Cooling stations. A UHI Hint

September 29, 2010
By

(This article was first published on Steven Mosher's Blog, and kindly contributed to R-bloggers)

Update: google earth files in the box: Personally I like to look at things backwards. Why are cool sites cool? So download the kml or kmz file and you can tour 62 sites: All with 90 years of data or more. All with a cooling trend. And all “supposedly” urban. what do you see at the meso scale. Anyway’s The next drop will have the animation code, the kml code, file download diagnostics, and a script to replicate the cool urban stations.

Let’s recap where we are. I went in search of the the biggest warming trend and biggest cooling trend in the data.( for an entirely different reason ) And after restricting our view to stations with long records, stations with 90 years of data in the period 1900 to 2009, we landed on this distribution of trends. The trend per decade.

Again, just looking at the distribution gives us some information. We’re seeing what appears “somewhat” normal, Just a quick look at the density,ECDF and QQ.

Density

ecdf

QQplot

By eye I was wondering if the trends were a gamma –err prolly not–  or log normal. A gamma sorta makes sense if one considers that a warming rate is going to take a while to appear. At some locations the “wait” time will be shorter while at other places the “wait” time will be longer. However, I didnt do any formal testing on this other than looking at the QQ– shrugs– just an observation to maybe come back to.

As we see we have 1492 stations. Looking at the metadata for urban/small town/ rural we have this:

table(Yr90Inv$Rural) R S U 757 320 415 Which shows us that for stations with long records over half of them are “rural” by the designation in the inventory. Now, of course, that designation has it flaws, but this is just exploratory data analysis. The next step I took what to isolate just the stations that had negative warming. otherwise known as cooling. And we pull up the most extreme case: Id Name Lat Lon Altitude Rural 4682 42572681004 KETCHUM RS 43.68 -114.35 R I noted a couple things. It’s a rural site in a Mountain valley. But seeing that drop in the later years and what may be a discontinuity ( undocumented station change) , I moved on to next station 42572438001 OOLITIC PURDUE EXP FM 38.88 -86.55 198 175 R A couple things: What we know from the distribution of all stations is that rural stations constitute half of the sample. Now, on the supposition that there is no UHI, that rural and urban see the same warming over 110 years of data one can expect this. One can expect that the sample of cooling stations drawn from the whole sample will have the same distribution of urban/small town/rural: roughly 50:25:25. So, when the second station I drew from the far end of the distribution was also rural, well thats like two heads in a row. Nothing special, but how long a streak would I get? Well, you have to look at 32 stations ( sorted from coolest to warmest) before you get to an urban station. hmm. And the other thing that was striking was this. That station ALSO happens to be the first NON US station. go figure: On one hand the US stations tend to have longer records so I can expect a lot of US stations. Still the number of us stations in the “cool” distribution seemed worthy of investigation ( later work if somebody wants to play) 61111518000 PRAHA RUZYNE 50.10 14.25 A couple points that I have discussed about UHI. UHI results from a few things. Chief among them is the disruption of the boundary layer that results from building tall buildings. And of course changes in the surface properties and lastly waste heat, probably the least important. Population is not a precise measure of any of these. So here we have a site probably at the airport with a clear fetch all around. We do not have an urban environment with tall buildings. We also see that airports are not categorically bad. There is another factor as well that Oke mentions that few have picked up on. That is the difference in wetness between the urban landscape and the surrounding rural environment. More on that later. lets hit the next “Urban site” 42572216003 ALBANY 3SE 31.53 -84.13 Map And the next: 12567083000 ANTANANARIVO -18.80 47.48 Map So in the top 50 cooling sites, here is what we have: 3 ”urban” sites. 48 sites from the US. At those “urban” sites we have two airports with what appear to be long fetches. If you have a long fetch, your UHI is going to be minimized. If you have buildings destroying that fetch, you have some of the preconditions to generate UHI. That’s why, for example, I think some of the concerns about waste heat at airports are potentially flawed. And one final note. Note the lake. More on that at a future date. So sum up the little exploration of the data, I’ll leave you with something to chew on: recall that the 1500 or so stations with long records were split 50:25:25: Rural:small town: urban. When we segregate the data by trend and look at cooling stations, is this structure preserved: > table(cold90Inv$Rural)

R   S   U

308 143  62

Nope:  Is it significant? Does it actually indicate anything? but it is interesting that when we look at the long records, and the cooling stations within those long records that urban sites are very few in number. Are they Actually urban? Nobody seems to ask questions like that. Partially that’s because people don’t understand everything that goes into UHI. Also because they tend to be mesmirized by the close up shot which focuses on waste heat or surface material. They often forget the bigger meso scale picture. That tarmac, sitting on the coast of an ocean has a long clean fetch and most students of Oke or Parker know what wind does to Tmin. Next up, I will need to re integrate some KML code and we can take a tour of these 62 ‘urban’ stations and all the ‘cooling’ stations. On the ground what do they have in common. Not in the “metadata”. on the ground.

R-bloggers.com offers daily e-mail updates about R news and tutorials on topics such as: visualization (ggplot2, Boxplots, maps, animation), programming (RStudio, Sweave, LaTeX, SQL, Eclipse, git, hadoop, Web Scraping) statistics (regression, PCA, time series, trading) and more...