After several years of work experience, he returned to graduate school to fill gaps in his technical knowledge of the environment. In addition to statistics and soil science, he pursued research on the forefront of remote sensing, machine learning, and drones.
In an interview with Michelle Winglee ’19 M.E.Sc., Wilcox discusses his research into drone technology, how it might yield insights into the efficacy of swidden agriculture, and the potential future for the traditional farming method.
Can you tell us about your work with drones?
Andrew Wilcox: The goal here is to evaluate whether a drone can go out into the landscape and recognize things — different trees, phenology, crops, water features — and assess relationships with positive or negative environmental outcomes. Drones can be a useful complement to satellite imagery by providing more detail and a different look-angle. Computer vision with drone images can extract patches of pixels and then derive meaning through machine learning of neural networks. Ideally you would have thousands of drone images to adjust the neural network and calibrate the artificial neurons through a training process.
Sounds like a useful resource. What got you into this?
Wilcox: You know, we already have computer vision in our lives on our phones. It’s a technology that’s out there, and it is going to happen. Why not make it work for natural resources?
Agreed. Can you tell me a little more about a way this technology can be applied?
Wilcox: While I was in the Peace Corps I saw a lot of swidden agriculture, which involves a fallow period where farmers allow the forest to regrow on land previously used for crops. During the fallow period, there’s varying levels of regeneration and involvement in manipulating what grows back.
To my knowledge, nobody's really looking at what constitutes the suitability of a fallow period and the length you should leave it before returning to cultivation. The secondary forest is treated as a monolith when there’s a lot of trajectories that regeneration can take. That's something where I think the drones would be useful. With the high resolution, drones could observe how fallow plots change over time — collecting data on species composition, spacing, et cetera. That could be used to develop canopy height models.
There are a lot of technical challenges. Relating a fallow to future productivity is really multidimensional because you’re talking about restoring soil structure, species composition, growth rate… It’d be tough, but fascinating.
What are some of the implications of this kind of research?
Wilcox: The reason I’m interested in swidden is that I really do think there’s logic in fallowing, especially tropical soils — and fallowing doesn't necessarily mean doing nothing. It could also mean switching from annuals to a tree crop. I think there's even a case to be made for fallowing land in the United States. Basically, I’m trying to do my best to look at the technologies we have today, how they're being applied, and how they can be relevant to sustainable food production.
Swidden has acquired a pretty bad connotation. Some people think swidden farmers just cut down trees, light a match, and let it burn, but there’s a lot of sophisticated knowledge that relates to forest ecology. Unfortunately that knowledge is rapidly eroding and being lost in traditional communities. Also, you have a lot of NGOs that will come in and place value on local forest ecological knowledge, but then when it comes to things like agricultural heritage, they somehow think that needs to be changed.
Swiddens aren’t perfect and they’re mostly appropriate when there’s a low population density. Central Peru is a case of a swidden system breaking down because you have small property sizes and lots of people, so the fallow cycle gets truncated. It also depends on the natural resource endowment in a given place, like whether or not soil properties need to be restored and what are the options to do so. Fallows might be one of many ways to do that. There's a lot of parallels with crop rotations, which are pretty uncontroversially accepted as a positive thing.