Comparing the effectiveness of random forest and generalized linear models in predicting ungulate browsing impact on Kyiv Polissya’s young pine forests

Autorzy

• Natalia Vysotska 1 Estonian University of Life Sciences, Institute of Forestry and Engineering
  Fr. R. Kreutzwaldi 5, Tartu, Estonia
  e-mail: natalia.vysotska@emu.ee
  2 Ukrainian Research Institute of Forestry and Forest Melioration
  Hryhoriia Skovorody 86, Kharkiv, Ukraine
• Oleksandr Khromuliak Dniprovsko-Teterivske State Forest and Hunting Enterprise
  Kyiv, Ukraine
• Oleksandr Borysenko 1 University of Tartu, Department of Remote Sensing
  Tartu Observatory, Toravere, Estonia
  2 National Aerospace University ‘Kharkiv Aviation Institute’ Department of Geoinformation Technologies and Space Monitoring of the Earth
  Vadim Manko 17, Kharkiv, Ukraine
• Maksym Rumiantsev Ukrainian Research Institute of Forestry and Forest Melioration
  Hryhoriia Skovorody 86, Kharkiv, Ukraine
• Iryna Yashchuk State Enterprise ‘Klavdiivska Forest Research Station’
  Kyiv, Ukraine
• Oleksandr Kipran Ukrainian Research Institute of Forestry and Forest Melioration
  Hryhoriia Skovorody 86, Kharkiv, Ukraine

Abstract

Effective management of forest ecosystems requires accurate predictions of damage by ungulates, a challenge particularly acute in the Kyiv Polissya. This study aims to identify key drivers of ungulate browsing intensity and compare the effectiveness of the random forest model (RFM) and generalized linear model (GLM) in forecasting damage to young forests. We analysed field data from three experimental enterprises in the Kyiv region, covering a combined area of 71.4 thousand hectares and involving 275 experimental plots. The study identified ungulate population density as the most influential factor affecting browsing intensity, surpassing variables such as tree age, tree species ratio and forest type. In comparing models, RFM demonstrated superior predictive accuracy over GLM, highlighting its effectiveness in forecasting damage to young forests. The study highlights how machine learning enhances the accuracy of ecological predictions and underscores the significance of selecting variables thoughtfully during model development. The findings point to the need for flexible forest management strategies focused on regulating wild ungulate populations and protecting young forests.

Keywords

• ungulate density
• machine learning
• tree species composition
• browsing
• tree damage