Last month’s Measuring Development Conference: Biodiversity on Land and at Sea focused on biodiversity panel data and causal inference. Why? We usually need panel data for causal inference, which helps us understand the effectiveness of wildlife conservation efforts and policy. This post shares themes and insights that emerged over the course of the day that point toward research opportunities and priorities.
Land and sea
It’s easier to observe animals and plants on land than in the ocean, so the conference had twice as much terrestrial research as marine. There’s no marine equivalent of the North American Breeding Bird Survey, which contains decades of consistently measured annual counts of bird species at a high spatial resolution. The exploitation of new data sources to produce publicly-available, research-grade panel data – like predicting bat occurrences from weather radar – is farther behind in the marine realm. Instead, the organization Global Fishing Watch is “measuring the hunters” – counting fishing vessels and their activities – with data from vessel transponders, satellite radar, and optical imagery.
Beyond transects
The traditional method for measuring biodiversity is “transect surveys”. On land, a person walks along a line and records the species within a fixed distance from the line. At sea, a vessel fishes along a straight line and crew record the catch. In both, scientists extrapolate the sampled data over the survey region to estimate the size of wildlife populations.
Transect surveys are the traditional gold standard measurement tool because they apply a known, constant methodology. “Citizen science” data sources like iNaturalist, to which anyone can contribute a wildlife observation, or even aggregators of data from scientific studies, such as the Living Planet Index, often don’t contain enough information about how the data were collected. Causal inference research usually cannot proceed without this metadata. Are there really more sloths in this location-time, or just more survey effort?
Camera traps and acoustic monitoring seem to offer a lower-cost alternative to terrestrial transect surveys. Just as transect lines are spaced evenly or placed randomly, cameras and microphones can be placed to enable representative population estimates. In the old days, only humans classified the animals photographed by camera traps or the bird songs recorded by microphones. Machine learning makes this task easier now. Were I designing an experiment, I would collect the biodiversity outcome data using camera traps and process (carefully, with human oversight) the images with a tool like MegaDetector, Wildlife Insights, Zamba Cloud, or BioCLIP. In the ocean, Baited Remote Underwater Video Systems are starting to offer an alternative to expensive vessel-based transect surveys.
Open data vs. Indigenous data sovereignty
Publicly-available data enables research by making accessing data inexpensive. If there was more public biodiversity panel data, there would be more biodiversity causal inference research. Many governments and non-governmental organizations collect these data, but accessing them as an outside researcher takes more networking muscle than I’ve been able to muster so far. This frustrates me and I wish the entities that recognize the importance of biodiversity research would share the incredible data they collect.
Emilio Tripp and Jessica Camarena of the Karuk Tribe Wildlife Team presented an interesting argument against opening their data or their lands to unregulated data collection or research. Instead, they collaborate in all research concerning them in order to shape the research questions toward their priorities, and they control access to data. Given the centuries of violence against Indigenous peoples, trusting relationships with non-Indigenous researchers is paramount for them.
Prioritize harmonization over technological progress
While technological progress is adequate, I am frustrated by the continuing dearth of publicly-available, research-grade biodiversity panel data. In helping to organize the conference, I particularly sought to identify individuals and institutions working to make biodiversity data comparable across locations and time periods. Scientists have been collecting these data for many decades, and sometimes they are uploaded to a single website, but progress toward exploiting the metadata to make the measurements comparable across studies is slow, as far as I can tell. I began research ten years ago, and I feel the data landscape to an economist interested in applying causal inference methods has barely improved.
Even though I would like to do more terrestrial conservation research, I mostly research marine fisheries because Global Fishing Watch (GFW) creates easily-useable panel data. (Fish in the ocean are wildlife, so I’m happy doing this research.) GFW is able to do this because they have a team of 20 full-time computer scientists. Funding 20 computer scientists to work full-time harmonizing existing biodiversity data seems like the ‘best buy’ in conservation to me. In five years or fewer, we could have the global panel data necessary for an explosion in biodiversity causal inference research.
Join the Conversation