The resolution of the data used to inform the Community Charts application is an important consideration. For example, it would not make sense to use raw CRU data at 0.5 degrees latitude and longitude spatial resolution or raw climate model outputs with grid cells upwards of a few degrees across to stand in for point data representing communities, even if a community is a large, sprawling city. Instead, downscaled data are used.
PRISM, the historical baseline, uses 2-km grid cells, as does anything downscaled to the PRISM climatology, e.g., SNAP’s CRU 3.2 data and five global climate models. Climate trends for communities are extracted from these gridded data products using the grid cell in which a community’s given spatial coordinates fall. These coordinates may fall in the center of a cell or they may fall near an edge or corner, but considering a cell is only four square kilometers and the focus is on very broad trends at a decadal time scale, this is not particularly problematic.
Nevertheless, some thought should be given to the accuracy of the coordinates. The communities data set from which community names, populations, and coordinates are pulled, is one I would consider to be of ill repute and quite unkempt as previously mentioned.
A glance at populations for many communities makes it immediately clear that the numbers in the data set are imprecise if not also inaccurate. It is technically sensible to be skeptical of the accurateness of a population like 100,763, as populations have constant turnover. We don’t care about that level of accuracy in that particular context. On the other hand, it makes a lot of sense to be skeptical of a city population of exactly 100,000. This generates concern regarding an unknown level of precision. And let’s forget about how outdated these numbers may be.
Population data is clearly not part of the Community Charts application, but it flows from the data set used to inform the application. It is not automatically expected that spatial coordinates, or any other particular variable in a data set, be of the same quality of measurement as any other variable. In this case I do not expect locations to be way off, nor do I expect them to be out of date (but communities do “move around” if they are defined by a proxy variable which can move).
Still, this leaves an open question regarding the accuracy and precision of spatial coordinates. Chiefly, if they are not to a high enough precision, they most assuredly will not be accurate. I don’t know offhand how many decimal places are required in longitude and latitude to precisely reference a specific 2-km grid cell, but it’s more than a few.
Another potential weakness of the data is in terms of strict accuracy. This is based on my observation that there are apparent duplicate communities, same name, same province, and almost the same coordinates, but not quite the same. Sometimes a community occurs many times within the data set and it must be considered whether this represents a change over time in location of a city center or public administration unit. It is safe to assume the data set does not contain communities which are truly nomadic.
With all that in mind, we proceed on the assumption that the spatial coordinates for each community are “good enough”. Again, our interest is in broad scale climate trends, enough so that some uncertainty in the spatial dimension based on the overall unkemptness of the communities data set is a moot point. However, spatial scale must still be considered.
Would you directly compare a sample of annual average temperatures from one community to a sample of 30-year annual temperature climatologies from another community and argue that the latter community has more stable climate? What about average July precipitation from a 2-km downscaled grid cell vs. that seen over a 10-minute grid cell?
Although the communities in the Community Charts application range roughly from northwestern Alaska down through the western half of Canada, a quick calculation for a typical community situated at a typical latitude in the application shows that a 10-minute by 10-minute grid cell can range anywhere from 100 to 200 square kilometers. That’s not typically what people think of as a community. Compare that to four square kilometers for a 2-km by 2-km grid cell.
It is not known in advance what kind of effect this will have on assessment of climate trends, in either temperature or precipitation, or whether any difference will be noteworthy. However, this is why we analyze these things, so that we know for sure and can make an informed decision on the legitimacy of mixing the two scales.
In the absence of context it is difficult to imagine why this mixture of scales, one to be used for some communities, another to be used for the rest, would even be a consideration. It’s not like the two scales are 1-km and 2-km, or 10-minute and 12-minute. So I will briefly expound upon that. The reason it is considered here is simply because there is interest giving the Community Charts application a nominal boost in the number of communities available. Inclusion of additional communities is sensible because they will be specifically from the Northwest Territories in Canada, which is currently not represented in the application.
The problem is that there is good reason this region was never included in versions 1, 2, or 3 of Community Charts (it will eventually be added to version 3 as well as this version 4 here). The PRISM climatology does not cover this area. Therefore coarser climate model outputs (and CRU data) could not be downscaled using the 2-km PRISM climatology. The finest resolution climatology available over this geographic region is the 10-minute resolution CRU climatology, which is world wide (over landmasses only, ocean excluded, since CRU is weather station based).
Whereas the vast majority of communities in the data set will be able to retain their basis in 2-km downscaled data, the nominal addition of communities based on 10-minute resolution data will likely incur some initial fanfare and be touted when and where possible. While there will surely be some caveat noted on the chart indicating the spatial resolution and what people should take away from it, this does not automatically make it acceptable to use. It depends on what kind of information comes through at that coarser scale in the context of community-level data.
Let’s take a look again at our test case, Fairbanks, Alaska. Error bars have been removed to help focus on mean values.
Across months and decades, temperature differences between the 2-km and 10-minute data sets tend to be within a degree Celcius or so of each other.
Next, the same data are plotted using floating range bars only. This makes it easier to focus on the variability at different scales.
Variability also appears similar in magnitude and change throughout time.