We’ve all used Google Earth — to explore remote destinations around the world or to check out our house from above. But Google Earth Engine is a valuable tool for conservationists and geographers like myself that allows us to tackle some tricky remote-sensing analysis.
After having completed a few smaller spatial science projects in the cloud (mostly on the Google Earth Engine, or GEE, platform), I decided to give it a real workout — by analyzing more than 300 gigabytes of data across 28 United States and seven Chinese cities.
This project was part of a larger study looking at trees in cities. Why trees? Trees provide numerous valuable ecosystem services to communities: benefits associated with air and water quality, energy conservation, cooler air temperatures, and many other environmental and social benefits.
It’s easy to understand the benefits of trees: stand outside on a hot sunny day and you immediately feel cooler in the shade of a tree. But what’s not as obvious as the cooling effect are tree’s ability to remove particulate matter (PM2.5) floating around in the air we breath. And this important, as this type of air pollution is implicated in the deaths of ~3 million people per year.
The Conservancy researched the relationship between city air quality and the cooling effects of trees. Results of this study will inform the Global Cities Program initiative on Planting Healthy Air for cities — the objective being to show how much trees can clean and cool, how much it will cost, and so forth.
The first step to understanding the cooling effect of trees is knowing the number and exact position of every tree in each of our 28 study cities — which is about as difficult as it sounds. Knowing exactly where individual trees are located will also enable us to target where trees should be planted for the maximum cooling and cleaning effect.
Getting this amount of detailed information is no small task, as it requires a lot of very high-resolution spatial data; imagery of 1 to 2 meter resolution covering more than 18 million acres. This works out to around 300 GB of data. Analyzing that amount of data could take a long time and potentially be both complicated and expensive.
We needed a solution, and cloud computing was the answer.
Luckily, the National Agriculture Imagery Program (NAIP) acquires aerial photography during the growing season — and covers both agricultural areas and U.S. cities. NAIP also recently added the important Near Infrared-Red (NIR) band, which allows geographers like me to accurately depict vegetation. And all of this imagery has been ingested into the GEE platform, along with an enormous amount of other satellite imagery. It’s a wealth of data just waiting for scientists to use.
The next step was pretty simple in terms of computation: find all the vegetation. Living, green plants use solar radiation for photosynthesis. By doing so, they absorb red light and reflect near-infrared light. We geographers use this characteristic to detect live biomass using an equation call the Normalized Differential Vegetation Index ratio (NDVI), a very well-studied indicator of above-ground live biomass that uses the red and near infrared bands. (In this case from the NAIP aerial photographs).
But there is a problem — all of the live vegetation, including golf courses, baseball fields and other grassy areas, would also be depicted as live vegetation during that time of year! But our analysis relies only on trees. To get around this problem, I used a texture analysis to remove the smooth areas — like grass — from the analysis.
The next step was to export the data and analyze using a desktop GIS to exclude the flat areas that have the same NDVI signal, or biomass, as the trees we are after. I could have done this in the cloud as well, but we needed the information for further analysis, including scaling up the analysis to 245 cities worldwide using other global datasets.
My method is a good example of how to use cloud computing to solve a complex and large spatial problem without downloading massive amounts of raw data. Some other advantages include the ability to easily share scripts with others, the ability to upload your own imagery for analysis, an active community with online help, and some very handy embedded algorithms (like the texture analyses I used). Oh, and did I mention that it is free?
But there are a few challenges: you might have to learn yet another programming language (Java scripting in this case, although GEE will now allow python scripting too). There are also limits on how much you can analyze at one time, requiring some juggling of datasets. And lastly, GEE is still a beta service, and there is the prospect that for-profit Google might discontinue this free service at some point.
Overall, remote-sensing specialists and geographers looking to analyze vast amounts of data quickly, efficiently, and simply should take a serious look at GEE. It could solve a number of problems.
As for our results about the link between trees and cooler air in cities, our research found that trees are already providing up to 68 million people with significantly cleaner and cooler air. What’s more, we found that planting and maintaining trees is as cost effective as other solutions to reduce pollution (and should be used in conjunction to those efforts).