I had the opportunity to take both the Intro and Advanced courses on Remote Sensing. I learned about the physics behind remote sensing, how to download, display, and pre-process and process satellite (raster) images.
During my time in my remote sensing courses, I took special interest in wildfires. At the time, the August Complex had only just begin to burn and had yet to become California’s largest wildfire. For my final project in the Intro class, I traced and compared the size of the burn scars for the Rim and August Complex fires. I animated their change over time and compared their damage.
Click the link to view the whole story map.
In the Advanced class, I decided to keep the subject of wildfires but greatly increase my scope. I wanted to process larger quantities of data as the previous semester showed me how time intensive my previous methods were.
For this project, I chose to follow the methods in the paper “Predictive modeling of wildfires” by Sayad et al. to write scripts to extract training data for a machine learning model to predict possible future wildfires. I processed and extracted the means for past fires from several MODIS products: Land Surface Temperature, NDVI, and Thermal Anomalies.
Most of this work was done in R though I did some steps at the Windows Command Line.
Google Earth Engine
In Advanced Remote Sensing, I worked with Google Earth Engine for the first time. We wrote scripts to compare Random Forest and CART (Classification and Regression Trees) classifications, create world wide elevation min/maxes and slops, and to create false color and NDVI images from Sentinel-2 data.
To view my full scripts, visit my Google Earth Engine Project.
One of my favorite parts of Remote Sensing was working with Lidar data. Yes, the datasets are quite large and they can take a while to work with, however, there is so much potential to glean useful information from point cloud images. I have used both LAStools and R to handle Lidar data.
It would be my dream to work with lidar data professionally full time, whether that be creating digital surface and canopy height models or helping to discover the vital role that trees play in carbon capture by calculating their biomass.
Working at the Command Line
Remote Sensing helped me conquer my fear of working at the Windows Command Line. My first exposure was working with LAStools to process and open Lidar images. While LAStools does have a GUI, it is extremely limited compared to working directly at the Command Line with the files.
The second time I worked at the Command Line was briefly to work with the GDAL package to project, resample, and process a thousand raster images. While it took some getting used to the particular syntax, after having some experience with it, I can say that there is a certain amount of elegance to the flag system in the functions.
eCognition — Object Based Image Analysis
In both Intro and Advanced Remote Sensing, we spent time processing images in eCognition. We built rule sets to group and classify objects. In many ways, this is the ultimate supervised learning process, as we are “teaching” the computer the rule sets in order to understand the image.