Using machine learning on tree‐ring data to determine the geographical provenance of historical construction timbers

Dendroclimatology offers the unique opportunity to reconstruct past climate at annual resolution and wood from historical buildings can be used extend such information back in time up several millennia. However, varying often unclear origin of timbers affects sensitivity individual tree-ring samples. Here, we compare width density 143 living larch (Larix decidua Mill.) trees seven sites along an elevational transect 1400 2200 m asl 99 series parametrize state-of-the-art classification models for European Alps. To achieve geographical provenance series, nine different supervised machine learning algorithms are trained tested their capability solve our problem. Based on this assessment, consider a density-based width-based dataset model building. For each these datasets, general not species-related larch-specific including cyclic budmoth influence built. From algorithms, Extreme Gradient Boosting showed best performance. The outperform ring-width with reaching highest skill (f1 score = 0.8). performance metrics reveal that also performs within particularly above 2000 asl, which show temperature sensitivities. application specific allows assigned high confidence particular elevation valley. procedure applied other studies using multiple tree growth characteristics. novel approach building based features omit common period between reference data finding relict will therefore help improve millennium-length reconstructions.