This is a cool example of a paper showing how iNat can be used to find new populations of rare species. In this case, the researchers were looking for the Rusty Patch Bumble Bee (RPBB), a Endangered pollinator that historically occured throughout the upper Midwest and Northeastern US. It is now mostly known to occur in a scattering of sites centered around Madison, Wisconsin.
The authors used a hierarchical approach to identify potential new populations:
First, they divided the study area into 10x10 km grid cells and noted whether RPBB had been observed in each cell or not. For the cells without observed populations, they counted the number of neighboring cells with observations. Their hypothesis, which they later confirmed, was that cells were more likely to host populations when more neighboring cells did.
Then, for each grid cell without known RPBB, they looked for potential habitat using GIS. They used a database of protected areas (PADUS 4) to find public land with some kind of conservation protection, and they used the National Land Cover Database (NLCD) to confirm that there was forest or grassland within the protected area.
Next, they used iNaturalist data to see where there were already observations of other species of bumblebees. Their hypothesis was that the verified presence of other pollinators indicated good general pollinator habitat that might also host RPBB. They also looked for iNat observations of host plants of RPBB (especially Monarda fistulosa) as direct evidence of potential habitat. They placed a sample point at the highest concentration of iNat observations in the protected area and went out to do surveys.
They detected RPBB at 57 of 105 sampled grid cells, which is a pretty good track record for finding a threatened species! Based on this success, and the presence of additional areas that meet their hierarchical criteria, they think there are another 145 grid cells that are likely occupied by RPBB but have not been surveyed yet.
this ends up creating cells that are ~12 sq km (vs 10 sq km), and you can sort of see how the cells that are highlighted (as having observations) change in distribution, meaning that that the neighboring cells are distributed differently, too. i’ve highlighted the cell in my visualization that i think roughly lines up with the cell that the researchers selected in the graphic in the original post. this cell in my visualization would have had 3 neighboring cells with observations (which is the same as in the researchers’ grid), but the cell just to the west would have had 4 neighboring cells with observations (vs 1 in the researchers’ grid). i know that our grids are constructed slightly differently, but it does make the grid scoring seem a bit arbitrary since it can change like that based on arbitrary decisions made upstream. maybe it wouldn’t affect the bigger picture though. i’m not sure. i didn’t read the paper. so i don’t know why they chose a grid approach like this.
Shows tens of thousands of unidentified observations
You click the Map Grid Tool in the Identify page (next to Filter), pops up a map like yours with a filter "Near Taxon: ".
You enter “Genus Salix”
The map grid tool shows you observations of Willows spp. And as you click the Map Grid Tool, it shows the observations in the cell you’ve clicked. You can filter observations that overlap Willow spp. or near.
Something like this could also help in digging up old observations. If you’re familiar with one species and that species occurs in some area, chances are there are observations of that same species initially ID’d at a higher taxon. So you’d search for Coleoptera and Low → Order. And use that map to show you “Near Taxon: Chrysomela scripta” (for example).
It’d enable everyone to do some of what those researchers did.
@pisum I was also wondering if the grid cell approach would lead to some arbitrariness. Thank you for doing that analysis to show that it does!
I read the paper but I’m not 100% sure why they went with the grid cells instead of a simple nearest neighbor search. Maybe the population models are standardized at 10km so it makes sense to fill in gaps at that scale?
The paper does say that The U.S. Fish and Wildlife Service (USFWS) tracks occupancy trends across a network of 100km2 grid cells to evaluate recovery progress across the historical distribution of RPBBs. Establishing contemporary information on the current distribution of RPBBs is an important first step in monitoring long-term population trends
@petezani - So they only found 57 new populations, but based on their estimated detection rate they predict that they would have found 67 new populations.
