Preprints
https://doi.org/10.5194/egusphere-2024-138
https://doi.org/10.5194/egusphere-2024-138
24 Jan 2024
 | 24 Jan 2024
Status: this preprint is open for discussion.

Random forests with spatial proxies for environmental modelling: opportunities and pitfalls

Carles Milà, Marvin Ludwig, Edzer Pebesma, Cathryn Tonne, and Hanna Meyer

Abstract. Spatial proxies such as coordinates and Euclidean distance fields are often added as predictors in random forest models; however, their suitability in different predictive conditions has not yet been thoroughly assessed. We investigated 1) the conditions under which spatial proxies are suitable, 2) the reasons for such adequacy, and 3) how proxy suitability can be assessed using cross-validation.

In a simulation and two case studies, we found that adding spatial proxies improved model performance when both residual spatial autocorrelation, and regularly or randomly-distributed training samples, were present. Otherwise, inclusion of proxies was neutral or counterproductive and resulted in feature extrapolation for clustered samples. Random k-fold cross-validation systematically favoured models with spatial proxies even when not appropriate.

As the benefits of spatial proxies are not universal, we recommend using spatial exploratory and validation analyses to determine their suitability, and considering alternative inherently spatial RF-GLS models.

Carles Milà, Marvin Ludwig, Edzer Pebesma, Cathryn Tonne, and Hanna Meyer

Status: open (until 20 Mar 2024)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on egusphere-2024-138', Anonymous Referee #1, 07 Feb 2024 reply
  • RC2: 'Comment on egusphere-2024-138', Carsten F. Dormann, 07 Feb 2024 reply
Carles Milà, Marvin Ludwig, Edzer Pebesma, Cathryn Tonne, and Hanna Meyer

Data sets

Code and data for "Random forests with spatial proxies for environmental modelling: opportunities and pitfalls" Carles Milà https://zenodo.org/records/10495235

Carles Milà, Marvin Ludwig, Edzer Pebesma, Cathryn Tonne, and Hanna Meyer

Viewed

Total article views: 228 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
172 46 10 228 1 0
  • HTML: 172
  • PDF: 46
  • XML: 10
  • Total: 228
  • BibTeX: 1
  • EndNote: 0
Views and downloads (calculated since 24 Jan 2024)
Cumulative views and downloads (calculated since 24 Jan 2024)

Viewed (geographical distribution)

Total article views: 227 (including HTML, PDF, and XML) Thereof 227 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 21 Feb 2024
Download
Short summary
Spatial proxies such as coordinates and distances are often included as predictors in random forest models for predictive mapping. In a simulation and two case studies, we investigated under which conditions this is appropriate. We found that spatial proxies are not always beneficial and thus we conclude that they should not be used as default approach without careful consideration. We also give insights on the reasons behind their suitability, how to detect it, and potential alternatives.