area of the cell so that the temperature of each cell receives its proper weight in the GMST calculation. But what if there are no thermometers at all in a given cell? This happens more frequently in the earlier years of all global datasets, and each dataset handles this problem a little differently. Some use a process of "infilling" to make educated guesses at the temperature of that cell from the cells around it. Others do not. But in all cases, the lack of coverage increases the uncertainty of the estimate, and the quality of any estimate can only be understood in light of the quantification of its uncertainties.
Bias correction and lack of coverage are reflected in the uncertainties that are quantified for each dataset. Consequently, all GMST datasets show very small uncertainties in recent decades, and they become larger in the earlier years of the dataset. The best graphs will show not only GMST anomalies but the uncertainties associated with them as well. One study from NASA has quantified the 95% confidence interval to be ±0.15°C in the late 19th century, decreasing gradually to ±0.05°C in recent decades in the most recent GISTEMP dataset.[3] This is representative of the kinds of uncertainties associated with the instrumental record. Above I plotted 5 GMST datasets, two from the US (NOAA and NASA), two from the UK (BEST and HadCRUT) and one from Japan. The 95% CI range comes from HadCRUT5. You can see how the uncertainties shrink with time and how each of the major datasets generally fall within the 95% CI for HadCRUT5.
Satellites
 |
| Two Major Satellite Datasets Produce Differing Trends, 0.21 C/decade for RSS and 0.14 C/decade for UAH |
With all these sources of bias and the coverage issues associated with the instrumental record, you may think that satellites would be the clear winner, but that's actually not the case. After all, satellite measurements of temperature only go back to 1979, and by that time most of the bias overage issues associated with surface thermometers had been resolved. Perhaps the greatest weakness of the satellite record is that it's simply too new. It only goes back 42 years at the time of this writing.
Another drawback of satellites is that they don't measure temperature at all. They measure radiances, from which temperatures are inferred. And in order to retrieve temperature data from what satellites do measure, there are a whole host of adjustments and corrections that need to be made that far exceed those made by surface thermometers. And the models used to calculate global temperatures from satellite data are frequently adjusted to address errors in the model calculations. One correction was so significant that it added a full 0.1°C/decade to the warming trend of the dataset. There is no bias adjustment or error correction even remotely similar to this magnitude in the instrumental record.
 |
| What Satellite Data Looks Like Prior to Adjustments |
And satellites cannot be used to infer surface temperatures either. They are limited to inferring the temperature of a layer (or portion of a layer) of the atmosphere, such as the lower troposphere or the lower stratosphere. And unfortunately they do not measure these layers purely. Measurements from the troposphere can be contaminated by other layers of the atmosphere, especially the lower stratosphere.[4] Since the lower stratosphere is cooling, this can introduce a cooling bias into the satellite record. Scientists such as Kevin Cowtan have estimated that the structural uncertainty of the satellite record is about 5 times greater than that of the instrumental record, though the graph below is based on older datasets.[5]
 |
| Accuracy of Temperature Trends (Kevin Cowtan) |
This also agrees with an assessment of satellite uncertainties of from Carl Mears.[6] Due to the effects of diurnal adjustment and satellite drift, the 1-sigma (68%) uncertainty is about ±0.1 C, suggesting a 2-sigma (95%) uncertainty of about ±0.2 C, about four times larger than the same uncertainty from NASA for recent years (about ±0.05 C).
 |
| Uncertainties of RSS and HadCRUT4 |
“Over the last three to five decades, global surface temperature records show increases of about +0.15ºC per decade. Explaining atmospheric and surface trends therefore demands relative accuracies of a few hundredths of a degree C per decade in global time series of both surface and upper-air observations. As this and subsequent chapters will show, the effects of instrumental biases on the global time series are significantly larger than a few hundredths of a degree for the upper-air data, though the global surface temperature compilations do appear to reach this level of accuracy in recent decades.”[8]
And lastly, since satellites do not give us surface temperatures, what they provide do not give us strictly an apples to apples comparison to surface thermometers. These datasets are complementary to GMST anomaly datasets, but they do not and cannot replace them. Yet most of model projections from the IPCC rely most heavily on projections for surface temperatures, since, after all, that's where we live. It seems readily apparent that the instrumental record is the most appropriate, accurate and reliable means we have to evaluate the theoretical predictions of climate science related to projected increases in surface temperatures due to greenhouse gas warming.
Conclusion
 |
| RSS Satellite Temperatures: Lower Stratosphere and Lower Troposphere |
So what is the value of the satellite record? I don't want to give the impression that I have a negative view of these satellite measurements. If nothing else, satellite data shows that warming is not limited to the surface. The troposphere is also warming. We may not have as much certainty about the magnitude of this warming, but we do know it's warming. Likewise, we can also determine that the stratosphere is cooling, and a warming troposphere with a cooling stratosphere is a good indicator of greenhouse gas warming (I will likely write a future post on this point). It may well be that with advancements in satellite technology and improvements to the models used to calculate temperatures, they may one day compete with surface thermometers in accuracy. But the data they provide is valuable to us even now, provided we accept it for what it is.
References:
[1] "Hadley Centre Central England Temperature (HadCET) dataset." https://www.metoffice.gov.uk/hadobs/hadcet/
[2] Smith, T. M., & Reynolds, R. W. (2002). Bias Corrections for Historical Sea Surface Temperatures Based on Marine Air Temperatures, Journal of Climate, 15(1), 73-87. Retrieved Jan 6, 2022, from https://journals.ametsoc.org/view/journals/clim/15/1/1520-0442_2002_015_0073_bcfhss_2.0.co_2.xmlhttps://www.ncdc.noaa.gov/monitoring-references/docs/Smith_Reynolds_2002.pdf
[8] “Temperature Trends in the Lower Atmosphere: Steps for Understanding and Reconciling Differences.” https://downloads.globalchange.gov/sap/sap1-1/sap1-1-final-all.pdf
Comments
Post a Comment