4 ms·

I'm not a climate scientist, so take this comment with a hefty serving of salt. > measuring temperatures over such long periods of time to these purported degr

I'm not a climate scientist, so take this comment with a hefty serving of salt. > measuring temperatures over such long periods of time to these purported degrees of accuracy and precision is simply impossible. 0.5 degrees C over 150 years? "Impossible" seems like a bit too strong of a statement. In terms of the raw data, temperature stations have been around since the mid-to-late-1800s or so, and the equipment back then seems more than accurate and precise enough to provide good readings [0]. After that, it's "just" a matter of good record-keeping and accounting for other possible sources of temperature change. As for other sources of uncertainty - I think climate scientists are well aware of those questions, and devote quite some time to analyzing uncertainties in the temperature record (e.g., [1]). You can probably find much more detailed information with some additional searching. I'd expect that the reasoning and/or math behind the estimates is available (e.g., Berkeley Earth's methodology is available at [2]). There may be paywalls for other datasets; haven't gone looking enough to know. > If I lump these measurements with measurements made at different location at different times, and average multiple times to aggregate across regions and time, well, there's no telling what the end result actually means. Are you sure that is what is actually done? The first two sources you provide seem to contradict the scenario you describe. As described in your first link, the need to remove confounders from data is well-known in the field: > The most important [confounders] are (1) changes in instrumentation, exposure, and measurement technique; (2) changes in station location (both position and elevation); (3) changes in observation times and the methods used to calculate monthly averages; and (4) changes in the environment of the station, particularly with reference to urbanization that affects the representativeness of the temperature records. Multiple techniques have been developed to identify and remove effects not caused by climate (e.g., Peterson et al. [3], referenced by your first link). In addition, your first link describes why absolute temperatures are not used: > Anomaly values overcome most of the problems with absolute temperatures such as differences between stations in elevation, in observation times, the methods used to calculate monthly mean temperatures, and screen types. This would mean that absolute temperature measurements from different locations and times are not naively combined as you described. Your NOAA link more or less backs up the use of anomalies instead of absolute temperatures for the given reasons. In general, I'd suspect that there are answers for most of the questions you bring up, but buried in one paper or another. [0]: https://www.chicagotribune.com/news/ct-xpm-2000-05-28-0005280042-story.html https://www.chicagotribune.com/news/ct-xpm-2000-05-28-000528... [1]: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2018JD029522 https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2018JD02... [2]: https://essd.copernicus.org/articles/12/3469/2020/essd-12-3469-2020.html https://essd.copernicus.org/articles/12/3469/2020/essd-12-34... [3]: https://hot-topic.co.nz/wp-content/uploads/2014/11/Peterson_HomogeneityAdjustments_IntJClim1998.pdf https://hot-topic.co.nz/wp-content/uploads/2014/11/Peterson_...

robomartin 5y ago
> "Impossible" seems like a bit too strong of a statement. You might be right. Rather than a statement of fact, this is my opinion based on some four decades in engineering spanning a range from consumer to military/aerospace work. The measurements taken as factual in these models, from my perspective, defy what is possible across 150 years around the planet. If you read through descriptions of the methodology you come across passages with a summary that looks like "average of the average of the average of the average". You can't just keep doing this without suffering a loss in significant figures. You would literally have to have accuracy and precision information on each sensor and for each measurement to then be able to propagate all of this carefully. Furthermore, you really need to know the standard deviation in order to understand the uncertainty in the measurements. If we have a set of measurements in region 1 and the average uncertainty in the measurements is 0.5, while, in region 2, it is 0.3...what should this mean when you combine them with other regions in order to fabricate (because it is being fabricated through computation) a number to represent the entire northern hemisphere. To that point: A number representing an entire region, state, country, continent or hemisphere MUST HAVE an uncertainty attached to it. There is no such thing as "The northern hemisphere is at 18.2 degrees", that would be a silly statement to make. Anyhow, I don't want to beat this one to death. Nothing is going to change. I just feel very uneasy with the way some of this data is presented. I can't remember ever seeing these presentations with a statement that quantifies the uncertainty in the numbers. When you do that, it is easy for the audience --particularly if they have no relevant training-- to assume the numbers given are an accurate and precise rendition of some measurement, when, in fact, that is never the case. What would the graph presented in the article look like with error bands? Don't know. An example of this is the typical graph for global air temperature that shows-up for most searches [0]. Not only does this not have any indication of measurement and propagated errors, it also resorts to calculating the temperature anomaly with respect to the average from 1961 onward rather than the entire period. Th graph would look very different if the average from 18xx onward was used. In fact, we would likely have to explain why and how everything from the 1920's and beyond is red. You will occasionally find graphs that do seem to compare to the average of all data [1]. This one, of course, looks a lot less "violent" than the other presentation. Using a late average pushes 75% of the graph into "it was cold" territory and makes it look like things went to hell after about 1980. One should not be blamed for thinking that one approach fits a desired narrative far better than the other. Anyhow, not sure what else to say. I am not convinced I am right. In order to get there I would have to do weeks of work reverse engineering the data and methodologies. No point in doing that. I am naturally very skeptical --which, I think, is a good thing-- so I'll stick to not being happy with the way things have been presented so far and hope to find something better in the future. [0] http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/globtemp.gif http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/gl... [1] http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/temper80.gif http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/te...

robomartin 5y ago

> "Impossible" seems like a bit too strong of a statement. You might be right. Rather than a statement of fact, this is my opinion based on some four decades in engineering spanning a range from consumer to military/aerospace work. The measurements taken as factual in these models, from my perspective, defy what is possible across 150 years around the planet. If you read through descriptions of the methodology you come across passages with a summary that looks like "average of the average of the average of the average". You can't just keep doing this without suffering a loss in significant figures. You would literally have to have accuracy and precision information on each sensor and for each measurement to then be able to propagate all of this carefully. Furthermore, you really need to know the standard deviation in order to understand the uncertainty in the measurements. If we have a set of measurements in region 1 and the average uncertainty in the measurements is 0.5, while, in region 2, it is 0.3...what should this mean when you combine them with other regions in order to fabricate (because it is being fabricated through computation) a number to represent the entire northern hemisphere. To that point: A number representing an entire region, state, country, continent or hemisphere MUST HAVE an uncertainty attached to it. There is no such thing as "The northern hemisphere is at 18.2 degrees", that would be a silly statement to make. Anyhow, I don't want to beat this one to death. Nothing is going to change. I just feel very uneasy with the way some of this data is presented. I can't remember ever seeing these presentations with a statement that quantifies the uncertainty in the numbers. When you do that, it is easy for the audience --particularly if they have no relevant training-- to assume the numbers given are an accurate and precise rendition of some measurement, when, in fact, that is never the case. What would the graph presented in the article look like with error bands? Don't know. An example of this is the typical graph for global air temperature that shows-up for most searches [0]. Not only does this not have any indication of measurement and propagated errors, it also resorts to calculating the temperature anomaly with respect to the average from 1961 onward rather than the entire period. Th graph would look very different if the average from 18xx onward was used. In fact, we would likely have to explain why and how everything from the 1920's and beyond is red. You will occasionally find graphs that do seem to compare to the average of all data [1]. This one, of course, looks a lot less "violent" than the other presentation. Using a late average pushes 75% of the graph into "it was cold" territory and makes it look like things went to hell after about 1980. One should not be blamed for thinking that one approach fits a desired narrative far better than the other. Anyhow, not sure what else to say. I am not convinced I am right. In order to get there I would have to do weeks of work reverse engineering the data and methodologies. No point in doing that. I am naturally very skeptical --which, I think, is a good thing-- so I'll stick to not being happy with the way things have been presented so far and hope to find something better in the future. [0] http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/globtemp.gif http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/gl... [1] http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/temper80.gif http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/imgheat/te...