By Keiko Nomura, Senior Analyst, Space Intelligence Ltd.
Editor’s note: Today’s post is by Keiko Nomura, a Senior Analyst at satellite data analysis firm Space Intelligence Ltd. in Edinburgh, Scotland. In mid-2019, Nomura began sharing tips and helpful videos on Twitter that guide Earth Engine users in creating visualizations. Below, she shares some of her 10 favorite tips.
During my final year as a PhD student at the University of Edinburgh, I used Earth Engine for my research into sustainable land use in Southeast Asia. I used Earth Engine to detect oil palm and rubber trees in order to distinguish from forests. As I got to know my way around Earth Engine, I started compiling a mental list of tips for addressing challenges like avoiding computational limits in classification, and shared my tips on Twitter.
If you haven’t seen my Earth Engine tips before, below is a curated list of the most-liked tips, including suggestions from my geosciences colleague, Morgan Crowley, who manages the group Ladies of Landsat. There’s also a special bonus tip that’s making its appearance here for the first time.
Brand-new tip: Removing color from earth imagery
It can be helpful to remove color from satellite images to magnify the objects you are studying — like specific buildings or vegetation. Tim Wallace, a creative lead at Descartes Labs, inspired me to investigate easy ways to do this in Earth Engine when he posted Photoshop images in which he removed color around his chosen subject. Harel Dan, a GIS coordinator at 4M Analytics in Israel, took the idea a step further by writing a script for the Sentinel Hub platform.
I decided to create a version of the script for Earth Engine. The examples below highlight the impact adjusting color can have.
Based on the script, I hoped I could apply this technique to highlight areas or pixels of interest, such as a deforestation area. Below, I highlight recent forest fires in Queensland and New South Wales, Australia by comparing high-vegetation area images. The grey background was used to mask low-vegetation areas, so the colors only show where there were dense trees in May 2019. The image on the right shows what happened to those areas with trees in December 2019. I added topography to the images to get more information about the locations of fire as well as to highlight the contrast further while adding a three-dimensional effect.
I also tried studying the deforestation area using the GLAD alert dataset. But as it turned out, the map of New South Wales, Australia, below left, was not as vibrant as I thought it would be. For the map on the right, I applied a “buffer” in red — but since the file is raster and not vector, I used the
focal_max morphological operation (there’s a bonus tip for you!).
Popular Earth Engine tips from my Twitter feed
Use NDVI time-series in RGB to see changes in trees or crops. This tip, inspired by remote sensing specialist Annamaria Luongo, explains how to see when trees or crops were cleared or have grown by stacking three different NDVI datasets in RGB. You can also use this approach to see new roads over time, or how roads become narrower as vegetation grows.
Add more maps to the Earth Engine console. Can’t get enough maps? Tap into the
ui.Map widget to add more maps in the console, and follow my example mapping the University of Edinburgh.
Identify objects using NDVI. If you just want to delineate non-vegetation objects and don’t want to use machine learning, you can do that by using the raster to vector conversion function and NDVI values. Make sure to use images across different seasons! This is a good low-cost and low-power alternative.
Display and estimate forest areas according to definitions. This is my first contribution to the Earth Engine community tutorial. You can learn how to extract forest areas by applying technical thresholds of canopy cover and minimum area requirements.
How to link maps. With
ui.Map.Linker, you can show images from different time periods by assigning unique band names, like this example I created to show Puzhal Lake in Chennai, India from 2016 to 2019. You can use this example in Code Editor to get you started.
Get time-series views of pixels. Get NDVI time-series views of a pixel with a click, using Landsat 8, in an example using oil palm data. This really shows the advantage of Earth Engine and the cloud computing — so much data, instantly! Great for quick checks.
See trends and years of tree loss. Use a
ui.Chart.image.histogram to see the trends and years of tree loss, with an example from Papua New Guinea.