Friday, March 30, 2018

GHCN data Part 5: Australia and Data Problems

In my last blog post I discussed creating accurate models of the existing data. I ended up having five separate models which describe what I would have to term different scenarios. Meaning the history of temperature changes varied greatly from location to location. Temperatures fell in some places and rose in others. In a typical location the over all change ended up being close to no change, even though there were lots of changes in between.

Obviously temperatures, or more accurately temperature measurements, changed over time for certain reasons. Those reasons presumably would fall into categories. Those categories could be broadly defined as local, regional, or global changes.

A global change, by definition, would be a change which affects every location in the world. However, it doesn't mean that change would be discernible in all locations. Global changes could be offset or augmented by local or regional changes. This is where the task becomes extremely difficult. It is impossible to know much less quantify all the local or regional changes which might distort a global change.

Because of this difficulty I decided to make a test case out of Australia. The advantages to this are the limited number of stations involved, the relative isolation of the region, and its location. Unlike the US, which is bordered by the Pacific, the Atlantic, and the Gulf. So the task of filtering the wheat from the chaff, as it were, should be easier. Should being the important word here.

My first pass at this task was looking at population growth as a proxy for changes in land usage. The urban heat island effect, where developed cities are often several degrees warmer than the surrounding areas, is well known. So this is a logical place to start. There does appear to be a correlation between population growth and temperature.


In general more people does equal higher temperatures. However, there are two stations where this pattern is broken rather severely. Those locations are Low Head, which is located on the north of Tasmania, and Otway Lighthouse. Both locations are somewhat similar, the Lead Head station is located at the Low Head Lighthouse. For now I am going to concentrate on Low Head.

Low Head breaks the population / temperature pattern because it has the highest overall temperature increase but without a corresponding population increase. It is listed in the 2016 census as home to 6,765 people with a population density of 331 per square kilometer.

This is the Low Head Temperature record.

Please note the last 10 temperature readings are estimated based upon the average of the previous 10 years. This is one of two instances where I imputed data. Now I will strip that data off and proceed to do some statistical analysis.

This is the same data converted to a standard normal distribution.

This is a graphical analysis which is generally used to look at data and see if there is evidence to suggest a shift in the average has occurred. This will become more clear in a moment, but it is obvious the average has changed over time.

 
This is where the advantages of this technique really come into play. This test shows on obvious change or changes in the site average. There is, however, more. It appears there are three distinctly different average occurring at different times. Meaning, what looks like a gradual rise in temperature from 1959 to 1974 is actually two discrete changes of .69° in 1959 and .82° in 1974. The inference is the site experienced a local change which affected the temperature measurements.

To state the hypothesis formally: The data shows three distinct periods of time which three distinctly different averages. These mean shifts occurred with no apparent transition period. There is no variation in the average between these shifts.

I will test this hypothesis, using the same technique, by splitting the graph into three parts and see if I can prove the null hypothesis. The null hypothesis is the average did shift inside these three time frames.

 
This test is somewhat inconclusive. There is evidence to suggest means shifts did occur, but, with the exception of the first point, nothing falls outside of a 95% confidence interval for the mean. The null hypothesis has not been proven.
 
 
This test is also inconclusive. There is insufficient evidence to support the null hypothesis.


This test is also inconclusive. There is no evidence to support the null hypothesis.

So, what does this mean? That is an excellent question. Failing to prove the null hypothesis, my hypothesis stands. Not proven, just not unproven. Meaning there is a reasonable probability I am correct and it is exactly as it looks. The inference is something changed in 1958 - 1959 and in 1973 - 1974 which permanently altered the temperature measurements. Beyond those singular changes, there is no other evidence of any change or trend.

At this point I want to go back to my designation of changes being local, regional, or global. Based upon the other nine graphs for Australian stations, these changes are not reflected in  eight the other stations. The 1959 shift does appear to be reflected in the Otway Lighthouse record, but the 1974 shift is not reflected. In fact, there appears to be no change at the Otway Lighthouse from 1959 onward. I will expound upon that in a future blog post.

Therefore, the assumption is these shifts reflect localized changes which impacted measurements on site. But what things could effect those kinds of changes?

Low Head Lighthouse
 
The first obvious thing to look at is the means of measuring temperature and how that is done. As mention in the first post of this series thermometers evolved from mercury and alcohol filled glass thermometers to digital thermometers which use a version of a thermocouple to measure temperature. That change really didn't get going until the nineties.
 
A Stevenson Screen
 
Weather monitoring stations use what is basically a white louvered box know as a Stevenson screen to hold measuring equipment. This is painted white to reflect sunlight and limit actual heating of the box. It is enclose on the sides by double walls. It holds the instruments at a standardized height from the ground. This design has essentially remained unchanged since the late 1800's.

There has been some speculation changes in paint formulas in the 1970's to eliminate lead may have resulted in changes to readings inside Stevenson screens. I have no data on that, so I will file it away for future consideration.

According to records on line the Low Head Lighthouse has undergone numerous changes over the years. Buildings have been added, electricity sources changed, equipment moved in and moved out. Probably the most important thing would be relocation, replacement, or repairs to the Stevenson screen. This is a location exposed to the ocean, so it is hard to imagine the same screen being in operation for  over 100 years. I have no doubt it has been painted, repaired, moved, and even replaced at some point. The location where it currently resides is on a rocky, grassless, portion of the site. There are several bushes and large rocks very close to the screen. Other locations nearby are grass covered, bare sand, or rock. The surface color varies quite a bit. All of these could have a measureable affect on a spot just three meters off the ground.

One of the facts most people don't know is temperature isn't uniform even in a single location. More to the point, temperature as measured can vary quite a bit even within a few yards, even in your own yard. For example, you could have a small area in your yard covered in black asphalt surrounded by a grassy area. The asphalt can be quite hot while the grass is cool. A thermometer held above the asphalt on a calm day will read higher than it would when held in a nearby location over grass. Have you ever noticed when walking towards a beach how the sand can go from just warm to too hot to walk on to actually being cool? Ever wonder why wet sand tends to be cooler and dry sand tends to be hotter? If you guessed water you are correct. When it comes to sand and dirt wet is cooler and dry is warmer. The point here is location and maintaining a location is important.

However, these things are just speculation. Ultimately the point of this is something changed suddenly. There is no gradual, incremental change apparent.



This final chart for Low Head was created by making adjustments to the data to eliminate those changes which occurred about 1959 and 1974. I am really showing this more for speculative purposes with respect to the concept of making adjustments to past data. How advisable is this? Is it really kosher to do this? Even though backed by data, I am still making assumptions. I am assuming what ever site changes may have happened did not augment or mitigate other changes which may have occurred. Whatever changes did occur, if they occurred, it is too late to quantify now except as an educated guess. Knowing exactly how to quantify such a change would require measuring in parallel for a period of time so any measurement change would be precisely defined. A more rigorous approach in situ would be to quantify that change and then take actions to eliminate it.

The ramifications of this are pretty high within the context of my study. Therefore, I am somewhat undecided. This is the central question to this whole endeavor. The options I have are to discard the record entirely, modify it and include it, or leave it in and hope it is offset somewhere else by negative impacts.

I have no answer at this point.

More to follow....






No comments:

Post a Comment