Authors: Bubnicki, Jakub; Angelstam, Per;Svensson, Johan;Mikusiński, Grzegorz;Jonsson, Bengt Gunnar;

High Conservation Value Forests (HCVF) play an especially important role as core biodiversity areas and functional components of Green infrastructure (GI), providing habitats to a large number of species and multiple well-recognized ecosystem-level services. However, identification of HCVF their importance for GI prioritization strategies, usually requires costly and time-consuming field surveys. In this work we applied a predictive modelling approach to scan all forest land in Sweden (divided into 5 ecoregions) for the (potential) occurrence of HCVF. We used Random Forest (RF) machine learning algorithm and available high-resolution (10x10m) wall-to-wall spatial datasets describing landscape configuration, topography, forest structural properties and various socio-economic factors. We trained our model and tested its performance using a country-wide HCVF inventory database. The final RF models generated high-accuracy predictions with the results of the 10-fold spatial cross-validation indicating a good predictive capabilities (ROC AUC in a range of 0.86 - 0.88 for all ecoregions). Using a comprehensive set of independent validation spatial datasets we confirmed that the predicted high probabilities of HCVF occurrence actually represent forests with high conservation values.