[PITCH] can Planet data stop illegal deforestation in tropics?

[guy1ziv2]

Deforestation is without doubt the biggest ecological disaster of our time. If you need conviction, watch David Attenborough 'Extinction' on Netflix. Deforestation can be easily spotted in Planet data with day(s) of occurring, but to stop deforestation we (a) need to know where it will happen enough time before it happens, (b) need to have a system that can pass such alerts to law enforcement with enough time for them to reach the site and stop it; (c) have a very low false positive given the former likely involve sending polcemen on helicopters...

The challenge I propose for this Planet Hack focuses on the first point - can we use Planet real-time daily feed to predict where deforestation will happen in 3 days? in 7 days? in two weeks? in a month?

What would a hack team for this project work towards in 2 days?
One nation that already uses satellite feed to send alerts to law enforcement is Brazil, with the DETER project run by INPE. DETER data is publicly available in http://terrabrasilis.dpi.inpe.br/en/download-2/ (search for 'Legal Amazon notices'), I used this dataset to find the coordinates of a small 0.1km2 deforestation event between 15 Sept and 10 Oct, and found that in fact, it was two events - a smaller deforestation between Sept 17-19 and a bigger one between Oct 7 and 10. In other words - if there was a real-time alert system that could pick up the first event, there were 18 days to stop the second event....

For this Hack, I propose to get a 'feel' of the effectiveness of a real-time alert system. The idea is NOT to develop such a deep-learning model with all its complexity, but rather to use Planet Explorer and the DETER data to assess 'how many km2 could have been saved in Brazilian Amazon if a system of this sort existed?'

How would this project use Planet's data & platform?
I am suggesting here a project that require only using Planet Explorer or basic QGIS/ArcGIS Pro experience. Using either online Planet Explorer or QGIS/ArcGIS Pro with Planet plug-in, it is straightforward to inspect individual DETER alerts (say we look at last 6 months, and split those alerts across the people in the group), Using daily data, pinpoint the day of deforestation (in practice this may be a short range because of clouds) which we call even 1. Then see if within a certain time window (say 2 weeks) there is another event within a certain geographic distance (say 5 km?) and date it as event 2. The centroid lat/lon, dates and area of both events would then be recorded in a Google Spreadsheet.

Using this data, we can make a plot - assuming a real-time Planet based system existed and would have detected event 1 and allowed respondents X days (X=1, 2, 5, 7...) to stop event 2, how much forest would have been saved? We can also plot how frequently event 2 exist close to event 1?

More elaborate data analysis (e.g. predictive modelling) could be envisioned but maybe not in a 2 day Hack. The data collection, however, can be utilized for attempting Machine Learning approach (to predict the probability of an event 2 type, so reducing false positive predictions)...

What obstacles or blockers do you think this project might run into?
I think this is a technically simple project. What we may find, however, is that future deforestation is unpredictable....

And finally... pitch your idea: why should people hack on THIS project in particular?
If you care about SOLUTIONS to global tropical DEFORESTATION, this is a project for you!

If you're interested in hacking on this project, add a 👍 reaction to this post.

[gilles-morain]

Hi Guy, I am interested in change detection using PlanetScope daily imagery, so I will be happy to be a part of this project. I'm familiar with open source desktop GIS applications (QGIS, + to some extent SNAP Toolbox) and Python / Jupyter Notebooks to locally process remote sensing data. I usually process Sentinel-1 and Sentinel-2 scenes locally, but also sometimes on GEE too. What I want to learn during the hackathon is mainly: 1/ what can Planet imagery (PlanetScope & SkySat hopefully) add to my toolbox in various use cases I work on, 2/ how to jump into the Planet ecosystem (including the APIs for finding, purchasing and processing imagery in the cloud). Your project is more focused on the first item but that's ok !

[gilles-morain]

Btw, one of my colleagues is currently working for a development agency on "deforestation" in several areas around capital cities in 3 Western Africa countries using Sentinel-2 and Google satellite background maps (we hope to task SkySat and/or purchase Pleiades archive if available for these areas). Hoping this hackathon project and Planet imagery can help us in those areas where forests do not have the same meaning as in Brazil !

[kevinlacaille]

This project sounds really exciting, I'd love to take part!

[guy1ziv2]

@gilles-morain do you know of any dataset of real time alerts in West Africa? I picked Brazil because I know of DETER, but happy to change to another part of the tropics, or try work on two regions - if similar data exists?

[gilles-morain]

@gilles-morain do you know of any dataset of real time alerts in West Africa? I picked Brazil because I know of DETER, but happy to change to another part of the tropics, or try work on two regions - if similar data exists?

@guy1ziv2 nothing I know about alas. Global Forest Watch does not even consider our AOIs as "forests". Happy to stick to the area you selected for now, so we have this "ground truth" element to start with.

[guy1ziv2]

If you want to take a view of DETER , I prepared a Spreadsheet for this Hack here

[ultraasi-atanas]

hey @guy1ziv2 thanks for posting in the chat just saw your message. I don't know half the stuff that @gilles-morain does, but I know my way round Jupiter Notebooks, I coach python fundamentals at codebar.io as a coding instructor with years of experience in tech and I am really keen to help. I have done some work with the fast.ai libraries as part of a study group. I am available for most of both days until the end

[karayigit-beyza]

Hi everyone! I am interested in this idea and would like to join the group if there is an open space. I am a graduate student in Information Systems in data science concentration at University of Cincinnati. I am comfortable with using python and enthusiastic to contribute and learn. I don't know how I need to proceed but I am positive that I can help by my seniors guide.