Impact of annual and seasonal precipitation and air 1 ! temperature on gross primary production in Mediterranean 2 ! ecosystems in Europe 3 ! 4 !

Abstract. Mediterranean ecosystems are significant carbon sinks but are also particularly sensitive to climate change but the carbon dynamic in such ecosystem is still not fully understood. An improved understanding of the drivers of the carbon fixation by plants is needed to better predict how such ecosystems will respond to climate change. Here, for the first time, a large dataset collected through the FLUXNET network is used to estimate how the gross primary production (GPP) of different Mediterranean ecosystems was affected by air temperature and precipitation between the years 1996 and 2013. We showed that annual precipitation was not a significant driver of annual GPP. Our results also indicated that seasonal variations of air temperature significantly affected seasonal variations of GPP but without major impact on inter annual variations. Inter-annual variations of GPP seemed largely controlled by the precipitation during early spring (March–April), making this period crucial for the future of Mediterranean ecosystems. Finally, we also observed that the sensitivity of GPP in Mediterranean ecosystems to climate drivers is not ecosystem type dependent.


38!
Mediterranean land ecosystems are of particular interest for research 39! because their outstanding biodiversity is one of the most important after the 40!
The statistical analyses were performed using RStudio (version 0.99.473,121! 2009-2015 RStudio).The impact of annual and seasonal PPT, T and the vegetation 122! type on annual and seasonal GPP was investigated by employing a linear mixed effect 123! model ANOVA.The sites were used as random effect, which enabled us to take 124! potential site-dependency effects into account.Because of non-normality, data were 125! rank-transformed before analysis, as previously done by e.g.Guenet et al. (2014).

126!
We tested seven different subsets (Tab.2, S0-S6).We first investigated if the 127! annual mean PPT, T and/or vegetation type significantly affected the annual mean 128! GPP (Tab.2, S0, case A).Then, we analyzed the annual GPP using bi-monthly mean 129! instead of annual mean PPT and T values (Tab.2, S1-S6, case A).Note that we 130! investigated the impact of PPT and T on the average annual GPP of the subsequent 131! year rather than on the actual year for the ND subset, because in this time of the year 132! the actual climatic factors hardly control the total growing strength of the actual year 133! (Tab.2).In a next step all tests were repeated using the total annual and bi-monthly

Inter-annual GPP variability 149!
Over the selected sites, the vegetation faced a typical Mediterranean climate, 150! with usually hot and dry summers as well as mostly mild and moist winters (Fig. 1).T 151! ranged from -0.1 to 28.4°C (Fig. 1C, bi-monthly averages) and the seasonal PPT from 152! 0 to 11.4 mm d -1 (Fig. 1D, bi-monthly averages).GPP values for shrubs were lowest 153! but show the highest variability across the different sites (Fig. 1A).For trees (ENT, 154! DBT, EBT) the GPP values were rather similar to each other.For grassland only two 155! suitable sites were available (Tab. 1, Fig. 1A).156! Surprisingly no significant correlation was found between annual GPP and 157! annual T or annual PPT across sites and years.A general trend over the vegetation 158! types was observed but this was not significant according to the Holm-Bonferroni 159! corrected threshold (Tab.3).Furthermore, the relationships between the applied 160! climatic factors and vegetation types were never found to be significant (Tab.3).We regression models (Fig. 2-5).Nevertheless, annual GPP averages could be explained 164! (significant p-values) by both precipitation during early spring (MA) and air 165! temperature during early winter (ND), when using bi-monthly averages or the sum as 166! explaining variables in the linear mixed effect model (Tab.3, Fig. 2 & 4).167!Furthermore, we found that the annual GPP was not significantly affected by climate 168! conditions in summer (MJ & JA), even though this time period is the hottest and the 169! driest for all sites (Fig. 1).Finally, we noticed that none of the interactions between 170! the explaining variables, vegetation type, T (bi-monthly & annual) and PPT (bi-171!monthly & annual), significantly impacted the annual average of GPP (Tab.3).172! 173!

Intra-annual GPP variability 174!
We observed that GPP was low at the beginning of the year (JF) and increased 175! till MJ (highest median value 6.8 gC m -2 d -1 ), when looking at the bi-monthly 176! distribution of GPP (Fig. 1B).During the summer GPP slowly decreased until the 177! lowest median value in ND (2.2 gC m -2 d -1 ).The highest variability in GPP was 178! observed in JA and dominated by broadleaf trees.
134! sum, instead of mean values for GPP and PPT (Tab.2, S0-S6, case B).As we applied 135! several hypotheses on one single data set, we faced the problem of multiple 136! Biogeosciences Discuss., doi:10.5194/bg-2016-491,2016 Manuscript under review for journal Biogeosciences Published: 15 November 2016 c Author(s) 2016.CC-BY 3.0 License.!7! comparisons minimizing the probability of receiving a Type I error.Accordingly, we 137! corrected the original significance level (p = 0.05) by applying the Holm-Bonferroni 138! method (Holm, 1979).In a last step, we investigated if the PPT and T of specific 139! seasons (bi-monthly time periods) significantly affected the GPP of the corresponding 140! seasons (Tab.2, S1-S6, case C).In the latter case, applying the Holm-Bonferroni 141! method was not necessary as we used an independent data set for every season and 142! subset.143!To interactively explore which predictors provided a good fit, we applied a 144! stepwise regression in all cases, which conducts an automatic stepwise model 145! selection by the AIC (Akaike information criterion).146! 147!
161! Biogeosciences Discuss., doi:10.5194/bg-2016-491,2016 Manuscript under review for journal Biogeosciences Published: 15 November 2016 c Author(s) 2016.CC-BY 3.0 License.!8! did not obtain a clearly relationship between annual and seasonal PPT and annual 162! GPP, or between annual and seasonal T and annual GPP by simply applying linear 163!

179 !Fig. 1 :Fig. 2 :TabFig. 3 :Fig. 4 :TabFig. 5 :!Tab. 3 :
Fig. 2: Seasonal mean PPT versus the annual mean GPP for the different vegetation 495! types and over the different bi-monthly time periods.See Fig. 1 for the abbreviations.496!A simple trend line & R-squared value (including all vegetation types) was added to 497! those plots where a significant p-value was obtained during our statistical tests (see 498! Tab. 3).499! 500!Fig. 3: Seasonal mean PPT versus the seasonal mean GPP for the different vegetation 501! types and over the different bi-monthly periods.See Fig. 1 for the abbreviations.502! Trend lines & R-squared values (including all vegetation types) were added to those 503!plots where a significant p-value was obtained during our statistical tests (see Tab. 3).504! 505! Fig. 4: Seasonal mean T versus the annual mean GPP for the different vegetation 506! types and over the different bi-monthly time periods.See Fig. 1 for the abbreviations.507!A simple trend line & R-squared value (including all vegetation types) was added to 508!