Modelling wood moisture content in outdoor conditions from measured data

Sustainable use of wood requires an understanding of expected service life, particularly when the material is exposed to outdoor conditions and, thus, fungal decay. Since moisture is the primary vector for fungal decay, accurate moisture prediction is a key component in service life assessment. For this purpose, the present study leverages existing measured data for linear regression of in-field moisture conditions of different wood species against climate parameters. Predictors of precipitation, relative humidity, and temperature were used in a finite distributed lag model to account for present and previous weather records. Issues of collinearity were addressed by ridge regression. The resulting model was, in general, able to describe the important features of different wood species. However, large errors were observed in certain periods, and it was hypothesized that these were related to thawing. Nevertheless, the results encourage additional effort into data-driven modelling of moisture content from measured data, and it is believed that non-linear models such as random forests and neural networks will be able to describe additional features and, in doing so, reduce the error. The study contributes to the ongoing efforts in developing effective, user-friendly, and open-source tools for performance-based service life assessment of wood. By improving our understanding of moisture content prediction in different softwoods, this research aims to enhance the reliability and sustainability of wood as a construction material.