Effects of climate change in the European croplands and grasslands: productivity, GHG balance and soil carbon storage

. 10 The knowledge of the effects of climate change on agro-ecosystems is fundamental to identify local actions aimed to maintain productivity and reduce environmental issues. This study investigates the effects of climate perturbation on the European crop and grassland production systems, combining the finding from two biogeochemical models. Accurate and high-resolution management and pedoclimatic data has been employed. Results has been verified for the period 1978-2004 (historical period) and projected until 2099 with two divergent intensities IPPC’s climate projections, RCP4.5 and RCP8.5. We provided a 15 detailed overview on productivity and the impacts on management (sowing dates, water demand, nitrogen use efficiency). Biogenic GHG budgets (N 2 O, CH 4 , CO 2 ) were calculated, including an assessment of their sensitivity to the leading drivers, and the compilation of a net carbon budget over production systems. Results confirmed that a significant reduction of productivity is expected during 2050-2099, caused by the shortening of the length of the plant growing cycle associated to the rising temperatures. This effect was more pronounced for the more pessimistic climate scenario (-13 % for croplands and -7.7 20 % for grasslands) and for Mediterranean regions, confirming a regionally distributed impact of climate change. Non-CO 2 GHG emissions were triggered by rising air temperatures and increased exponentially over the century, being often higher than the CO 2 accumulation of the explored agro-ecosystems, which acted as potential C sinks. Emission factor for N 2 O was 1.82 ± 0.07 % during the historical period, rising up to 2.05 ± 0.11 % for both climate projections. The biomass removal (crop yield, residues exports, mowing and animal intake) converted croplands and grasslands into net C sources (236 ± 107 Tg CO 2 eq y -1 25 in the historical period), increasing of more than 20 % during the climate projections. Nonetheless, crop residues restitution demonstrates to be a potential management strategy to overturn the C balance. Although with a marked latitudinal gradient, water demand will double over the next few decades in the European croplands, whereas the benefit in terms of yield will not contribute substantially to balance the C losses due to climate perturbation.


Introduction 30
Agriculture is facing a major challenge to meet growing food demand while limiting soil degradation, air and water pollution, and adapting to climate change impacts (Chaudhary et al., 2018;Olesen, 2017). Agricultural sector is the main source of non-CO2 anthropogenic greenhouse gases (GHG) and is responsible for 78.6 % of nitrous oxide (N2O) and 39.1 % of methane (CH4) emissions worldwide (IPCC, 2018). Agricultural practice, which directly affect soil, plant and atmosphere, represent a strategic lever to counteract climate change by mitigating GHG emission and fostering soil C storage (Chabbi et al., 2017, 35 Smith et al., 2008 achieving long-term (i.e. 2100) climate objectives (Fuss et al., 2016;Minasny et al., 2017;Smith et al., 2013).
Evaluating the impacts of climate on agricultural productions at local, regional and global scales is still a challenge nowadays (Fitton et al., 2019;Olesen and Bindi, 2002). The main source of uncertainty come from the representation of agroecosystems in models' framework, or from the approaches used to upscale data network and local experiments to regional scales (Ewert 40 et al., 2011;Hansen and Jones, 2000;Tubiello et al., 2007). Notwithstanding that, is commonly recognised that a decrease in crop yields is expected towards the mid and the end of the century, with reductions extending to more than 10 % in some region of the world (Challinor et al., 2014). Decline in productivity is likely to be combined with an increase of the interannual yield variability due to climate extremes (Dono et al., 2016), and with a strong latitudinal gradient (Rosenzweig et al., 2013).
In the northern hemisphere, which will benefit from the lengthening of the growing season, milder temperatures and wet 45 conditions in the next decades, crop and grassland productions are expected to rise (Yang et al., 2015). Conversely, lower latitudes are going to face a rise of drought frequencies with a decline of winter rainfall, accompanied by a potential decline in productivity (Stagge et al., 2017). This geographical gap would lead to an intensification of farming systems in northern regions, as north Europe, to an extensification in the southern regions, as the Mediterranean basin (Olesen and Bindi, 2012).
In line with the commitment to the Paris Agreement and the European Green Deal, EU set the objective to cut net GHG 50 emissions by at least 55 % by 2030, compared to 1990 levels. In addition, EU aims to become climate neutral by 2050 (EC, Soil C and N dynamics in the ploughed layer are simulated by means of the NCSOIL model (Molina et al., 1983;Nicolardot et al., 1994), which is a nested module in CERES-EGC. NCSOIL compute nitrification, immobilisation and mineralisation of N, the decomposition of soil organic matter (SOM) after incorporation of crop residues and SOM formation. The module 100 works with a series of specific pools, three pools for crop residues (easily fermentable carbohydrates, cellulose and lignin) and four endogenous pools (zymogenous and microbial biomass, active and passive humus), where CO2 is released from the decomposition of each pool. N uptake by plants is calculated through a specific supply/demand scheme depending on mineral nitrogen availability and root length density. CERES-EGC includes the model NOE (Hénault et al., 2005) for simulating N2O emissions from denitrification and nitrification processes in the topsoil (0-20 cm depth). Denitrification and nitrification are 105 computed from a soil-specific potential rate limited by unitless factors related to soil water content, soil temperature and substrate content (nitrates, NO3, and ammonium, NH4, for denitrification and nitrification, respectively). Plant growth is simulated according to the crop specific genetic potential and the photosynthetically active solar radiation absorbed by the canopy. Potential dry matter production is constrained by air temperatures, soil water availability and N deficit.
PaSim (Pasture Simulation model) is a biogeochemical process-based model able to simulate C, N and water dynamics in the 110 soil-plant-animal-atmosphere grassland system . Five interacting sub-models of soil biology and physics, microclimate, vegetation and grazing herbivores constitute the model structure. The model runs on daily (or hourly) time step and inputs require soil property data, management and meteorological characteristics (global solar radiation, minimum and maximum air temperature, relative humidity, wind speed, precipitation and atmospheric CO2 concentration). The soil is described in six sublayers allowing to parametrise different soil depths with site-specific soil physical and chemical 115 characteristics. Management includes grazing, mowing, N fertilisation. Grazing is considered as a dairy or suckling system managed by grazing periods with specific stocking density and live weight. Indoor periods are not simulated. Vegetation cover is considered as a homogeneous cover with a fixed legume fraction. The vegetation cover comprises root system and three shoot compartments (laminae, sheaths and stems, ears) divided into age classes. Soil C dynamics (based on CENTURY model; Parton et al., 1994) are computed in five pools, a structural and a metabolic pool for fresh organic carbon (plant residues), and 120 an active, a slow and a passive pool for the microbial processed organic carbon. Photosynthetic C is allocated in plant (root and shoot) and can be lost as CO2 by ecosystem respiration and as CH4 through enteric fermentation.
Soil N inputs are represented by atmospheric N deposition, symbiotic N2 fixation, mineral or organic fertilisation, animal faeces and urine. These inputs, together with the nitrogen mineralised from the organic carbon pools, constitute the mineral N pool. N availability for plants is reduced by losses via processes of immobilization, NO3 leaching, NH3 volatilization, 125 nitrification and denitrification processes. N2O emission from nitrification and denitrification depends on substrate availability (NO3 or NH4). These emissions are modulated by factors controlling the effects of soil temperature and water content.
Furthermore, the release of N2O produced into the soil toward the atmosphere is calculated with a resistance model in the rooting zone and plant canopy (Schmid et al., 2001).
CERES-EGC and PaSim were selected as the most suitable models for a spatialized assessment since they have been calibrated 130 and evaluated in different worldwide (Brilli et al., 2017;Ehrhardt et al., 2018;Sándor et al., 2018) and European conditions, i.e. France, Denmark, Germany, Italy, Sweden, UK for CERES-EGC (Rolland et al., 2008;Lehuger et al., 2009;Wattenbach et al., 2010, Drouet et al., 2011Lehuger et al., 2011;Goglio et al., 2013;Ferrara et al. 2021;Haas et al., 2021) and France, Germany, Hungary, Ireland, Italy, Portugal, Spain, The Netherlands, UK for PaSim (Lawton et al., 2006 ;Calanca et al., 2007;Gottschalk et al., 2007;Vuichard et al. 2007;Ma et al., 2015;Sándor et al., 2016). These models can globally simulate a 135 number of crops and rotations, mown or grazed grasslands, and the effects of management practices on plant-soil-atmosphere interactions. Besides, they are able to simulate GHG emissions and the carbon budget at the field scale through the C assimilated from the photosynthesis, C emitted into the atmosphere from autotrophic and heterotrophic respirations, C recycled into the system (dung, plant residues) or introduced from external sources (fertilisers, soil improvers), and the C exported from the system by production activity. CO2 fertilisation was not simulated for croplands (see S.4 in the supplementary material). 140 Also, the two models used in this study do not represent potential impacts of air pollution, pest and disease effects on plant production.

Climate data
Historical and climate projections data were used in this study to analyse the likely effect on GHG, productions and soil C 145 stocks in European production systems. We selected two of the four climate scenarios, or "Representative Concentration Pathways" (RCPs) adopted by the IPCC for the fifth Assessment Report (AR5) (IPCC, 2013), an intermediate scenario, RCP4.5, and a pessimistic one, RCP8.5.
Climate data were provided by the Earth System model HadGEM2-ES (Collins et al., 2011) and were downscaled to a horizontal grid with 0.5° side resolution in the framework of the Inter-Sectoral Impact Model Intercomparison Project (ISI-150 MIP; Warszawski et al., 2014). Since the spatial resolution of the climatic data is larger than the size selected for the simulation units (0.25°), 4 adjacent simulation units were subjected to the same meteorological data. Data were not downscaled to maintain data accuracy as much as possible. Data has been shaped for the European surface (29.0° to 71.5° Latitude and -24.0° to 45.5° Longitude). HadGEM2-ES model provided daily values of minimum and maximum air temperature, total precipitation, air specific humidity, shortwave radiation and near surface wind speed, for the period 1951-2099. Based on these 155 data, input variables for each model were assigned. The simulation protocol consists of a historical dataset, from 1978 to 2004, constituted in accordance with HadGEM2-ES model using the historical record of climate forcing factors , and two climate projections RCP4.5 and RCP8.5, from 2005 to 2099.

Soil data
Soil data were obtained from the European Soil Database (ESDB; Hiederer, 2013). The ESDB is composed by 1 km × 1 km 160 raster files containing topsoil (0 to 30 cm) and subsoil (30 cm to maximum soil depth) data of clay, silt, sand, gravel, and soil organic carbon (SOC) content, bulk density and maximum root depth. Soil pH for the topsoil was derived at the same spatial detail from the ESDB dataset provided by Reuter et al. (2008). To define soil characteristics for each spatial simulation unit of 0.25° side, the most recurrent soil was selected, accordingly with their characteristics. Model-specific soil input parameters were calculated on the base of the elementary characteristics (see supplementary material S.1 for details). For both models, a 165 fixed number of six soil layers was established with a thickness defined as a function of the maximum soil depth. Organic soils with SOC content over 30 % were excluded from the simulations (3.4 % of the total simulation units).

Crop data
Crop species and N fertilisation amount for European Union on the 1 km × 1 km grid were provided in the framework of the GHG-Europe project (EU FP7; Wattenbach et al., 2015). These data are based on the statistical crop distribution of Eurostat 170 database (European Statistical Office, 2019a) at regional scale (NUTS2 regions), and the simulation of the CAPRI model (Common Agricultural Policy Regionalized Impact; Britz and Witzke, 2008;see Leip et al., 2008). Nitrogen fertilisation amount and the repartition between mineral and organic forms were also provided at NUTS2 scale.
Crop successions were available for the period 1976 to 2010. We only considered the crop successions from the time interval 1978 to 2010 since the crop species used in the two discarded years (1976)(1977) were never reused over the time series, and 175 represented less than 1 % of the crop successions (i.e., summer cereal mixes without triticale, other cereals including triticale, winter barley, flax, hemp and set aside). The two most frequent crop successions were selected as a reference for each simulation unit. In fact, they cover on average up to 93 % of the total agricultural area of the simulation units (median over all the simulation units with two rotations equal to 100 %). Based on this aggregation, the simulated crops were: summer/spring soft wheat, winter soft wheat, durum wheat, summer/spring barley, grain maize, fodder maize, rapeseed, sunflower, pulses, 180 oats and sugar beet. Crop rotations included also winter rye, rice and potato, which were not explicitly parameterised in CERES-EGC model, and were respectively substituted with specific varieties of soft wheat for rye and rice (the latter in the end was not represented in the rotations) and with sugar beet for potato. To define the crop species in the period 1951 to 2099, primary and secondary successions were replicated for all the years preceding and succeeding the time interval of available data . Furthermore, most adapted and calibrated crop varieties were designated in function of the latitude, based 185 on previous work and modellers' experience by using the CERES-EGC crop database.
Sowing dates were defined for each crop species, for each simulation unit and per year, based on a crop-specific time window, as well as a minimum and a maximum threshold temperature. Crop-specific windows were extracted from the assessments of USDA (1994) and Sacks (2010), selecting the minimum and the maximum typical sowing span over Europe, whereas threshold temperatures were extracted from Steduto et al. (2012). Given their width, time windows were not changed over time. The 190 sowing date was defined as the earliest within the time window, when minimum and maximum temperatures were respectively higher and lower than the thresholds. An additional constraint (no precipitation for three days in a row) was applied to consider farmers practice concerning access to the field. If a suitable sowing date was not identified, a fixed date was imposed in the middle of the time window. Residues were managed based on crop specie, exporting half (50%) of the aboveground cereal straws, 80% of the fodder maize and removing 20% from the residues of all the other crop types (harvesting losses), including 195 grain maize (Scarlat et al., 2019). Typical sowing crop densities were imposed based on Steduto et al. (2012). Fertilisation amounts (kg N ha -1 year -1 ) were defined as the average amounts designed for each of the crops in the most frequent succession in the simulation unit. Splitting and fertilisation dates were established based on crop type and the sowing date, total nitrogen amount and mineral to organic repartition (see supplementary material S.2 for details).
Irrigation was automatically supplied to the simulation unit which are defined as irrigable in the EU for the year 2016; 200 "irrigable" is considered as the area equipped for irrigation greater than 5 % of the utilised agricultural area (Eurostat 2019b).
This share was 36 % of the EU and is mainly in the Mediterranean area, southern France and north-west of France, the Netherland and some regions in Denmark, Germany and the UK. The amount of water was distributed automatically at the rate of 10 mm d -1 when the soil available water content was below 90 %. This means that non-irrigated crops had access to irrigation water. Even if in the coming decades the global irrigated area is not expected to grow further due to water scarcity 205 and limited land (Turral et al., 2011), to account for a possible increase of the irrigable share toward 2100, a management scenario to observe the maximum potential irrigation water demand for today's crops grown in Europe was simulated and discussed. This management is evaluated over the century by the two scenarios i_RCP4.5 and i_RCP8.5.

Grassland and livestock data
Grasslands data considered permanent grassland and rainfed temporary grassland. Nitrogen fertiliser application for European 210 grasslands in a 0.25° side resolution grid was estimated on the base of regional and national statistics (Eurostat) and CAPRI model (Leip et al., 2008). Data were generated combining fertilization managements and nitrogen doses, together with number of mowing events, animal loads, amount of mineral fertilizers and / or organics, and the fraction of leguminous. Mowing dates were defined from temperature using thermal sums (500 °C-days from the first of January) on a base of 5 °C. No cutting was performed before such thermal sums was not obtained. Fertilisation events occurred three days after mowing. Grazing started 215 30 days after the first mowing event and ended either at the end of the year or at the first freezing period of five consecutive days. Livestock were represented in the model only by cattle. Livestock densities (LSU ha -1 ) were obtained from 0.05° side regional statistics (Wint and Robinson, 2007) multiplying the total number of animals per surface unit to 0.8, 0.1, and 0.1 for cattle, sheep and goats, respectively. Finally, LSU density distribution was aggregated to the 0.25° side grid. As for cutting and fertilization, if no thermal sums were reached, then no events were performed. Biomass production is considered as the 220 sum of the grazer intake and the cut biomass. For each grid cell, livestock is only fed by grass (i.e. no external feed is considered). If the amount of daily aboveground biomass is not sufficient to grazing animals, animals are moved from the pasture. In this study we simulate livestock as they contribute to N cycling and since are an important source of nitrogen in grassland, although we do not discuss here their productions.

2.2.5
Models spin-up and computation 225 CERES-EGC and PaSim were first initialised with soil C, along with the chemical and physical soil parameters, taken from the ESDB for the year 2013. Then, for cropland, an equilibrium was set through a spin-up run using the weather period from year 1951 to 1977 assuming that the cultivated area during this period was likely to have been continuously cultivated with the same crop successions. Equilibrium was reached before 1971 for all the pixels with an estimation error lower than 0.1% of the relative variation in the soil C balance in 5 years. For grasslands, we first let derive the simulation for each pixel from 1840 230 based on HadGEM2-ES weather data. Following, transformation rules were applied to move from past towards current management practices, i.e. from 1901-1950, a low intensification management level with no mineral fertilization and cut at 900°C-days were applied. From 1951 to 2010, there was a gradual management intensification up to achieving the target levels (linear increase of quantities, progressive earlier shift of cutting date). In this period, mineral nitrogen fertilization was applied, starting with a low level in 1951. Finally, from 2010 to 2100, constant management according to the protocol come into effect. 235 A total of 86724 run divided in two land uses (8861 for arable, with two climate scenarios, two crop rotations and two irrigation scenarios; 7918 unit for grasslands, with two climate scenarios) were simulated in a dedicated server.

Greenhouse gas exchange and balance
For assessing the net greenhouse gas exchange (NGHGE) of the investigated ecosystems, the contribution of the biogenic GHG (CO2, N2O, CH4) is combined and normalised to grams CO2-equivalents by using the relative global warming potential 240 ( _ ) at the 100-year time horizon (298 for N2O, 25 for CH4 and 1 for CO2; IPCC, 2018), following the approach presented by : The net ecosystem production (NEP) is the amount of organic C available for net ecosystem C storage, export or loss in an ecosystem, in terms of CO2. NEP represents the difference between the gross primary production, or photosynthesis, and the 245 ecosystem respiration, which is the sum of the autotrophic and heterotrophic respirations (HR); ruminant respiration from grasslands ecosystems is not accounted in the HR term. Conventionally, a negative value of NEP stating an uptake of CO2 by the system, whereas a positive value is a release in atmosphere.
The annual net greenhouse gas balance (NGB) is calculated on the base of Ammann et al. (2020) by including the export of C by harvested biomass (crop yield, mowing and animal intake), the export of crop residues and the import of C by manure 250 (organic fertilizers and the excreta from grazers):

! "
As livestock were not grazing all year, their contribution to the carbon balance is represented by the intake of biomass, enteric fermentation (CH4) and C in excreta. Carbon emissions from farm operations (i.e. tractor emissions), erosion and leaching processes, fire or off-farm emissions (i.e. fertiliser manufacture, barns) are not included in the C budget, as well the effects of 255 volatile organic compounds and CH4 emissions from manure and from soil are considered as negligible. Moreover, the C exported from animal production (body mass increase and milk production) is neglected from NGB calculation (e.g. Chang et al., 2015).

Model validation
Simulated crop yields during the historical period ranged between 1.4 and 44.8 t ha -1 (as standard humidity) and were in good agreement with EU statistics reported in the Eurostat database (Eurostat, 2020) for the time span 1978-2004; the time span considered represents the original crop rotation data and complies with the beginning of the climate scenarios). Root Mean Squared Error (RMSE) was equal to 2.24 t ha -1 , Mean Absolute Error (MAE) to 1.32 t ha -1 and the modelling efficiency 265 (Nash and Sutcliff, E) scored 0.96. Simulations with CERES-EGC overestimated the yields for grain maize, wheat, rye, oats, soybean and sunflower, whereas potato, pulses, rapeseed, fodder maize, barley and sugar beet were slightly underestimated.
The relative RMSE (RRMSE) for each crop, individually, ranged from 12.8 to 38.6 % (Table S3). Furthermore, reducing the simulation period to 1994-2004 to limit the effect of the crop annual genetic gain on measured data, the statistics above  Representative data for grassland productions are still scarce at EU-level. Smit et al. (2008) computed the production of permanent grassland (pastures and meadows) across Europe based on national and international statistics for the period 1995-2004. The productivity simulated with PaSim ( Fig. 1b) and aggregated to NUTS2 level (257 regions in this study) shown a significant positive spatial correlation (r = 0.68, p < 0.05) with the statistics reported by Smit et al., (2008), following the 280 environmental stratification of Europe (Metzger et al., 2005). Compared to these statistics, PaSim scored a RMSE of 2.37 t DM ha -1 y -1 , a MAE of 2.04 t DM ha -1 y -1 and a negative E (-0.34). Simulated productivity was generally overestimated in the Mediterranean area (+55 %; representing 16 % of the surface) and eastern Europe (+20 %; representing 25 % of the surface).
The overestimation in these areas is verified also by other modelling interpretation (van Oijen et al., 2014, Chang et al., 2015Chang et al., 2017;Blanke et al., 2018) and is due to the gap between potential (maximum) simulated productivity and real 285 harvest data. A slight underestimation of the simulated productions was recorded for the Atlantic North zone (-15 %; representing 8 % of the surface). Finally, livestock density and distribution were in line with the Eurostat findings at country scale for the period 1995-2004, ranging from 0 to 1.35 LSU ha -1 (mean: 0.34 LSU ha -1 ). Livestock densities were higher in Belgium, the Netherlands, Denmark and Ireland, and in some regions of Germany, France, Italy and Spain, as also reported by Lesschen et al., (2011). Further details regarding grassland productivity are reported in the supplementary material S.3. 290

Effects of climate change scenarios on productive systems
Our results showed increasing cropland and grassland productions in Europe during the historical scenarios (Fig. 2).
Productions were positively correlated with the increasing air temperatures over this period. Mann-Kendall test highlighted a positive linear increase (p << 0.01) in the mean annual maximum air temperature (0.05 °C year -1 ) and minimum air temperature (0.04 °C year -1 ), as well as in solar radiation (0.02 MJ m -2 year -1 ). 295 Crop production in Europe assumed a positive yearly increase during the historical period (18.1 kg DM ha y -1 ; Fig. 2a), which persisted until 2020, reaching 4.6 t DM ha -1 (average 2005-2020). Crop production raised in the first half of the century for both climatic scenarios (+5 % compared to the average of the historical period; Table 1), even if the rate of increase slow over time, especially from year 2020 to 2050. In the second part of the century, crop production remained stable for the scenario RCP4.5 (+2 % compared to the average of the historical period), while a reduction of -6 % is forecasted for the RCP8.5 300 scenario; this decline reached -13 % in the end of the century (period 2080-2099). The extension of irrigation to all European croplands foster crop productions to +10 % in the first half of the century, while in the second part of the century crop productions were sustained only for RCP4.5 and irrigation mitigates the projected decline for RCP8.5 (+2 % compared to historical period). 305 Figure 2: a) Crop yield trends in Europe from 1978 to 2099 with the two climatic scenarios RCP4.5 and RCP8.5, and two irrigation conditions following the irrigable agricultural area in Europe or extending the irrigation to all the arable lands (i_RCP4.5 and i_RCP8.5); all crops confounded. b) Grassland yield reported as the sum of biomass mowed and ruminant intake. 310 Crop production showed a clear trend over latitudes and over time. During the historical period, crops were more productive in low latitudes (< 45°) than in mid latitudes (45°-55°; -25 % compared to low latitudes, p >> 0.05) and higher latitudes (>55°; -46 % compared to low latitudes, p >> 0.05). These gaps were reduced during the climate scenarios (Table S1). In low latitudes yields were comparable with the historical period in the first half of the century, undergoing to severe reductions towards the 315 end of the century (-4 % and -11 % for RCP4.5 and RCP8.5, respectively). Moving to mid and to high latitudes crop productions increased in the first part of the century for both climatic scenarios (from +5 to +12 %). In the second part of the century, productivity was maintained only for mid latitudes in the RCP4.5 (+3 %), whereas declined for the RCP8.5 scenario (-8 %).
High latitudes were characterised by a further increase towards the end of the century (+14 % and +12 % for RCP4.5 and RCP8.5, respectively). 320 The yield of the two most cultivated crops in terms of area in Europe, grain maize and winter soft wheat, were not negatively affected by climate perturbations in first half of the century with the RCP4.5 scenario, while a slight increase is expected in the RCP8.5 for grain maize (+2 %; average 2030-2049) and a decrease for winter soft wheat (-4 %). Drastic reductions are projected for grain maize yield in the end of the century for both climate scenarios (-5 % in RCP4.5 and -19 % for the RCP8.5, average 2080-2099). Conversely, production is expected to increase for winter soft wheat for RCP4.5 (up to +8 %), and a 325 decline (-1 %) for RCP8.5 (Fig. S3a,b). The adoption of irrigation for all European croplands increased the productivity of grain maize compared to the irrigable scenario (+8 % toward mid-century for both scenarios; +13 % and +16 % toward the end of the century for i_RCP4.5 and i_RCP8.5, respectively). On the other hand, small yield increases are expected with the irrigation scenario for winter soft wheat. Compared to the historical period, in the middle of the century there was an average reduction of the growing season of -8 days for grain maize (-12, -5 and +9 days for low, mid and high latitudes, respectively) and -20 days for winter soft wheat (-20, -19 and -6 days for low, mid and high latitudes, respectively). This trend remained constant for RCP4.5 scenario toward 2100, whereas worsened for RCP8.5, with averaged reductions of -27 and -36 days for grain maize and winter wheat, 335 respectively. Severe reductions are expected at mid and low latitudes for grain maize (-34 and -24 days), and at mid and high latitudes for winter soft wheat (-49 and -38 days). The length of the growing cycle for all the crop, except for potato and sugar beet, was reduced of -12 days in the middle of the century and reached -19 days in the second part of the century (Fig. S4).
Conversely, potato and sugar beet shown an extension of the length of the cropping cycle over time in all scenarios, especially during the end of the century. 340 Figure 3: Yield, length of the cropping season and irrigation needed over the cropping cycle for fodder maize (a) and winter soft wheat (b) in the two climatic scenarios RCP4.5 and RCP8.5; figure reports results for the European the irrigable area and extension of the irrigation to all the European arable area (scenarios i_RCP4.5 and i_RCP8.5). 345 Considering the mild climate projections, positive yield increases from 4 to 20 % are expected for durum and soft wheat, soybean, rye and spring wheat for low latitudes and toward the end of the century. On the other hand, grain and fodder maize, potato, barley, sugar beet, pulses and oats, are affected by substantial reductions (from -1 % to -44 %). The extension of irrigation is able to increase yields for the more water demanding crops (grain and fodder maize, sunflower, sugar beet and 350 potato) with increases of more than +10 %. At mid latitudes strong reductions, in the range of -2 to -17 %, are expected for the large part of the main European crops (durum and soft wheat, potato, rapeseed, barley, soybean, spring soft wheat, sugar beet, and sunflower), whereas fodder maize and winter rye were projected to increase (+30 and +9 %, respectively). High latitudes displayed reductions in yields for pulses and barley (-22 % and -11, respectively), and an increase (+7 to over +100 %) for rapeseed, sugar beet, potato grain and fodder maize. The extension of irrigation to all European croplands will not make 355 sensible improvement for mid and high latitudes yields for i_RCP4.5, while a substantial reduction is projected for all the crops in i_RCP8.5. kg N-N2O ha -1 y -1 g N-N2O ha -1 y -1 kg C-CH4 ha -1 y -1 g C-CH4 ha -1 y-1 kg C-CO2 ha -1 y -1 g C-CO2 ha -1 y -1 kg DM ha -1 y -1 g DM ha -1 y -1  (290) -13988 a Yield for croplands; the sum of harvested biomass and animal intake for grasslands.
Irrigation was applied to 93 % of all the simulation units, doubling the volumes needed to fulfil the evapotranspiration deficit (160 mm y -1 in the first half of the century) compared to the historical period (82 mm y -1 ). Then, water volumes needed in the second half of the century are reduced for i_RCP4.5 (114 mm y -1 ) and are slightly increased for i_RCP8.5 (176 mm y -1 ). 365 Compared to the scenario with actual irrigable surface, these volumes increased by more than 2 and 5 times at mid and high latitudes and only by +30 % at low latitudes, indicating that the extension of irrigable areas became an essential to ensure adequate levels of crop production, especially in the Mediterranean regions.
Grassland productivity showed a trend over time similar to croplands ( Fig. 2b; Table 1). Compared to the historical period, grassland productivity slightly increased until 2020 to decline toward the middle of the century, with an average production of 370 5.6 t DM ha -1 (average 2030-2049). Biomass productivity is maintained during the progress of the RCP4.5 scenario, whereas an averaged reduction of about 0.45 t DM ha -1 (-7.7 % compared to the historical period) is expected for the RCP8.5 scenario in the second part of the century. During the historical period, grassland productivity at low latitudes was about 30 % lower compared both to mid and high latitudes, with higher productions concentrated in the north-west Europe. A substantial increase of production was observed toward 2050 both for low latitudes (+9 % for RCP4.5 and +10 % RCP8.5) and high latitudes (+13 375 % and +14 % for RCP4.5 and RCP8.5, respectively; Fig. S3c and Table S1 in the supplementary material). Moving to the end of the century, grass production increased further compared to the historical period, especially for RCP4.5 (+16 % and +22 % for low and high latitudes, respectively), while a less marked increase is expected for RCP8.5 (+6 % and +13 % for low and high latitudes, respectively). At central EU latitudes, characterised by a higher livestock density than low and high latitudes (+42 % and +13 %, respectively), productivity was reduced of -6 % in RCP4.5 and -5 % in RCP8.5 in the middle of the century. 380 This reduction remains constant for the RCP4.5 scenario toward the end of the century and was more pronounced for RCP8.5 (-24 %).

N2O emissions
385 Figure 4: a) N2O emissions (kg N ha -1 y -1 ) for croplands and b) grassland with two climate change scenarios (RCP4.5 and RCP8.5). N2O emissions for croplands consider two irrigation conditions, following the irrigable agricultural area in Europe or extending the irrigation to all the arable lands (i_RCP4.5 and i_RCP8.5).
N2O emissions increased sharply for croplands along the century for both climate scenarios (Fig. 4a). During the historical 390 period, a stable growth of the emission at the rate of 2.2 g N-N2O ha -1 y -1 is observed, with a mean value of 1.44 kg N-N2O ha -1 y -1 (Table 1). This rate decreased to 1.3 g N-N2O ha -1 y -1 in the first half of the century for RCP4.5 scenario, while a rise of 2.9 g N-N2O ha -1 y -1 is forecasted for the RCP8.5 scenario. In the second part of the century, the rate of emission was nearly tripled for RCP4.5 compared to the emission in first half of the century. A strong increase of emissions is observed for RCP8.5, with a rate of 10 g N-N2O ha -1 y -1 , reaching a mean of 2.09 kg N-N2O ha -1 (average 2080-2099). RCP4.5 scenario, instead, 395 reach a total of 1.69 kg N-N2O ha -1 . The extension of irrigation to all European croplands amplified the emission rates in the first half of the century for both i_RCP4.5 and i_RCP8.5, compared with the irrigable scenario (+3 % and +26 %, respectively).
Emission rates decreased in the second part of the century for i_RCP4.5 (-34 %), whereas grown up to +17 % for i_RCP8.5.
Furthermore, the interannual variance of N2O emissions increased from the historical period to the first half of the century (+15 % in both scenarios) and continued for the second part of the century (+41 % and +75 % for RCP4.5 and RCP8.5, 400 respectively). While, the extension of irrigation was able to reduce the interannual variance for both i_RCP4.5 and i_RCP8.5 scenarios (+17 % and +61 %). N2O emissions from grasslands described a similar trend over the years as for croplands (Fig. 4b), characterised by lower rates.
During the historical period, the emission increased at a rate of 2.4 g N-N2O ha -1 y -1 , with a mean value of 0.81 kg N-N2O ha -1 y -1 ( Table 1). The rate raised to about 3.6 g N-N2O ha -1 y -1 during the first half of the century, afterwards the two different 405 climate scenarios shown different trends. RCP4.5 was characterised by a significant reduction of the emission rate to 0.5 g N-N2O ha -1 y -1 , while the rate tripled for RCP8.5, reaching 1.32 kg N-N2O ha -1 y -1 in the end of the century (average 2080-2099).
A total emission of 1.05 kg N-N2O ha -1 is expected for RCP4.5. Total N2O emissions from croplands and grasslands were reported to the surface allocated for arable crops and permanent 415 grasslands for each simulation unit, by using the share of Corine Land Cover inventory of 2018 (Fig. 5). Emissions ranged between 0 and 2.5 kg N ha -1 y -1 and were concentrated in hotspots, such as northern Italy, north-east Germany and Poland, southern England, Bulgaria, eastern Romania, the Scandinavian peninsula, the north-western of Spain and Portugal. During the climatic projections is observed a general worsening of N2O emissions towards the end of the century, reaching up and often over ±1 kg N ha -1 y -1 , especially for the strongest climatic scenario. An average of 1.02 kg N-N2O ha -1 y -1 (corresponding 420 to 0.163 Mt N-N2O y -1 ) were emitted during the historical period. This amount raised 1.06 and 1.08 kg N-N2O ha -1 y -1 (0.166 and 0.170 Mt N-N2O y -1 ) in the first half of the century for RCP4.5 and RCP8.5, respectively. In the second half of the century total N2O emissions assumed a further increase to 1.11 and 1.13 kg N-N2O ha -1 y -1 (0.169 and 0.174 Mt N-N2O y -1 ) for RCP4.5 and RCP8.5, respectively. Separate emissions of N2O from croplands and grasslands are reported in Fig. S5a,b.
The N2O emission factor (EF), intended as the ratio between the N emitted as N2O from croplands and grasslands, and the N 425 introduced into the system (not including the N added by animal excretion, crop residue, atmospheric deposition, soil mineralisation and fixation), assumed the same trend described for N2O over time. During the historical period the averaged EF for croplands was 1.88 % ± 0.32 %, while the EF for grasslands was 1.99 % ± 0.16 %, see Fig. S6a,b. 430 Figure 6. N2O emission factor (EF %) for croplands and grasslands in European administrative borders (NUTS2). EF is reported for the historical period (1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004), and the difference "Δ" with the middle (2030-2049) and the end of the century (2080-2099) for the two climatic scenarios RCP4.5 and RCP8.5. EF is calculated as ratio between the N emitted as N2O from croplands (irrigable surfaces) and grasslands, and the N introduced into the system (not including the N added by animal excretion, crop residue, atmospheric deposition, soil mineralisation and fixation). 435 Combining cropland and grassland emissions over each simulation unit, the resulting EF was 1.82 ± 0.07 % during the historical period, and rose to 1.90 ± 0.09 % for RCP4.5 and 1.94 ± 0.09 % for RCP8.5 in the first half of the century. EF was 2.02 ± 0.11 % and 2.05 ± 0.11 % for RCP4.5 and RCP8.5, respectively, in the second part of the century. The spatial distribution of EF values at NUTS2 scale, as reported in Fig. 6, varies from 0.1 % to over 5 % in the historical period, to assume variations 440 between ±1 % in RCP4.5, and up to ±10 % in RCP8.5. The hotspots are the same described for the N2O emissions. The specific EF for the simulated crops, calculated in the period from sowing (including pre-sowing management) to the sowing of the next crop in a succession (excluding pre-sowing management), ranged from 0.9 % to 3.4 % in the historical period, and is reported in Fig. 7. EFs toward mid-and the end of the century raised for all the crops, with a greater impact for the RCP8.5 scenario, except for winter soft wheat, which exhibited lower EFs values over the century, and soybeans which presented a low EF at 445 the end of the century for RCP8.5 scenario compared to the less strong scenario. Fig. 7 reported the EF for N2O also for grasslands, which assumed an increasing behaviour following the course of the century and the strength of climate scenarios. EF is ratio between the N emitted as N2O from crops and grasslands, and the N applied.

CH4 emissions
The emissions of CH4 from enteric fermentation are reported in Fig. 8. During the historical period, a mean emission of 6.71 kg C-CH4 ha -1 y -1 was observed, with a rate of 15.6 g C-CH4 ha -1 y -1 (Table 1). The emission rate halved in the first part of the 455 century, to increase slightly in the second part of the century for RCP4.5 (4.3 g C-CH4 ha -1 y -1 ) and strongly decrease for RCP8.5 scenario (-23.7 g C-CH4 ha -1 y -1 ). Emissions toward the end of the year were 6.73 kg C-CH4 ha -1 y -1 in RCP4.5 (average 2080-2099), and 5.74 kg C-CH4 ha -1 y -1 for RCP8.5. The averaged CH4 emissions per head ranged from 2.99 kg CH4 head -1 y -1 in the historical period to reach 3.03 and 3.01 kg CH4 head -1 y -1 In the first half of the century for RCP4.5 and RCP8.5, respectively. In the second half of the century a reduction to 2.98 and 2.73 kg CH4 head -1 y -1 are expected for RCP4.5 and 460 RCP8.5, respectively. The spatial distribution of CH4 emissions at NUTS2 scale is reported in Fig. 9 and ranged from 0 to over 20 kg C-CH4 ha -1 y -1 in the historical period and resulted concentrated in the north-west part of Europe. During the climate projections, methane emissions assumed variations in the range of ±12 kg C-CH4 ha -1 y -1 , with increases mostly concentrated in northern Europe.

Carbon fluxes
Results are presented with sign convention indicating CO2 accumulation as negative, and CO2 losses as positive. Net Ecosystem Production (NEP) for European croplands showed a clear intensification of CO2 accumulation until 2050. Rates 475 were contrasting for RCP4.5 and RCP8.5, -3.27 and +1.44 kg C-CO2 ha -2 y -1 , respectively ( Fig. 10a; Table 1). In the second part of the century, a net divergence between the trends of the two climate scenarios is expected. CO2 is continuously accumulated for RCP4.5 (-1.66 kg C-CO2 ha -1 y -1 ), whereas a decrease is projected for RCP8.5 (9.84 kg C-CO2 ha -1 y -1 ).
Extending the irrigation area over all European croplands, which taken advantage of irrigation volumes according to crop needs and soil water status, produced a proportional increase of CO2 accumulation in the climatic scenarios for both the first 480 half of the century (+6 %) and the second half (+7 %). NEP for croplands is expected to increase toward 2050 at low latitudes (+3 %) for both RCP4.5 and RCP8.5 ( Fig. S7a; Table S1). This trend is inverted toward the end of the century for RCP4.5 scenario (-1 %), while turn to more severe for RCP8.5 (-8 %). In central European latitudes CO2 is accumulated in the first part of the century for both climate scenarios (+9 %) and tended to be released in the end of the century for RCP8.5 (-3 %).
Compared to central European latitudes, higher latitudes shown a trend to store more CO2 for the RCP4.5 scenario respect to 485 the historical period (+5 % in the middle of the century and +9 % in the end of the century), whereas a tendency to release CO2 is forecasted for the RCP8.5 scenario, especially toward the end of century (-5 %). The extension of irrigation to all European area showed a clear CO2 loss towards the end of the century for low and mid latitudes, while a potential accumulation was observed at high latitudes. 490 Figure 10: Net ecosystem production (NEP; g C ha -1 ) for croplands (a) and grasslands (b), with two climate change scenarios (RCP4.5 and RCP8.5), and two irrigation conditions following the irrigable agricultural area in Europe or extending the irrigation to all the arable lands (i_RCP4.5 and i_RCP8.5) NEP in grasslands indicated a clear trend to CO2 accumulation into the system during the historical period ( Fig. 10b; Table 1), 495 with a rate of -0.77 kg C ha -1 y -1 . Towards 2050 a slight imbalance and a tendency to release CO2 is observed for both climate scenarios. Towards 2100, the amount of CO2 potentially stored into the system is maintained for RCP4.5 with a loss of about 100 kg C-CO2 ha -1 y -1 compared to the historical period (-622 kg C-CO2 ha -1 y -1 ), while a clear tendency to CO2 release was forecasted, on average, for the scenario without adaptation, RCP8.5. In the second half of the century RCP8.5 scenario projected a potential loss of 50 % of the CO2 annually stored in the historical period. A potential release of CO2 is also projected 500 for the RCP4.5 for low latitudes, both in the middle (-7 %) and towards the end of the century (-16 %), compared to the historical period (Table S1). Higher decreases are forecasted for the RCP8.5 for the lower latitudes, -13 % and -37 % towards the first and the second half of the century, respectively. Conversely, for latitudes > 55° a potential storage of CO2 is expected for RCP4.5 (+2 % and +3 % for the mid and the end of the century, compared to the historical period), whereas the scenario RCP8.5 gain more CO2 in the middle of the century (+2 %) and turns to negative (-31 %) toward the end of the century. The 505 intermediate latitudes, corresponding to the central Europe, displayed a strong susceptibility to CO2 release in both climatic scenarios, ranging between -19 % and -31 % for RCP4.5 in the middle and at the end of the century, respectively, and turning to more negative and equal to -50 % and -100 % for the RCP8.5 scenario (Fig. S7b).
NEP of the European cropland and grasslands system, obtained reporting emissions the surface allocated for arable crops and permanent grasslands in each simulation unit, is reported in Fig. 11. During the historical period, NEP varied between -7500 510 and +200 kg C-CO2 ha -1 y -1 within the European regions. Climate projections showed variation up to ± 2800 kg C-CO2 ha -1 y -1 from the historical values, identifying a tendency to store less CO2 towards the first half of the century, especially for the Mediterranean regions. CO2 stock is further reduced in central European latitudes towards the end of the century for RCP4.5 scenario, and came to a strong reduction on all regions during RCP8.5. A total of -1865 kg C-CO2 ha -1 y -1 (corresponding to -338 Tg C-CO2 y -1 ) were stocked during the historical period. This amount raised in the first half of the century to -1845 kg C-515 CO2 ha -1 y -1 (-336 Tg C-CO2 y -1 ) for RCP4.5 and -1859 kg C-CO2 ha -1 y -1 (-339 Tg C-CO2 y -1 ) for RCP8.5. In the second half of the century NEP emissions assumed a further increase for both climatic scenarios to -1771 kg C-CO2 ha -1 y -1 (-321 Tg C-CO2 y -1 ) for RCP4.5, and -1620 kg C-CO2 ha -1 y -1 (-293 Tg C-CO2 y -1 ) for RCP8.5.
NGHGE indicated a potential capacity of the European production systems to store an average of -1155 ± 82 Tg C-CO2eq y -1 during the historical period (Table 2). N2O and CH4 were able to offset the NEP by 6.2 % and 0.8 %, respectively. In the first 520 half of the century, the NGHGE assumed a slight reduction for RCP4.5, indicating a potential C stock, whereas remained substantially unvaried for RCP8.5. In the second part of the century NGHGE increased for both RCP4.5 (-1087 ± 119 Tg C-CO2eq y -1 ) and RCP8.5 (-997 ± 159 Tg C-CO2eq y -1 ), indicating a slowdown of C accumulation. Irrigation scenarios highlights an increased potential of C stock of about 3-4 %, mainly due to the greater NEP values. NGB indicated losses from European agricultural surfaces in the range of 236 ± 107 Tg CO2eq y -1 for the historical period (Table 2). Losses increased both in the 525 first and the second half of the century and for both climate scenarios, being higher for RCP4.5 than RCP8.5.

Productions
Results from this study confirmed that the effects of climate change, implying shift of temperature, precipitation, and plant growing length among other factors, represents a serious drawback to plant production.
Air Temperature: Our findings pointed out that the increase of air temperature during the climate scenarios were negatively correlated with productivity, leading to a persistent reductions of biomass production in both grassland and 545 croplands. This behaviour is confirmed also by previous studies (e.g. Challinor et al., 2014;Lobell and Tebaldi, 2014;Olesen and Bindi, 2002;Zhang et al., 2017), and was more pronounced for the more pessimistic climate scenario (-0.15 and -0.29 t DM ha -1 y -1 °C -1 for RCP4.5 and RCP8.5, respectively, in the 2050-2099 period). Effects of air temperature in the European crop yields ranges from +5 % to -11 % for every degree of rising temperature for both climatic scenarios (remaining negative along most of the climate projected scenarios: -1 % and -5 %, for RCP4.5 and RCP8.5, respectively 550 in the period 2025-2099), as also reported by recent studies using modelling and multi-modelling approaches (e.g. Asseng et al., 2015: Bassu et al. 2014Zhao et al., 2017;Yang et al., 2019). The extension of irrigable areas to all European croplands reduce the dependence of daily maximum and minimum air temperatures on crop production (Fig. S8). This demonstrates the fact that even with access to water (no limitation in irrigation), biomass production will decline due to increasing air temperatures, as reported by Minoli et al. (2019). This can be seen also from biomass projections in Fig. 2,  555 considering an increase in temperatures over time. Interestingly, grassland productivity assumed a less pronounced correlation with air temperature during climate scenarios compared to croplands (Fig.12). RCP8.5, characterised by a strong reduction of grassland productions in the second half of the century, has an evident negative correlation with minimum and maximum daily air temperatures (r = -0.6, p < 0.01), up to a null correlation under RCP4.5. Furthermore, crop yields were strictly correlated with minimum and maximum air temperatures (r = 0.64 and r = 0.57, respectively; p < 0.01) compared 560 to grasslands, which did not show such a dependence (r = 0.1 for both minimum and maximum air temperatures), highlighting a greater sensitivity of CERES-EGC model to air temperatures compared to PaSim.
Length of crop growing cycle: Apart from increases in temperature and reduction in precipitation, our simulation highlights that crop yield is affected by the shortening of the length of the growing cycle, as confirmed by Zhang. (2011), andBassu et al. (2014). Bassu et al. (2014)  simulations. Our findings confirm that climate change will have a regionally distributed impact 580 Challinor et al., 2014;Parry et al., 2005;Lobell and Tebaldi, 2014) even in scenario that include mitigation measures to offset climate change (RCP4.5), creating the possibility to the design cropping systems with multiple crops in a year.
Furthermore, a certain number of crops can be cultivated in the Europe even in the worst climate scenario and can potentially yield higher productions than today at high latitudes, while a whole reduction in crop production is expected for low latitudes. 585 Figure 12. Correlation matrixes for croplands and grasslands considering the most interesting indicators for the objectives of this study. Correlation is presented for the historical period  and for the RCP4.5 and RCP8.5 scenarios; for croplands the irrigated and irrigable scenarios are reported in Fig. S8.

590
Finally, as reported in the result for the historical data (chapter 3.1.1), the productions of cropland and grasslands are in line with available data and the recent, albeit scarce, literature, making this study coherent and representative. Regarding the climatic projections, our study predicted an average yield for croplands of 4.49 t DM ha -1 y -1 (ranging from 3.55 to 5.49 t DM ha -1 y -1 ) in the period 2015-2099 with the RCP4.5 scenario, which is in line with the previous estimated yields reported by Lugato et al. (2018) of 4.34 t DM ha -1 y -1 (ranging from 3.69 to 4.90 t DM ha -1 y -1 ) for the same period and climate scenario 595 by using DayCent model.
Assessing the effects of climate change in the European croplands and grasslands, our study give a support for the identification of climate smart practices. Among these, the modulation of crop sowing dates or the implementation of irrigation, represent possible solutions in the short to medium term to prevent water stress (Lehmann et al., 2013).
Sowing date: Shifting sowing dates represents a promising adaptation to overcome yield drop (Olesen et al., 2012). 600 Accordingly, our results showed that earlier sowing dates are expected for spring-sown crops under future climate scenarios, compared to historical dates. Differences between historical and future sowing dates ranged from 0 to -5 days for both RCP4.5 and RCP8.5 scenarios towards 2050, whereas at 2100 horizon earlier sowing dates are predicted with differences of -5 and -7 days for RCP4.5 and RCP8.5, respectively. This clearly shows that the climate change allows significantly more advanced sowing in Europe, as confirmed by Tubiello and Rosenzweig (2008). For winter-sown crops, 605 sowing dates extended in a range from +5 to +9 days toward 2050, to +13 days in the end of the century for RCP4.5. These increases raised in RCP8.5, ranging from +7 to +13 toward 2050 and reaching +19 days on the way to 2100. The extension of irrigation in all simulated crops in Europe had a negligible influence on the length of the crop cycles, as discussed by Minoli et al. (2019), despite an increasing demand of water over the course of the century.
Irrigation: Water demand has been shown to increase by +6 %, during the first half of the century, to slightly decrease in 610 its second half for RCP4.5 (-2 %) and increase again for RCP8.5 (+23 %) scenario. These changes are in line with the results of the multi-model approach used by Wada et al., (2013) analysing the uncertainty of the response of different hydrological models over Europe. Wada et al (2013) showed a decrease in water demand for irrigation toward 2100 of <5 % for RCP4.5, and a rise of >20 % for RCP8.5 in Europe. Furthermore, from our study we observed that water demand would assume a strong regionally variation in Europe, with low latitudes needing 227 mm y -1 on average in the historical 615 period (mean 1985-2004), an order of magnitude higher, respectively, than mid latitudes (29 mm y -1 ) and high latitudes (9 mm y -1 ). These proportions between the latitudes remained unvaried over the course of the century, whereas mid and high latitudes displayed a 20 % increase of water demand towards 2050 (mean 2030-2049) compared to historical period, in both climate scenarios. This phenomenon for low latitudes is strictly related to climate perturbation (i.e. strong increase of air temperature and reduction of rainfall), which increased crop water demand (Olesen et al., 2011). Furthermore, the 620 potential increase of water demand even in mid and high latitudes, confirm that irrigation need to be supplied even for the crops that are now commonly rainfed (e.g. spring and winter soft wheat, spring barley, sunflower, rapeseed). Towards 2100, the water volumes needed for European croplands were largely reduced to under the amounts observed during the historical period, especially for low latitudes. These findings underline that even with high availability of irrigation water, the reduction of the crop growing cycle for the actual crop varieties, which sharpens toward the end of the century, is 625 decisive to determine drops of yields. This is more evident for grain maize, the most water-demanding crop (Fig. 3), which needs an additional +35 mm y -1 (average over Europe) to support production toward 2050, compared to the historical period. Towards 2100 water demand for maize remains identical to the historical period for RCP4.5, while increased (+25 mm y -1 ) for RCP8.5. Conversely, water demand for winter soft wheat remained constant along the century for both RCP4.5 and RCP8.5 scenarios, whereas i_RCP4.5 and i_RCP8.5 scenarios confirmed an increasing water demand of about 50 mm 630 (average over Europe; Fig. 3), as confirmed by Yang et al. (2019) for the Mediterranean regions.

Effect of climate on N2O and CH4 emissions
Emission of non-CO2 GHG such as N2O and CH4 form enteric fermentation are strictly related to the trend of biomass productions (Maaz et al., 2021).

N2O.
The estimation and the projection of N2O emissions in the historical and the climate change scenarios were in line 635 with other model integrations over Europe. Lugato et al. (2017) estimated averaged emissions ranging from 1.18 to 2.63 kg N-N2O ha -1 y -1 in the period 2010-2014 for both cropland and grasslands production systems with the DayCent model.
In comparison with Lugato et al. (2017), we found similar results for the Mediterranean latitudes (about 1 kg N-N2O ha -1 y -1 ), while we predicted significantly lower emissions for Central Europe (1.1 kg N-N2O ha -1 y -1 , this study), as well as at higher latitude (0.96 kg N ha -1 y -1 , this study), compared to 3 kg N-N2O ha -1 y -1 forecasted by Lugato et al. (2017). Indeed, 640 lower emissions at high latitudes were also reported by other studies (World Bank; Eurostat, 2017;Stehfest and Bouwman, 2006;Wells et al., 2018). Other research in the field were also within the range of our results e.g. Reinds et al., (2012)  significantly higher than those found in our study (0.17 Mt N-N2O y -1 ) for the same period. In addition, the estimation by Tian et al. (2020) included also manure management and aquaculture, and suffers from high uncertainties given by the quality of the data and statistics used as input and, foremost, by the use of default emission factors. Regarding climate 650 projection studies, Lugato et al. (2018) quantified N2O emissions for croplands in the RCP4.5 scenario, reporting losses of 1.81 and 1.77 kg N-N2O ha -1 y -1 for the first and the second part of the century, respectively. These estimations resulted comparable, while slightly higher, to the emissions for croplands issued from our study, both for the first part of the century (1.53 ± 0.23 kg N ha -1 y -1 ) and for the second (1.66 ± 0.28 kg N ha -1 y -1 ). Our study highlighted that crop type is a significant determinant of N2O EFs of fertilisers, with most of the cereals having low EF (barley, fodder maize, soft spring wheat and 655 rapeseed; mean = 1.1 %), and pulses, soybean and potato a high (mean EF = 3.1 %), during 1985-2004 integration period.
The highest EF for leguminous crops indicates that the management of fertilisation for these crops, or for the rotation itself, can be improved on the input data. Finally, information about crop-specific EF turns to be useful to design crop successions and compiling emission inventories (Myrgiotis et al., 2019). However, our results were higher than the 1 % default value defined by the IPCC guidelines for the N applied to agricultural soils, mainly because we consider only the N applied as 660 fertiliser, neglecting animal excretions, crop residues, deposition, mineralization and fixation. Anyway, this default factor shows large uncertainties at local to regional scales, especially for agricultural N2O emissions, due to the scarce captured dependence of emission factors on spatial diversity of management, pedoclimatic, soil physical and biochemical conditions Reay et al., 2012;Shcherbak et al., 2014;Cayuela et al., 2017). We observed that N2O emitted from croplands had a significant and positive correlation (p < 0.05) with rainfall (r = 0.47), as well as minimum and maximum 665 air temperatures during the historical period (Fig. 12). The correlation with the minimum and maximum air temperatures increased (p < 0.01) depending of the climatic scenarios (r > 0.5 for RCP4.5 and r > 0.9 for RCP8.5; Fig. 12), while the relation with rain turned to negative for RCP8.5 (r = -0.32, p < 0.01). This trend inversion is probably connected to the strict dependency of N2O emissions to the length of crop growing period rather than the yearly cumulated rainfall, which can occur outside of the cultivation period, as also stated by Shcherbak et al. (2014). Accordingly, the correlation from 670 N2O and the irrigation amount occurring during the cultivation period raised in the climate scenarios (r = 0.23 and 0.59 for RCP4.5 and RCP8.5, respectively; p < 0.01). Moreover, N2O emissions from cropland and grasslands were both positively correlated with soil clay content (r > 0.5, p < 0.01; data not shown) for values lower than 32 %, as higher clay content can promote complete denitrification (Weitz et al., 2001).

CH4.
Methane emissions in EU were mainly concentrated in the regions with the highest density of grazing animals 675 . The range of the emissions simulated in this study were in line with the simulation of Chang et al. (2015), which found emissions of 18.7 ± 7.9 kg C-CH4 ha -1 y -1 (period 1961-2010) and by Hörtnagl et al. (2018) by experimental trials from central European grasslands.  reported emissions over Europe higher then to our study, 41 kg C-CH4 ha -1 y -1 with comparable animal densities, as well as Viuchard et al. (2007) with 108 kg C-CH4 ha -1 y -1 using PaSim model, but with a higher stocking rate. CH4 emissions decreased towards the end of the century, 680 especially in the RCP8.5 scenario, due to reduced biomass productivity of grasslands that reduced animal intake (Fig. 2) and the stocking density, which is reduced to 8 % compared to RCP4.5 in the last decade of the century. Reduction of stoking density was also found by Chang et al. (2015). Furthermore, rising temperatures and reduced precipitations could be able to decrease the protein content and the digestibility of the forage, resulting in a possible reduction of N2O losses from dung and urine in pastures. However, this mechanism could be compensated by in an increase of methane (CH4) 685 losses (Wilkinson and Lee, 2017).

Nitrogen Use Efficiency (NUE).
We observed an increase in the NUE for the European croplands, especially for the mild climate change projection. Compared to the historical period, in the RCP4.5 scenario, there is a reduction in the correlation between N2O emitted with the other N losses (NO3 and NH3) and crop yield (Fig. 12). Conversely, there is an intensification of the dependence with the N dose. In fact, with an identical amount of N applied in the rotations over the simulated years, 690 both NO3and NH3 losses were reduced along the century (data not show), and crop yields increased, at least, until 2050.
This indicated a potential increase in the NUE. Kanter et al. (2016) observed an increase of the NUE by 2050 due to the increasing yields, support our findings. The improvement of NUE is a key factor to reduce environmental negative effects and mitigating GHG emissions. Bouwman et al. (2013) indicated that NUE improvement could reduce N2O emissions by more than 30 % by 2050 in the RCP8.5 scenario. On the other hand, in the RCP8.5 scenario the correlation between N2O 695 emissions and N dose is lost (p > 0.01), and a strong negative (p < 0.01) score between yield and nitrogen losses took place, indicating a reduction of NUE. This lack of relationship is most probably connected to the interannual variability of N2O emissions in the strongest scenario and in the second part of the century. Higher NUE are typical for low European latitudes than mid and high latitudes, since yields are generally higher and the N losses lower (Sutton et al., 2011). Improving actual agronomic practices to improve NUE could have several benefits. These practices can increase crop yields and reduces 700 reactive N losses, including N2O emissions (Lassaletta et al., 2014;Myrgiotis et al., 2019). In this context, irrigation represents a fundamental intensification practice to counteract the effects of climate change in crop productions (Minoli et al., 2019). In our case, the extension of the irrigation to all cropping systems in EU significantly decreased N losses as (-5 % for NO3 leaching and -4 % for NH3 emissions, on average, for both i_RCP4.5 and i_RCP8.5 scenarios), and increased crop yield (Fig. 2), leading to a potential increase in the NUE. 705 Concerning grasslands, we noted weak relationship between N2O emissions and N application doses. This is mainly due to the calculation of N doses and management as a function of animal loads, fraction of leguminous, mowing events and available amount of mineral fertilizers and / or organics. During the historical period, N2O emissions were positively correlated with NO3 and NH3 losses and negatively correlated with productions, representing a potential low NUE.
Moreover, N2O emissions in grassland were anti-correlated with CH4 emissions. CH4 emissions are rather positively related 710 to biomass production and livestock intake. Therefore, low biomass production could potentially increase N2O emissions, due to low NUE, and can decrease CH4 losses, due to low livestock intake. Surprisingly, N2O emissions in grasslands were weakly correlated with meteorological variables, especially minimum and maximum air temperatures, whereas a relation with rain and solar radiation is noticeable for RCP4.5, while is not evident for RCP8.5. As observed for croplands, the relation between N2O and NH3 emissions is positive, especially for the RCP8.5 scenario, to indicate a possible reduction 715 of the NUE.
Potential carbon stock NEP. The NEP represents a simple indicator of carbon storage potential, since does not account for C removal in terms of yield, animal intake or crop residues. Concerning cropland, our results are directly comparable with Kutsch et al. (2010) during the historical period who observed fluxes of −2400 ± 1130 kg C-CO2 ha −1 y −1 based on field measurements in 720 multiple sites in Europe (see Table 1), confirming a net potential storage of C. Regarding climate scenarios, a noticeable decline of C uptake was predicted in north-western cropping systems (British islands, Scandinavian peninsula) and Mediterranean area. This is most probably due, respectively, to the increase of soil heterotrophic respiration caused by climatic factors, and to a potential reduction of NEP, as also reported by Kirschbaum (1995) (Fig. S9). Further decreasing values of NEP (towards carbon stock) were evident in the central and in the north-eastern European, especially in the first 725 part of the century. A substantial increase of NEP in croplands was predicted towards the end of the century for the RCP8.5 scenario. This increase is most probably due to the low levels of heterotrophic respiration (i.e. microbial respiration due to soil organic matter decomposition processes) related to a partial soil coverage (e.g. no cover crops) of the simulated crop successions (Emmel et al., 2018). Conversely, in grasslands systems we observed lower averaged values compared to arable lands. This is related to the continuous biomass removal from grazers, the general higher content of SOC in the 730 topsoil, the long-term land use (Morais et al., 2019), and the larger heterotrophic respiration that characterises these soils, especially if extensively managed (Bahn et al., 2008). These evidences were also described by Chang et al. (2015) who simulated an average of -570 kg C-CO2 ha -1 y -1 between 1961 and 2010 for EU (close to -622 ± 62 kg C-CO2 ha -1 y -1 in the historical period from our study). In general, areas where the heterotrophic respiration is enhanced by climatic drivers or by high amount of SOC, would lead to lower values of NEP (Chang et al., 2017). This is the case for the north-east of 735 France and the British islands, while for the Scandinavian peninsula and north-east Europe, characterised by low C and low heterotrophic respiration, NEP reached higher values. These findings point out that in view of a growing productivity expected towards 2050, storing additional (new) carbon will be more challenging in areas characterised by high levels of SOC (Hassink and Whitmore, 1997), mainly due to high levels heterotrophic respiration. Finally, grasslands remained a potential sink for C during the historical period, which was in line with experimental measurements performed in the last 740 two decades, e.g. -2470 ± 670 kg C-CO2 ha -1 y -1 reported by , and -25 to -486 g kg C-CO2 ha -1 y -1 reported by Hörtnagl et al. (2018). However, our results were slightly higher in absolute value than the mean value simulated by Chang et al. (2015) from 1961 to 2010 (-570 ± 210 kg C-CO2 ha -1 y -1 ). N2O emissions from croplands were able to offset (reduce) the C sequestration potential. Offsets were in the order of 5.4 % for the historical period, and up to 6.1 % and 7.5 % in the end of the century for RCP4.5 and RCP8.5, respectively. The 745 extension of irrigation to all European arable lands reduced these gaps, mainly due to the increase values of NEP (5.4 % and 7.1 % for i_RCP4.5 and i_RCP8.5). Few data are available in the literature regarding the CO2 storage potential for croplands (Emmel et al., 2018). Our results confirmed that croplands may act as a potential sink of C when ignoring C exports by harvest (Buysse et al., 2017;Ceschia et al., 2010).
N2O and CH4 emissions in grasslands were able to offset NEP during the historical period by 17 % and 1 %, respectively. 750 These results are compatible with the studies reported by Soussana et al. (2010) who displayed offsets over EU of 34 % and 10 % for N2O and CH4, respectively. During climate projection, the offset rises to 22 % for N2O and 1.2 % for CH4 towards 2050 for both RCP4.5 and RCP8.5. In the second part of the century N2O emission offset the potential carbon sequestration by 26 % and 52 % for RCP4.5 and RCP8.5, respectively, while CH4 offsets varied between 1.2 % and 1.9 % for RCP4.5 and RCP8.5, respectively. 755

GHG emission budget
For both cropland and grasslands, CO2 storage potential (estimated from NEP) provided the largest term in the net greenhouse gas exchange balance (NGHGE), confirming the statement by Jones et al. (2016). The NGB, calculated as the balance between NGHGE and other C forms (i.e. harvest, manure and crop residues), indicated that European agricultural surfaces are a net C source. The most important components that determined these losses were the C exports, yield (FC-harvest) and crop residues 760 (FC-residues), which varied proportionally to the NEP in the various climatic projections, i.e. the lower the NEP, the lower the yields. The non-CO2 GHGs, despite being high especially in the RCP8.5 scenario towards the end of the century, had a minor impact in the differentiation of the two climatic scenarios, although they represent an important component in the overall carbon balance at the European level (see Table 3).
The values observed for NGB highlight that C inputs into the system such as organic fertilizers (the manures used in this study 765 have a C:N ratio of 25 and represent 2/3 of the component FC-manure), or actions aimed to recycle a portion of biomass in the field (crop residues management), are essential to improve the overall C budget toward a net storage, as reported by Ceschia et al. (2010) and Buysse et al. (2017). Moreover, our findings shown that the contribution of crop residue roughly corresponded to the carbon deficit in Europe. Therefore, crop residue could play a key-role in land-based mitigation of anthropogenic emissions, as also reported by Stella et al. (2019). This is in line with the "4 per 1000" initiative (Rumpel et al., 2019) promoting 770 the maintenance of soil fertility as a key to achieve GHG mitigation strategies. In addition to the spatial diversity observed on the European agricultural area, the achievement of this goal depends on the complexity of rural, economic, and political structure of the territories (Amundson and Biardeau, 2018). Furthermore, local policies can be supported by simulation tools as used in this study, bearing in mind that their effectiveness can be affected by the omission of large variances given by varied characteristics of small extents. Finally, irrigation management extended to all European cropping land is able to increase the 775 stored C (NEP = -3 %), but increases the contribution of FC-residues and the non-CO2 GHG (up to +4 %), leading to slightly higher C deficits.

Uncertainty, limitations and novelty
The extension of field-scale models to a regional scale faces several challenges associated with the representation of the 780 systems under study, which can affect the confidence of model outputs (Challinor et al., 2014;Folberth et al., 2019).
1. Input data requirement for such models for large and heterogeneous areas are difficult to fulfil . Soil and climate inputs are directly available from European databases at different spatial resolutions. Details on crop and grassland management (e.g. type and amount of inputs, timing of operation, tillage system, crop varieties) are less readily available and are an important source of uncertainty (Molina-Herrera et al., 2016). In our assessment 785 we used a dataset for cropland, with data of crop rotations resulting from a spatial crop cultivation distribution (NUTS2) and by crop succession likelihoods on a high-resolution scale (1 km). For instance, this dataset do not report details about management or crop growth parameters. The absence of plant phenological development data, constitutes a relevant source of uncertainty in regional assessments (Minoli et al., 2019) since they affect crop growth and growing length, the biogeochemical cycles at different scale and are key for future projection. To deal with this 790 lack of information, we calculated crop-specific sowing and fertilisation dates as a function of climate (Ramirez-Villegas et al., 2015), together with the uses of different crop varieties following a latitudinal gradient to fulfil the thermal units needed, N doses and the crop-specific residue management, aiming to reduce the uncertainty of input data (Hansen and Jones, 2000). Furthermore, the use of two different crop rotations per simulation unit attempt to cover a range of uncertainties existing below the spatial resolution of 0.25°, which, however, cannot be assumed to 795 be fully covered by the range of setups presented here. In the present work CERES-EGC model used fixed parameters issued from a calibration over different sites in EU Lehuger et al., 2011; goodness of fit, R² = 0.59 to 0.76 for NEP; error of prediction reduced by 6-40% for N2O compared with the model's standard parameters).
Grasslands, as previously reported, were simulated with a parameter set resulting from a multi-site calibration for a network of EU grasslands (i.e. flux tower network, see Ma et al., 2015; goodness of fit, R² =0.4 to 0.9). Likewise, 800 PaSim follows an adaptive management based on climate. Since the information concerning the input data are already the result of a scaling process, we retain that an uncertainty analysis concerning the input data is not appropriate (Hansen and Jones, 2000).
2. Calibration of models: to fulfil this task over large areas, data representing the spatial and temporal variation of models' parameters are required. Although both models have been calibrated and verified with direct observations 805 under various pedo-climatic and management conditions at the field-scale, comprehensive studies aimed to calibrate these and other models with spatially extensive time series are still scarce (Balkovič et al., 2013;Lehuger et al., 2010;Lugato et al., 2010;Vuichard et al., 2007). Data aggregation over the same extent can be used to assess model representations, even if they do not represent the field-scale conditions for which the models have been originally calibrated (Lugato et al., 2017;Therond et al., 2011;van der Velde et al., 2009), exposing them to a broader range of 810 conditions (e.g. weather and soil characteristics). Indeed, dealing with lacking and heterogeneous input data requires different procedures of downscaling and upscaling for the different data types, which potentially contribute to feed the uncertainty of the representation. Consequently, projecting regional model responses under future climate scenarios requires careful understanding of input and model uncertainty (Asseng et al., 2013;Challinor et al., 2009). This is the reason why the two periods of temporal aggregation considered in the present study, historical and climate 815 scenarios, provide outcomes with different levels of confidence. In the historical period, results are obtained based on the spatial aggregation of real (statistical) data by means of models parameterised with current soil, climate and vegetation conditions. The outcomes of the climate scenarios deal with the uncertainty related to the sensitivity of the model parameters and the algorithms to climate variables, which is expected to be different due to the diverging intensities of the two projections and the different conditions in the near-and the long-term (2100). For this reason, 820 direct comparisons between the two aggregation periods should be done with caution.
3. Model validation: Data quality and availability prevent also the validation of regional scale models, even if literature report some effort (Challinor et al., 2009;Faivre et al., 2004;Niu et al., 2009). Comparing the outputs with statistical data aggregated at regional scale (productions) allowed to obtain indications about the magnitude of simulated variables at the same spatial extent. Furthermore, assessing the ranges of the model outputs, e.g. yield, with measured 825 data and over EU (r = 0.92, p<0.01 for croplands, Fig S1; r = 0.68, p < 0.05 for grasslands, Fig 1), as well as other modelling interpretations (even if grounded on different approaches), contributed decidedly to increase the reliability of our estimations.
Finally, literature reports similar studies aiming to estimate crop and/or grasslands productions, GHG emissions and carbon storage at regional scale (Chang et al., 2017;Lugato et al., 2018;Blanke et al., 2018). Compared to them, the novelty introduced 830 by this study grounds on the combination of two specific models for the systems under study, for the detailed and dynamic management options (Leip et al., 2008;Lugato et al., 2017) and for the finer spatial resolution (Ciais et al., 2010;Iglesias et al., 2012;Lugato et al., 2014;Vuichard et al., 2007). Moreover, albeit aggregated in simulation units, our work considers a variety of pedoclimates over EU and is not based on an extrapolation of a few points or on a single European area Kutsch et al., 2010;Myrgiotis et al., 2019;Soussanna et al., 2010). Knowing and controlling the sources of 835 uncertainty from regional applications could be a key to the improvement of decision-support tools for the design of policies.
In this context, providing a range of possible outcomes, the application of multi-model ensemble (Ehrhardt et al., 2018;Martre et al., 2015;Sándor et al., 2018;Rosenzweig et al., 2013) at regional scale, could represents a valuable tool to tackle this uncertainty. Increasing spatial resolution of the input dataset we used (weather data) could represent also a key to further reduce uncertainties from input data in future large-scale applications (Folberth et al., 2019;Hoffmann et al., 2016;Stella et 840 al., 2019).

Conclusions and perspectives
In this study we presented the combined spatial analysis of two specific models for crops and grassland, to quantify the effects of climate change on the European agricultural systems. Results clearly showed that the productions will be stable in the first half of the century, while a strong reduction will occur during the second half of the century, especially at low latitudes, and 845 mainly due to a reduction in the length of growing cycle. Non-CO2 greenhouse gas emissions were triggered by the rising temperatures, increasing significantly in the second part of the century. At the EU scale, both grasslands and croplands are potential carbon sinks, although this potential is reduced by the negative effects of climate change on productivity. Biomass removal from the agricultural surfaces (yield and hay), combined with the animal intake and the removal of crop residues, transformed the production systems into a net source of carbon. In this framework, the introduction of carbon with fertilizers 850 and dung was not able to counterbalance these removals of C. Crop residues restitution could be a potential strategy to improve the overall carbon balance towards a C neutrality, or even towards a C storage. The effects of crop residues recycle on N2O emissions and the greenhouse gas balance need to be investigated with further researches. Our study highlights that storing further carbon in areas characterised by high levels of SOC will be more challenging in the future. The extension of irrigation to all European croplands highlights a significant increase of water demand over the next few decades for most of the European 855 croplands, whereas the benefit in terms of crop yield will not contribute substantially to fill the gap of carbon losses.
Our findings show that productivity, GHG emissions and changes in soil C-stock have a heterogeneous spatial distribution.
This underlines the need of targeted agricultural policies at territorial scale aimed to avoid the risk of significant reductions of productivity and mitigate the negative effects of climate change, foremost expected in the second half of the century.
Accordingly, this transformational adaptation has to deal with socio-economic and political dynamics, as well as land 860 suitability (Fischer et al., 2005;Chaudhary et al., 2018, Martin-Lopez et al., 2019. This work provides a database on cultivation and management of cropland and grassland at a detailed spatial level, which can be improved and exploited in future work to test different management options, new or a combination of agro-ecosystem models, climate change projections, crop varieties or floristic compositions, and the support for future actions.

Data availability 870
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. European Commission, 2020. Communication from the commission to the European parliament, the council, the European economic and social committee and the committee of the regions Stepping up Europe's 2030 climate ambition Investing 985 in a climate-neutral future for the benefit of our people,