My group’s project focused on using land cover data and GLOBE Observer images to create a machine learning model that can predict land surface temperature - a measure of how hot the surface of the Earth feels in a given location. For context, GLOBE is short for “Global Learning and Observations to Benefit the Environment”. It’s an app that lets volunteers across the world take observations of their local community (these could be of cloud cover, land cover, or mosquito habitats) and these observations help scientists track changes in the environment. It’s part of a wider movement called “citizen science”, where everyday people (people who aren’t experts in Earth Science) can contribute to scientific research.
Probably my nicest GLOBE Observer location!
The goal was to help develop a better understanding of the impact that land cover has on the temperature of the environment. This is super important now more than ever because of climate change! Temperatures everywhere are increasing. The 18th and 19th of July both set records for the hottest days on record (since 1940). Although this summer’s extreme heat was partially caused by the El Niño event (a naturally occurring global weather phenomenon), average temperatures are continuing to increase globally due to climate change. One way to try and slightly reduce surface temperatures - apart from tackling the direct causes of climate change - is to alter the land cover of our towns and cities.
Major cities generally suffer from a phenomenon known as the Urban Heat Island (UHI). This is when cities experience temperatures that can be up to 10˚C higher than those of the surrounding environment, and can be visualized as an “island” of heat that is trapped within the city and even lasts during the nighttime! There are many reasons this might occur: tall buildings act as barriers to physically prevent heat from escaping the city (see image below); cities are full of low albedo materials such as concrete and asphalt, trapping heat at the surface of the city; and anthropogenic heat sources such as air conditioners and vehicles generate more heat to reinforce the severity of the UHI.
Diagram of the light/energy transfers that cause the urban heat island phenomenon. Light is converted to heat when it is absorbed, and this waste heat is trapped by the heat island. Additionally, light can be reflected several times between tall buildings, resulting in additional heat absorption. (Ren, 2015)
The UHI can have devastating impacts on the urban population. As of July 2024, urban heat islands are present in 65 American cities that are home to about 50 million people (15% of the US population) (Axios, 2024)! 68% of residents live in environments where the UHI can increase temperatures by up to 5˚C. Dozens of people have died as a result of the extreme heat in these urban areas this summer alone.
Illustration of causes of and solutions to the UHI. (Pavement Technology Inc, 2021)
There is a solution, however! And this is where our project comes in. I think our machine learning model, with additional training, could be a very useful tool for urban planners and city officials to gain a better understanding of how altering their land cover could help mitigate the intensity of the urban heat island in their cities. To train our model, we used:
An example of Collect Earth Online labels I completed for one of my locations. Each dot in the grid can be assigned a label depending on what category of land cover is underneath it. The category options can be seen on the right.
The Collect Earth Online labels and GLOBE Observer images give an indication of the primary land cover of that specific location, which the model would then use to train on and better understand the relationship between the land cover of an area and the land surface temperature of that area. The land surface temperature data we used was sourced from Landsat 8 - a satellite launched in 2013 to track urban expansion, forest loss and regrowth, glacier melting, changes in land use, and water use by crops. Conveniently, it also collects data on the land surface temperature of the entire globe. We used this data to train our model. Unfortunately, we weren’t able to obtain this data for all 950 or so locations we were using from our SEES group; we could only get it for just less than half (~450).
I think our research could be a good first step towards developing a better understanding of how land cover can impact temperature. Although we were met with a lot of setbacks while collecting our data, I am happy with the accuracy we were able to reach with the limited data we had. I am confident that given more time, and if we were able to find the Landsat 8 data we were missing, we would definitely see much more accurate predictions in the models we developed. With additional training I think our models could demonstrate the strong link between vegetation and a reduced UHI intensity.
Overall, I feel I learned a lot through my experience with the program! I’ve gained a much better understanding of the crucial role vegetation can play in regulating surface temperatures - and this information is relevant now more than ever given recent events. I’d love to see how future interns could help take our project a step further - I think there’s a lot of potential for great improvement here!
Thanks for reading!
About the author, Maako is a rising senior from Preverenges, Vaud, Switzerland. This virtual internship is part of a collaboration between the Institute for Global Environmental Strategies (IGES) and the NASA Texas Space Grant Consortium (TSGC) to extend the TSGC Summer Enhancement in Earth Science (SEES) internship for U.S. high school (http://www.tsgc.utexas.edu/sees-internship/). This guest blog shares the NASA SEES Earth System Explorers virtual internship in 2024.
