Spatial Implicit Neural Representations for Global-Scale Species Mapping
Motivation
Estimating the geographical range of a species from sparse observations is an important geospatial prediction problem with a long history in ecology and conservation.
However, traditional methods struggle to take advantage of emerging large-scale crowdsourced datasets which can include tens of millions of records for hundreds of thousands of species.
How can we use these large datasets to scale up species range mapping?
Our Work
We train Spatial Implicit Neural Representations (SINRs) to jointly estimate the geographical range of 47k species simultaneously.
Using only noisy and biased crowdsourced data , SINRs can approximate expert-developed range maps for many species.
SINRs learn rich representations of the environment from species occurrences alone.
SINRs scale gracefully, performing better as we increase the number of species in the training set and the number of training examples per species.
To facilitate future research on this important problem, we provide four new benchmarks that measure different aspects of species range estimation and spatial representation learning.
