Increasing atmospheric CO2 concentrations correlate with declining nutritional status of European forests

The drivers of global change, including increases in atmospheric CO2 concentrations, N and S deposition, and climate change, likely affect the nutritional status of forests. Here we show forest foliar concentrations of N, P, K, S and Mg decreased significantly in Europe by 5%, 11%, 8%, 6% and 7%, respectively during the last three decades. The decrease in nutritional status was especially large in Mediterranean and temperate forests. Increasing atmospheric CO2 concentration was well correlated with the decreases in N, P, K, Mg, S concentrations and the increase of N:P ratio. Regional analyses indicated that increases in some foliar nutrient concentrations such as N, S and Ca in northern Europe occurred associated with increasingly favourable conditions of mean annual precipitation and temperature. Crucial changes in forest health, structure, functioning and services, including negative feedbacks on C capture can be expected if these trends are not reversed.

and southern (c) Europe. See Table S1 for detailed results of the model lme (foliar variable ~ year, random=~1|country/plot/species, data=dades, method="REML"). d) Geographical distribution of the annual rate of variation for N:P. The pixel estimations are based on Neural Networks using 80% of the trees with more than 5 measurements for training and 20% for validation. We replicated the process 1000 times and averaged the results. were adjusted to the same mean to remove forest-specific variability. See Table S1 for detailed results of the model lme (foliar variable ~ year, random=~1|country/plot/species, data=dades, method="REML").  Table S1 for detailed results of the model lme (foliar variable ~ year, random=~1|country/plot/species, data=dades, method="REML").  Table S1 for detailed results of the model lme (foliar variable ~ year, random=~1|country/plot/species, data=dades, method="REML").  Table S1 for detailed results of the model lme (foliar variable ~ year, random=~1|country/plot/species, data=dades, method="REML").

Supplementary Table 5. Squared Mahalanobis distances (M) among the elementomes of the species sampled in 1990-2005 and 2005-2016 in southern
Europe, with foliar N, P, S, Ca, Mg and K concentrations and N:P ratios as variables (Fig. 5).

Models to assess the temporal contributions of the environmental factors to the temporal trends of foliar nutrient concentrations at the site level
Saturated formulation of the models: foliar nutrient ~ (mean annual precipitation + precipitation annual anomaly + CO 2 )^2 + (mean annual temperature + temperature annual anomaly + CO 2 )^2 + (mean oxidised N deposition + oxidised N deposition annual anomaly + CO 2 )^2 + (mean reduced N deposition + reduced N deposition annual anomaly + CO 2 )^2 + (mean S deposition + S deposition annual anomaly + CO 2 )^2 + (annual anomalies of CO 2 + annual anomalies of temperature + annual anomalies of precipitation + annual anomalies of oxidised N deposition + annual anomalies of reduced N deposition + annual anomalies of S deposition)^2 + (mean annual temperature + mean annual precipitation + mean annual oxidised N deposition + mean annual reduced N deposition + mean annual of S deposition)^2, where ^2 indicates a firstorder interaction of the variables within the brackets.

Selection of models that best accounted for the foliar nutrient concentrations at the tree level
We selected the models conducted at the tree level that best accounted for the foliar nutrient We also constructed different subset models to identify patterns at the latitudinal or species level. We used species and tree ID nested inside plot and country as random factors for the continental analysis and subsets by latitude and used only tree ID nested inside plot and country as a random factor for subsets including only one species. All analyses were performed in R using the MuMIn and lme4packages.
We calculated the importance of each predictor as the sum of the AIC-based weight of the models that included the predictor.

Interannual variability in foliar concentration data
The trends reported in this study were found despite the strong interannual variability in foliar concentrations clearly shown in Figures, for example see figure 1. Even continuously measured sites present a relatively large interannual variability in their foliar concentrations. There are many factors contributing to these temporal variability, among which we can find: i) measurement error (we will never find two leaves with the exact same concentration of elements) + the systematic error of the sampling and analytical techniques, and ii) environmental variability can drastically affect element composition from year to year (e.g., late frosts, drought, storms or pests), iii) the inclusion of different number of sites within each year.
4. PRISMA flow diagrams of the meta-analysis for N, P and N:P