Evaluating the Effectiveness of Location Encoding Methods for Species Distribution Modelling
Organizations: University of Edinburgh
Abstract
Understanding the spatial distribution of species is a key question in ecology. A variety of solutions, from statistical, traditional machine learning, through to deep learning-based methods have been proposed. These methods take sparse observations for a species of interest as input and aim to generate dense spatial outputs indicating the spatial range of the species. Orthogonal to the choice of statistical model used is the question of what input features are most informative. Thanks to recent developments, the number of options has greatly expanded, with choices spanning different climate variables, learning-based encodings, earth embedding models, or raw transformed geographic coordinates. In this work, we introduce a standardized evaluation framework and quantify the effectiveness of different location encodings in the context of species range estimation, both at the local and global scale.
Figures & tables
| Global | UK | California | |||||||||
| Features | Type | Dim | RF | LR | ME | RF | LR | ME | RF | LR | ME |
| coords | loc | 2 | 0.72 | 0.16 | 0.58 † | 0.69 | 0.63 | 0.66 | 0.56 | 0.47 | 0.54 |
| sin cos | loc | 16 | 0.72 | 0.66 | 0.69 | 0.68 | 0.65 | 0.67 | 0.55 | 0.52 | 0.55 |
| rand fourier | loc | 128 | 0.76 | 0.67 | 0.71 | 0.69 | 0.67 | 0.68 | 0.59 | 0.55 | 0.56 |
| ERA5 | env | 117 | 0.63 | 0.51 | 0.50 | 0.68 | 0.68 | 0.67 | 0.60 | 0.56 | 0.54 |
| bioclim | env | 20 | 0.54 | 0.24 | 0.47 | 0.66 | 0.65 | 0.66 | 0.61 | 0.58 | 0.61 |
Appendix figures & tables20 assets
Supplementary material from the paper’s appendix.
Appendix
| Species | Group | Species | Group | Species | Group | |||
| Pipistrellus pygmaeus | B | 10,894 | Pipistrellus pipistrellus | B | 8,778 | Plecotus auritus | B | 8,284 |
| Myotis daubentonii | B | 6,047 | Nyctalus noctula | B | 5,631 | Myotis nattereri | B | 3,216 |
| Rhinolophus hipposideros | B | 2,791 | Cnephaeus serotinus | B | 2,304 | Rhinolophus ferrumequinum | B | 1,231 |
| Myotis mystacinus | B | 1,191 | Nyctalus leisleri | B | 991 | Barbastella barbastellus | B | 856 |
| Pipistrellus nathusii | B | 700 | Myotis brandtii | B | 346 | Myotis bechsteinii | B | 180 |
| Plecotus austriacus | B | 62 | Myotis alcathoe | B | 28 |
| Species | Group | Species | Group | Species | Group | |||
| Pseudacris regilla | A | 9,380 | Anaxyrus boreas | A | 4,649 | Batrachoseps attenuatus | A | 3,248 |
| Lithobates catesbeianus | A | 1,917 | Aneides lugubris | A | 1,908 | Taricha torosa | A | 1,766 |
| Ensatina eschscholtzii | A | 1,688 | Taricha granulosa | A | 1,072 | Pseudacris cadaverina | A | 681 |
| Batrachoseps major | A | 585 | Batrachoseps nigriventris | A | 500 | Taricha sierrae | A | 470 |
| Dicamptodon ensatus | A | 458 | Dicamptodon tenebrosus | A | 419 | Aneides vagrans | A | 174 |
| Aneides flavipunctatus | A | 173 | Ambystoma gracile | A | 145 | Anaxyrus punctatus | A | 128 |
| Group | Species | Training cells | Range cells |
|---|---|---|---|
| Birds | 1,614 | 5 / 98 / 11,774 | 4 / 13,271 / 524,718 |
| Reptiles | 444 | 5 / 60 / 4,727 | 1 / 3,139 / 70,124 |
| Plants | 286 | 5 / 41 / 4,482 | 3 / 436 / 133,698 |
| Amphibians | 254 | 5 / 62 / 5,069 | 2 / 2,736 / 45,962 |
| Mammals | 228 | 8 / 91 / 8,080 | 16 / 9,249 / 138,575 |
| Dragonflies and damselflies | 106 | 5 / 83 / 2,295 | 19 / 19,615 / 313,016 |
| Name | Type | Dim. | Description | Native | Used |
|---|---|---|---|---|---|
| coords | loc | 2 | longitude and latitude in degrees | – | exact |
| sin cos | loc | 16 | sin and cos of longitude and latitude at 4 frequencies | – | exact |
| rand fourier | loc | 128 | 64 random Fourier features of the 3-D position on the unit sphere | – | exact |
| ERA5 | env | 117 | ERA5 biodiversity climatology (Copernicus CDS), 1979–2018 | 0.5 ∘ | 0.5 ∘ |
| bioclim | env | 20 | WorldClim 2.1 BIO1–19 and elevation, 1970–2000 | 5 arcmin | 5 arcmin |
| CHELSA | env | 19 | CHELSA v2.1 BIO1–19, 1981–2010 | 30 arcsec | 2.5 arcmin |
| Global | UK | California | |||||||||
| Features | Type | Dim | RF | LR | ME | RF | LR | ME | RF | LR | ME |
| coords | loc | 2 | 0.98 | 0.89 | 0.98 † | 0.83 | 0.80 | 0.83 | 0.87 | 0.83 | 0.87 |
| sin cos | loc | 16 | 0.98 | 0.98 | 0.99 | 0.82 | 0.82 | 0.82 | 0.87 | 0.86 | 0.87 |
| rand fourier | loc | 128 | 0.97 | 0.96 | 0.97 | 0.84 | 0.83 | 0.84 | 0.89 | 0.88 | 0.87 |
| ERA5 | env | 117 | 0.96 | 0.96 | 0.95 | 0.83 | 0.84 | 0.81 | 0.90 | 0.88 | 0.87 |
| bioclim | env | 20 | 0.96 | 0.93 | 0.96 | 0.82 | 0.82 | 0.82 | 0.90 | 0.89 | 0.91 |
| Global | UK | California | |||||||||
| Features | Type | Dim | RF | LR | ME | RF | LR | ME | RF | LR | ME |
| coords | loc | 2 | 0.60 | 0.51 | 0.60 † | 0.68 | 0.66 | 0.69 | 0.64 | 0.64 | 0.65 |
| sin cos | loc | 16 | 0.62 | 0.61 | 0.61 | 0.68 | 0.69 | 0.68 | 0.64 | 0.65 | 0.65 |
| rand fourier | loc | 128 | 0.63 | 0.59 | 0.63 | 0.68 | 0.70 | 0.68 | 0.67 | 0.66 | 0.65 |
| ERA5 | env | 117 | 0.68 | 0.64 | 0.65 | 0.69 | 0.72 | 0.67 | 0.68 | 0.67 | 0.63 |
| bioclim | env | 20 | 0.68 | 0.62 | 0.66 | 0.68 | 0.70 | 0.69 | 0.71 | 0.70 | 0.71 |
| Global | UK | California | |||||||||
| Features | Type | Dim | RF | LR | ME | RF | LR | ME | RF | LR | ME |
| coords | loc | 2 | 0.59 | 0.50 | 0.58 † | 0.66 | 0.66 | 0.67 | 0.64 | 0.64 | 0.65 |
| sin cos | loc | 16 | 0.62 | 0.60 | 0.60 | 0.65 | 0.68 | 0.67 | 0.65 | 0.65 | 0.66 |
| rand fourier | loc | 128 | 0.64 | 0.58 | 0.63 | 0.66 | 0.69 | 0.67 | 0.67 | 0.66 | 0.65 |
| ERA5 | env | 117 | 0.68 | 0.62 | 0.63 | 0.67 | 0.71 | 0.66 | 0.68 | 0.67 | 0.64 |
| bioclim | env | 20 | 0.67 | 0.60 | 0.65 | 0.67 | 0.69 | 0.68 | 0.71 | 0.70 | 0.71 |
Geospatial Representation Learning
Geospatial representation learning builds reusable embeddings of places, imagery, maps, mobility, and Earth data for spatial prediction and reasoning.
See all 142 papers