Digitala Vetenskapliga Arkivet

Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Random forest with generalized least squares for spatial modeling: A case study on lake pH in Sweden
Uppsala University, Disciplinary Domain of Humanities and Social Sciences, Faculty of Social Sciences, Department of Statistics. SLU.
2025 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Environmental monitoring increasingly relies on flexible, data‑driven methods to model complex spatial processes. Random Forest (RF) algorithms are widely used for capturing nonlinear relationships among numerous predictors, but they assume independent observations and thus fail to account for spatial autocorrelation. Random Forest–Generalized Least Squares (RF–GLS) addresses this limitation by modeling a Gaussian-process covariance structure into each fit. This thesis evaluates the performance of Random Forest Generalized Least Squares (RF–GLS) in comparison to standard Random Forest (RF) through simulation experiments and a real-world case study of lake-water pH across Sweden. The study aims to identify the order of importance of various variables influencing pH levels, providing insights into the most significant environmental factors affecting lake-water acidity. Simulation results show that RF–GLS consistently outperforms standard RF when spatial autocorrelation is strong, but this advantage diminishes as the nugget effect (non-spatial error) becomes comparable in magnitude to the spatial variance component. In the case study, both methods perform similarly in predictive accuracy across four environmentally distinct subregions. Permutation-based importance analysis identifies calcium concentration as the most influential predictor of pH, while geographic variables contribute minimally. These findings indicate that, in practice, RF and RF–GLS yield broadly similar results in both prediction and variable importance, even when moderate spatial dependence is present.

Place, publisher, year, edition, pages
2025. , p. 43
National Category
Probability Theory and Statistics
Identifiers
URN: urn:nbn:se:uu:diva-562688OAI: oai:DiVA.org:uu-562688DiVA, id: diva2:1979462
Educational program
Master Programme in Statistics
Supervisors
Examiners
Available from: 2025-07-01 Created: 2025-06-30 Last updated: 2025-07-01Bibliographically approved

Open Access in DiVA

fulltext(4057 kB)213 downloads
File information
File name FULLTEXT01.pdfFile size 4057 kBChecksum SHA-512
76c2d99a6d8d1da20edacc8a667ea5047a3226630ad640bb3995f8ba81137dcceadad36577b18b23b2c82f2ea2cab8515db8f2ea2af5f5426c381f1d891aabdc
Type fulltextMimetype application/pdf

By organisation
Department of Statistics
Probability Theory and Statistics

Search outside of DiVA

GoogleGoogle Scholar
Total: 217 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 263 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf