DECODAGE, 25 avril 2023, 13h00

Cette semaine, Jeudi 25 Avril entre 13h et 14h, nous aurons le plaisir d’assister au DECODage de Nicole Augustin (Reader à l’Université de Edinburgh, School of Mathematics), qui sera en visite à Ifremer Nantes. Nicole présentera ses recherches sur la modélisation des facteurs impactant les forêts allemandes afin d’aider à leur gestion dans le cadre du changement climatique.
La présentation sera en anglais, depuis Nantes. Plus de détails ci-dessous !

Understanding forest damage in Germany: Finding key drivers to help with future forest conversion of climate sensitive stands

Abstract:
Recently climate change has contributed to the decline in forest health, and yearly European forest health monitoring data are increasingly being used to investigate the effects of climate change on forests in order to decide on forest management strategies for mitigation. Forests in Germany have been badly affected and climate change now appears to be the major cause of defoliation (Eickenscheidt et al., 2019; Augustin et al.,2009). Thus, large scale forest conversions to more mixed forests with drought and heat resistant species are planned in some areas of Germany. This talk will cover the statistical aspects of a  modelling project which has been informing decisions regarding this future forest conversion. Model selection is a challenge because of spatial confounding and the large number of correlated time varying environmental predictors.  In addition there are computational challenges due to the large number of parameters and large sample sizes. A generalized additive mixed model is used for estimating spatio-temporal trends of defoliation, an indicator for tree health. Defoliation is modelled as a function of  site characteristics (topography, soil and climate) with the aim of identifying the main factors associated with tree damage. The minimal model contains a space-time smoother and an AR1 process for temporal correlation. Initially we exclude redundant predictors in the large set of more than 70 predictors in discussion with experts. To eliminate predictors with negligible effects in the remaining set of predictors we use stability selection. Variable selection  using integrated backward selection is carried out repeatedly with resampled data yielding selection inclusion frequencies. The final set of predictors are the predictors with selection inclusion frequencies above a certain threshold.  To assess predictive performance we use a cross-validation approach which takes the spatio-temporal dependence structure of the data into account.

 

References:

Augustin, N., Musio, M., von Wilpert, K., Kublin, E., Wood, S., and Schumacher, M. (2009). Modelling spatio-temporal forest health monitoring data. Journal of the American Statistical Society, 104(487):899–911.

Eickenscheidt, N., Augustin, N. H., and Wellbrock, N. (2019). Spatio-temporal modelling of forest monitoring data: Modelling German tree defoliation data collected between 1989 and 2015 for trend estimation and survey grid examination using gamms. iForest Biogeosciences and Forestry, 12:338–348.

 

Contact : Robin Faillettaz - robin.faillettaz@ifremer.fr et Olivier Dézerald - olivier.dezerald@inrae.fr

Planning prévisionnel des DECODages : 

- 22/05/2024 : Chalikakis – Fourno (chercheurs Avignon Université)
- 30/05/2024 : Thomas Outrequin (Doc DECOD)
- 06/06/2024 : Pierre-Yves Hernvann (PostDoc DECOD)