High stomatal ozone (O

Tropospheric ozone (O

In recent years, many studies have been conducted to investigate the
mechanisms underlying the O

To assess the impact of O

While high stomatal O

In this study, we investigated the effect of O

The study area consisted of a 2 ha Scots pine stand in a 150 ha
coniferous and deciduous forest named “De Inslag”, situated in Brasschaat
(

The pine stand was planted in 1929 (Neirynck et al., 2008). Until the autumn
of 1999, when the forest was thinned, the tree density amounted to
542 trees ha

The stand is part of the ICP Forests Level II and the Fluxnet CarboEurope-IP
networks and is equipped with a 41 m tall instrumentation tower.
Meteorological measurements and measurements of ecosystem CO

The study covered the period 1998–2013, with the years 1999 and 2003 excluded due to poor data quality or coverage.

Air temperature (

The O

A continuous time series with daily LAI values was reconstructed for the pine
stand based on the historical data. The general approach was to keep the
seasonal pattern measured in 2009 by Op de Beeck et al. (2010) fixed for each
year and to scale it year per year to the seasonal maximum LAI
(LAI

Gross primary productivity (

The measurements of stomatal conductance to H

The stomatal O

The aerodynamic resistance was calculated following Grünhage (2002)
with

for unstable atmospheric stratification (

for stable atmospheric stratification (

and for neutral atmospheric stratification (

The canopy resistance was calculated from a stomatal resistance
(

The non-stomatal resistance

Total and stomatal O

The multiplicative stomatal model was parameterised and validated against the
dataset of

The parameterised model was then tested against the validation dataset.
The model performance was evaluated with the linear regression

We adopted a modelling approach to detect possible O

We used a feed-forward back propagation artificial neural
network (ANN) as a GPP model in Matlab (Matlab, Natick, MA, USA). The
ANN contained 10 nodes organised in 1 layer, which came out as the best
performing network after comparing networks containing different numbers of
nodes and/or layers (data not shown). The default settings of the Matlab
Neural Network Toolbox were used. A normalisation process was applied for
training and testing the data; data were scaled to [

To obtain an O

The model overestimation of daily GPP was evaluated (1) from the linear
regression on the data of the measured versus the modelled GPP for the days on which
an O

The fingerprint of air temperature (

The time series of the weekly total precipitation and mean soil water potential (SWP). The precipitation and SWP data are averaged over the period 1998–2013. The error bars represent the 95 % confidence intervals.

Since it may take some time for trees to repair the damage to the photosynthetic apparatus
induced by O

High O

One of the assumptions in our approach is that the O

All statistics were performed with R version 3.2.3 (The R Project for Statistical Computing Core
Team, Vienna, Austria
2015) at a
significance level of

Figure 1 shows a fingerprint of the multi-annual average diel and seasonal
patterns of the main meteorological variables

The seasonal course of the LAI for each of the 14 growing seasons used in this study.

The optimised parameter values of the model are presented in Table 1. The
different statistics to evaluate the model performance are presented in
Table 2, and this is for both the parameterisation and the validation dataset. For
the parameterisation dataset, the measured data were plotted against modelled

The measured versus the modelled stomatal conductance (

The optimised parameter values of the multiplicative stomatal model.

Figure 5 shows the scatter plots of the measured

The measured stomatal conductance (

The average daily O

Figure 6 shows the frequency distributions of

Histograms of the meteorological variables for the training dataset
(red) and the high O

All parameters in the GPP model were ranked according to their contribution
to GPP prediction (Table 3). Global radiation is the most important parameter
in defining GPP, with a mean square error (MSE) of
37 500.81 mol m

To test for carry-over O

Performance statistics for the multiplicative stomatal model: mean
bias (MB), relative mean error (RME), systematic and unsystematic root mean
square error (RMSE

Figure 8 shows the measured versus the modelled daily GPP for the model trained
without the days with the highest stomatal O

The measured GPP plotted as a function of the modelled GPP for two
different datasets:

The ranking of the parameters defining GPP in the ANN by replacing each
input variable with a random permutation of its values.

No statistically significant correlations were found between the model
residuals of growing season GPP and total stomatal O

The measured versus the modelled gross primary productivity (GPP) for days
used for model training and testing

The residuals of growing season gross primary productivity (GPP)
as a function of

All statistics shown in Table 2 clearly indicated that the fitted
multiplicative stomatal model performed well. For both the parameterisation and
the validation datasets, the model explained 72 % of the variance in

As explained in the mapping manual of the Convention on Long-Range
Transboundary Air Pollution (CLRTAP), Scots pine is the representative
species to assess the risk of O

The stomatal O

The low relative stomatal O

A comparison of the frequency distributions of radiation, temperature, and
VPD between the training dataset and the dataset with the days on which we
expected an O

The statistical tests on the datasets of the measured and the modelled GPP did
not reveal a statistically significant model overestimation of daily GPP for
the days on which we assumed an O

Some earlier studies have investigated the effect of O

AOT40 is, at present, the European standard for forest protection (EEA, 2014) with a critical level of 5000 ppb h, equivalent to a growth reduction of 5 % (CLRTAP, 2015). In this study on Scots pine in Brasschaat, this value was far exceeded in all years (Fig. 9), yet no negative effect on GPP was observed in years with higher AOT40 values.

POD

Notwithstanding the absence of a statistically significant positive
correlation between GPP residuals and both AOT40 and POD

Figure S4 shows a negative relationship between the measured growing season GPP
and the O

Overall, no significant O

The lack of detected O

We parameterised a multiplicative stomatal model for a Scots pine stand in
Brasschaat. This species- and site-specific parameterised model performed
very well. With this model embedded in a resistance scheme, stomatal O

Data to this paper can be found in the Supplement.

This study investigates O

Gross primary productivity (

In this work, the multiplicative stomatal model described by Jarvis (1976) is modified specifically for the Scots pine stand in Brasschaat. The basic model is explained below.

The stomatal conductance to O

Phenology modifies

The stomatal response to PAR is described by a rectangular hyperbola, where

In order to test how well the modified stomatal model performed, several model statistics were calculated. These model statistics are explained below.

The mean bias (MB) is the mean difference between the simulations (

Lore T. Verryckt, Maarten Op de Beeck, Bert Gielen,
Marilyn Roland, and Ivan A. Janssens designed the study. Johan Neirynck provided the
O

The authors declare that they have no conflict of interest.

The measurements for this work were funded by the Hercules Foundation through the support of the Brasschaat ICOS ecosystem station. Ivan A. Janssens acknowledges support from the European Research Council Synergy Grant; ERC-2013-SyG-610028 IMBALANCE-P. Edited by: I. Trebs Reviewed by: five anonymous referees