Open California for education at Wageningen University

This past January I have got the chance to be part of the first use of the Planet Lab's open California data-set in an educational context, and the results are pretty exciting. We've got two student teams working on assignments involving planet's API and data.

The experiment took place at Wageningen University in the Netherlands with the students of the Master of Geo-Information science.

It was in San Francisco when I joined Planet Lab's happy hour organized as a side event of the AGU fall meeting that I heard of the Open California data-set. I thought: "A massive archive of Very High Resolution images covering the entire state of California; we can do something really cool with that !" And I decided to try using it as part of the Geo-Scripting course taught at Wageningen University.

One of the team took advantage of differences in temporal signatures among crops to perform crop type mapping using multi-temporal images. A color composite using NDVI at different times of the year for the red, blue and green bands can reveal different temporal patterns of greenness among agricultural fields.

Adapting to the changing requirements of the Geo-industry

I remember when Arjen Vrielink — an experienced software developer and geographer we invited to give a guest lecture on python geo-programming — mentioned in 2014 during his lecture: "10–15 years ago having some programming experience for a GIS job was a small plus, but now not having it is a big minus". It is precisely to respond to this increasing demand of the Geo-industry that the Geo Scripting course was created three years ago. The course runs once a year in January and is coordinated by Jan Verbesselt, researcher of the Laboratory of Geo-information Science and Remote Sensing (the research group coordinating the master program).

The course focuses on using R and python to solve geographical questions, but not only. The importance of keeping a project well organized and managed (using revision control with git), or knowing how to properly ask for help on programming fora like stackoverflow is emphasized as well. We also encourage students to adopt an open source attitude; we teach essentially open source tools, all assignments have to be submitted via github, and the course itself is open source.

Practical session with the students of the Master of Geo-Information at Wageningen University

The right balance of theory and practical experience required

Successful learning requires a balance of theory and practice. The first three weeks of the course are spent on both; concepts and theory in the mornings and practice on simple cases in the afternoons. Practice is taken to the next level during the last week of the course when students are asked to work in teams on (larger) projects of their choice. And that is where Open California played a role this year.

The new generation of remote sensing specialists have to become familiar with modern tools

I believe there are two important points for using such platform for education:

1. It provides an unique opportunity to work with recent high resolution images. High resolution data are otherwise not cheap and we don't get to use them a lot in universities.
2. The platform and the API used to access the data illustrate well what I call "modern remote sensing", and we have to teach the new generation of remote sensing specialists to use such modern tools.

2016 projects using Open California data

Two teams of two students chose to work with the Open California data this year.

Automated detection of trees over Sunset district in San Francisco

One team worked on urban tree detection and mapping, which gave them a good opportunity to explore the planet API as well as the numerous features of the Orfeo ToolBox, an open source toolbox designed specifically for high resolution remote sensing image analysis. They focused on San Francisco, were able to test different classification algorithms and experienced the joy of working with rasters in python. Have a look at their github repository for further details.

I found the API to be extremely well documented and the fact that all the students could access the data within a few hours confirmed that impression.

A time-laps of RapieEye images shows agricultural fields being cultivated at different times over the year 2015

The high temporal frequency of Planet's data offers the possibility to explore the temporal dimension of the data-set. The other team took advantage of that to work on crop type mapping over a small area of the California Central valley. Different crops have different cropping calendars, and they therefore all have a unique temporal signature, as shown in the graph below. Link to their github repository.

Field level temporal patterns of greenness differ among four agricultural fields (look for polygons with magenta borders in the time-laps animation for corresponding fields)

Wrapping up

Using planet's Open California data in an educational context was a great experience. Setting up the topics and making sure students got their API key on time required a bit of logistic and planing but all was well worth it in the end. By providing access to their platform and data, Planet Labs can play an important role in education and I sincerely hope they pursue this effort in the future.