- @lesserj Talks about his thesis project to identify landscape disturbances using big data statistics
- @emiliom Seeks out brilliant suggestions about his trail mapping and related geo-volunteering work on the Green River
- @lovegis Talks about the use of open source on the Maps Team at Amazon.
- @powersa will lead a rally for Spring Fling 2016
- @cliffordsnow will announce our venue choice for 2016 SOTM-US conference. And International Women’s Day Event in Bellingham.
- @you tell us about what you’re trying to solve.
Jacob Lesser, Satellite Imagery Change Detection Algorithms
Heard of the EROS data center? They do sweet downloads for satellite data like Landsat and Modis. Originally, if you weren’t downloading data as satellites sent it in, you didn’t get it. So they built some data centers to capture all of the information (I think). In the basement of EROS is the largest collection of remotely sensed images in the world.
He worked for the Fire Science Division that mananges USGS’s fire data. Used a dataset called LANDFIRE that includes information at 30m resolution and anything that you’d need to monitor fire. These inputs are data needed for prediction models. Includes vegetation type, forest canopy cover, fuel classifications, etc.
How do they update it? They incorporate detected changes from Landsat imagery and disturbances and incorporate them into the new models. How much data? 224 billion pixels per day. BAM.
Current methods for change detection: Real-time fire specific detections, pattern matching, and the tool he developed!
His method JUST detects fire, ignores all other changes. It uses Modis for picking up “hot spots” like fires and volcanoes. Because resolution is low it’s bad for fire analysis. Looking for particular shapes and patterns you can find disturbances. These shapes are different depending on the ecosystem, which makes things complicated!
Methods & Tools:
- extract & clean data: using python & GDAL
- stats: run decision trees that look for patters that define output and tell you how likely the pixel is important for fire analysis with C using Cubist, QGIS for viewing/testing
- apply the model from the Cubist model and apply it to a new scene, you get outputs with liklihood of disturbance
- set disturbance threshold that says “anything greater than this value is likely fire”
- validate results: compare detected fires to the fire database, compare to BARC data for the scene and year in question, and human assessment
- results: for most scenes it looks like it works pretty well. It’s dependent on the amount of training data available. Where there’s a lot of training data, the model is good at predicting.
Ubunutu computer, yah never know when it’s gonna project… And it does!
Emilio, couldn’t wrap his head around this idea that there was a Gorge 40 minutes outside of Seattle. The Green River Gorge is an oddity in Geology of the region. Beautiful place, really worth a visit. Trout fishing, swimming holes, great foliage.
As an environmental scientist, he wants to know his area. So he researched the gorge online. But he found online data sources (Google) to be quite misleading. There are a lot of different types of public land, but these online resources just mark them as “green”.
There is a long history of locals taking down signage that helps you access these areas. It’s a great area, but really hard to know what you can and can’t do.
This inspired Emilio to document the area to provide better access. Text descriptions only get you so far. Especially in the area of ubiquitous maps, we can do better. Emilio and a group have started sporadic mapping while hiking the area. They start data collection when they start, but then the device (phone) just runs and doesn’t require active participation. You can just hike! They’re seeking the simplest possible work flow.
Over the years, they’ve collected a lot of hiking path data. The results are really cool. Great gpx data for many hikes. How do you put it out there? The data is messy, too messy to make a public map. Accurate and detailed geo data takes a lot of time. It’s prohibitive.
How do you go from raw gpx data to something other people can actually use? He wants to get the data somewhere that people can use it… Without spending days data munging…
Seth suggests Strava. That can be a good first stop for this type of data, and makes it accessible to lots of different folks.
Nick is building apps with Mapbox. He’s been working with TileJSON and
mapbox-gl-native) that allows him to add custom imagery tiles to a web map.
He’s been using Dropchop to load in geojson files, but dropchop doesn’t handle big data files because it has to draw the data every time. He’s been using
mapbox-gl-js to turn Dropchop into a vector editor so he can drag in huge geojson files and they can be rendered as tiles instantly. Check out the progress of this feature here: https://github.com/cugos/dropchop/issues/216
If you want to help, head on over to github.com/cugos/dropchop to see what’s up.
Reily works at Amazon on the maps team. They are developing a maps API. The Fire phone does some sweet enhanced 3d points of interest based on your face position and the phone using 4 cameras. WHOA that’s a lot of cameras!
They have their own tilesets and datasets. Now the mapping team is working with packaging and they have been tasked as the “Address Authority”. They are matching and conflating a ton of address data. Addresses have TONS of different representations, which is a problem for determining where actual addresses exist and which are formatted correctly or not. Because of differences in addresses across countries, address validation/geocoding is incredibly difficult.
powersa & the spring fling crew
There are some teams put together preparing for the Spring Fling on Friday, April 15th. If you would like to be a part of the volunteer/planning, send a message to firstname.lastname@example.org.
- Captain (Andrew)
- Sponsorship (Aaron Racicot & Peter Keum)
- Speakers & workshops (Greg Corradini)
- Operations (Ryan Small & Sam Matthews)
- website, AV/equipment, code of conduct
- Food committee (YOU)
Cliff submitted the Seattle bid for FOSS4G-NA! Hopefully we hear back by next week, but if things go well we’ll be hosting it at Seattle U in late July. WOop!