Fig. 5

Modelling flow implemented for each tree species. In order to isolate the temporal signal in growth, other factors must be first controlled including growing stock, stage of stand development and average/permanent site conditions. While indicators exist for the two formers, the latter cannot be controlled easily, e.g. by site index that changes over time, and environmental indicators must be used. Step 1 is intended for their selection, using a PLS regression approach. Once these environmental predictors are identified (X _i), it turns possible to resample NFI plots from the older NFI sampling sketch where all administrative units were not sampled at the same time, resulting in a sampling design imbalance likely to cause false temporal effects. Thus, NFI plots in the older NFI design are bootstrapped to get samples in which the predictors X _i match at any date (pair of years) the whole country distribution of these. A PCA is further implemented to reduce the initial dimensionality of the X _i and only match 2 distributions defined by PCA axes. Then, step 2 can be implemented where all factors are controlled by relevant proxies and the samples are balanced over time. In this step, an appropriate mathematical approximation for temporal growth changes can be selected from bootstrap tests, and its uncertainty can be measured (Fig. 2 of the manuscript)