Climate change related lessons learned from a long-term field experiment with maize

doi:10.21203/rs.3.rs-5241040/v1

Download PDF

Research Article

Climate change related lessons learned from a long-term field experiment with maize

https://doi.org/10.21203/rs.3.rs-5241040/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Maize is the second most important cereal crop in European agriculture and a widely used raw material for feed, food and energy production. Climate change studies over Europe project a significant negative change in maize production. Finding appropriate and feasible adaptation strategies is a top priority for agriculture in the 21^st century. Long-term agricultural experiments (LTE) provide a useful resource for evaluating biological, biogeochemical, and environmental aspects of agricultural sustainability and for predicting future global changes. The objective of the study was to analyze a 30-year period of a multi-factorial (Variety × Fertilization × Planting date) LTE at Martonvásár (Hungary) searching for traces of climate change as well as for favorable combinations of agro-management factors that can be used as adaptation options in the future. According to the results: (1) intensification of fertilization would not promote sustainable development in the region; (2) late hybrids (FAO number > 400) have no perspective in the Pannonian climatic zone and (3) Earlier planting (first decade of April or even earlier) may become an effective adaptation option in the future. Our comprehensive methodology combines long-term historical weather and climate projection data with statistical and simulation models for the first time to provide agricultural stakeholders with more reliable adaptation strategies than ever before.

Planting date

Genotype

Fertilization

Mixed model

Crop simulation model

Maize (Zea mays L.) is the second most important cereal crop in European agriculture (EC 2022a) and the first regarding its harvested area in Hungary (KSH 2022). Maize is used for feed, food and energy production (EC 2022b) all of which is expected to be significantly affected by climate change.

Climate change studies over Europe show a consistent increase in temperature and different patterns of precipitation with widespread increases in northern Europe and decreases over parts of Southern and Eastern Europe (Olesen et al. 2016). In the Carpathian Basin, Hungary is going to face significant warming due to climate change as well, with the largest temperature increase likely to occur in summer while the summer precipitation is projected to decrease significantly during the 21st century (Pongrácz et al. 2011). It is important to note that the uncertainty of changes predicted by the latest climate projections is also significant (see Fig. 5.)

A substantial number of large (continent-global) and small (country or state) scale modelling studies have already investigated the possible impacts of climate change on maize production. In addition to studies on potential climate change impacts, also the likely effects of some agro-management options (irrigation, fertilization level, planting date, genotype selection etc.) for adapting to anticipated climate changes have been taken into account. The Europe-wide drought stress study of Webber et al. (2018) suggested that average drought losses will increase for maize and elevated CO₂ will not be able to mitigate yield losses from drought. Parent et al. (2018) found that European maize yields could be increased by 4–7% if farmers would make the best use of the current genetic variability of maize hybrids and the genotype selection as well as the sowing dates were optimized in each local environment. Additionally, for a given set of climate change projections ex ante evaluation of farmer's adaptation measures has shown that adaptation would be likely to reduce the yield gap between south and north of Europe and could result in an increased European maize production if farmers use the available best practices. Moore and Lobell (2014) showed high adaptation potential for maize to future climate change in Europe. Agricultural profits could increase slightly if farmers take adaptation measures but could decrease in many areas if there will be no adaptation. Local studies give a more detailed picture of the expected changes. In their high resolution study, Mereu et al. (2021) concluded that maize will be a largely affected crop, with yield reductions homogeneously distributed from North to South Italy. Using different modelling approaches, Bassu et al. (2021) found that by 2060 the potential maize yield is expected to decrease by 14–17% in the Mediterranean region. The projected losses may only be partially offset by changes in genotypes and sowing dates. Extended growing season due to warming climate will likely create opportunities for the northward expansion of crop production and the introduction of ‘new’ crops (e.g. maize) and varieties. Examining the possibility of silage maize production, Eckersten et al., (2012) were found that by the end of the twenty-first century, an adequate fodder quality would be obtained every year in southern Sweden, however in the middle of Sweden (60°N), approximately 30% of the years would be unsuccessful, even for the earliest cultivars. According to Žydelis et al. (2021), since climate change will bring warmer and wetter conditions in the Nemoral climate zone (South Scandinavia and the Baltic countries), increasing maize yield trends are expected with 200–300 kg ha^− 1 per decade yield growth depending on the underlying scenario. Cuculeanu et al. (1999) found that under different scenarios maize showed negative responses to climate change in southern Romania. They also found that the dominant adaptation strategy should include irrigation, use of longer maturing hybrids, later sowing, and the increase of nitrogen levels. Parker et al. (2017) reported that in Germany earlier planting potentially increases yield but this is offset by additional management costs and risks. Buhinicek et al. (2021) investigated the future suitability of five maize maturity groups in Southeast Europe and pointed out that choosing early hybrids does not seem to be an optimum selection in the future. According to a study on Hungary (Fodor et al. 2014) the average yield loss for maize production could reach 2000 kg ha^− 1 y^− 1 by the end of the 21st century having a present slightly over 6000 kg ha^− 1 annual average yield level. On the other hand, sustaining human impact has the potential to reverse the projected decreasing trend in Hungary (Marton et al. 2020).

Region-specific adaptation strategies: what works, where and how much, depend greatly on local conditions and opportunities. This fact increases the importance of local field experiments as their results can be used for model calibration and validation. Every model should go through these steps before being used for addressing scientific or practical problems (Kersebaum et al., 2015; Ginaldi et al., 2016; Choruma et al., 2019, Liang et al., 2018). To test a model’s capability of accounting for the effects of different agro-management options, special experiments are required. Multi-factorial experiments investigate the effects on plant production of usually two or more factors among the following: variety selection (V), fertilization level and date (F and FD), irrigation (IR), planting date (P), plant density (PD), plant protection (PP), and soil management (SM). Russelle et al. (1987) determined the optimum F×P combination(s) at two N rates for irrigated maize in Nebraska, USA based on a 3-year-long experiment. Based on results from two years, Tsimba et al. (2013) reported on the V×P interaction for maize concluding that delayed planting is an unfavorable option in New Zealand. Bassu et al. (2021) investigated V×P×PD interactions (though not all the combinations) in a 3-year-long experiment with maize under optimum management (no water and nutrient limitation) in Italy. After calibrating and using two models of different yield calculation approaches they concluded that in absence of abiotic yield-reducing factors, the highest yield can be realized with early sowings, and regarding the final yield the effect of sowing date is larger than the effect of the cultivar selection. Without neglecting the importance of shorter field trials, it should be noted that these experiments of 2–5 years in length, are not necessarily suitable for detecting climate×agro-management interactions, as the weather patterns over such short periods may not be good representations of local climatic conditions, not to mention climate change induced trends.

Although long-term agricultural experiments (LTEs) have numerous constraints and weaknesses (e.g. change of genotypes) they are the only way to identify long-term trends (Berti et al. 2016) as well as robust, site specific features of the interactions between the investigated factors. By definition, these experiments are carried out for at least 20 consecutive years and study crop and/or livestock production, nutrient cycling, and environmental impacts of agriculture (Grosse et al., 2020; Macholdt et al., 2020; Li et al., 2023; Pereyra‑Goday et al., 2024). They provide a useful resource for evaluating biological, biogeochemical, and environmental dimensions of agricultural sustainability and for predicting future global changes (Rasmussen et al. 1998; Reckling et al., 2018). These experiments are valuable for spatially differentiated data analyses and reuse of data in modelling studies for validating model capability and performance (Grosse et al. 2020; Rasmussen et al., 1998). LTEs are long enough to detect the effects of climate change on the factors investigated in the experiments (Donmez et al., 2023) or perhaps more accurately, LTEs are long enough to represent the climate variability and change of the historical period (of weather and agronomy) yet, they cannot give us final answers on the expected changes in climate in the future - and their implications on maize crop performance. The evolution of LTEs in the past decades show a clear change in the focus of research investigating not only a specific practice but rather a combination of practices and the possible interactions. Although the range of factors that, at one location, are worth investigating is constrained by pedo-climatic variables, the explicit intent for using a more holistic approach promises a more ecological and climate-smart management of agroecosystems (Blanchy et al., 2024).

The objective of the current study is to analyze a 30-year period of a multi-factorial (Variety × Fertilization × Planting date; V×F×P) LTE at Martonvásár, Hungary (Fig. 1) searching for the traces of climate change in the yield trends as well as for favorable combinations of agro-management factors that can be used as adaptation options in the future.

Figure 1

2.1. Soil and climatic characteristics of the experiment area

The field trial was carried out on the experimental farm of the Centre for Agricultural Research, Martonvásár, Central Hungary (N 47°19’, E 18°47’, 110 m asl). The soil is classified by FAO-WRB (IUSS Working Group, 2015) as a Haplic Chernozem (34% sand, 42% silt and 24% clay in the 0–25 cm layer), with average pH_H2O of 7.59, 1.84% CaCO₃, 3.39% Soil Organic Matter, and 1799/374/429 mg kg^–1 total N/P/K content. Based on the water retention curve measured in the laboratory, the saturated water capacity, the field capacity and the water content at wilting point are 0.476, 0.322 and 0.134 cm³ cm^–3, respectively. In order to gain insight into the spatial scalability of the results that will be presented subsequently, an investigation was conducted into the prevalence of the soil type of the Martonvásár area. According to the global soil map of the World Reference Base for Soil Resources (IUSS Working Group, 2015) Chernozems cover 20.7% (1.17 million hectares) and 15.6% (2.39 million hectares) of the arable area in Hungary and in the Carpathian Basis, respectively. In Hungary, Chernozems account for the largest share of arable land, ahead of Vertisols and Gleysols.

Long-term, annual meteorological data for the area, recorded at an on-site station, are shown in Fig. 2. For each year, precipitation sum (Psum), mean temperature (Tmean), the number of hot days when the daily maximum temperature is over 30°C (nrHotD), and the number of days with precipitation (nrPD) when precipitation exceeds 0.1 mm, vapor pressure deficit (VPD) and total reference evapotranspiration (refET₀, defined by Allen et al., 1998) are plotted.

Figure 2

In the Supplementary Material, these indicators are presented for the vegetation period, the flowering period and also for the grain filling period (Fig. SM1). The significance of the trend in climatic characteristics was tested using t-test, and t-test conditions (normality of the residuals and absence of auto-correlation) were tested using Jarque-Bera and Durbin-Watson tests, respectively, with the help of the statsmodels 0.13.5 Python package. The required conditions for applicability were met for all the characteristics examined.

Trends of the above climatic indicators were investigated for the whole study area to see how the changes at Martonvásár are representative of trends across the region. For this purpose, the FORESEE database (Kern et al., 2024) was used. Its 10×10 km resolution grid covers the area of Hungary with 1014 cells containing observation based, spatially interpolated, daily weather data.

2.2. Experimental design

The experiment involves three factors: four planting dates (P), five fertilization doses (F), and five varieties (V) in every single year. The choice of varieties changed over the years, reflecting breeding progress but in each year (Y) five different varieties were sown from the early, medium and late maturity groups. The list of varieties used in the trial is shown in Table 1. FAO number is a characteristic of maize maturity groups (Jugenheimer, 1958): the lower the number, the fewer heat units that are required to reach grain maturity. According to their FAO numbers, early (FAO 290–320) medium (FAO 330–420) and late (FAO 430–550) varieties were sown in each year. In the five fertilization treatments (F = 1 to 5) 0, 60, 120, 180 and 240 kg ha^− 1 N was applied annually, two weeks before the first planting date. Planting date treatments are described in Table 2.

Table 1

List of varieties used in the experiment. The Years column indicates the period of years when the varieties were used in the experiment
Variety	FAO number	Years
Mv Tc 1287	320	1992–1994
Mara	290	1995–1998
Mara	290	1999–2000
Mv 272	300	1999–2001
Mv 277	300	2002–2021
Norma	370	1992 − 1913
Dáma	330	2001–2003
Hunor	350	2004–2021
Furio	390	1992–1994
Mv 355	400	1995–2020
Tarján	380	2014–2021
DK 524	530	1992–1994
Maya	430	1995–1997
Botond	420	1998
DK 608	550	1992–1993
Mv 1514	540	1994
Mv 484	480	1995–1996
Maraton	450	1997–2008
Miranda	460	2009–2015
Danietta	500	2016–2020
Mv 352	500	2021

Table 2

Planting date treatments used in the experiment. DOY denotes the day of the year.
Planting date (P)	Planting dates (DOY): min / median / max
1	94 / 103 / 116
2	104 / 113 / 122
3	115 / 123 / 131
4	127 / 134 / 141

Table 1

We aimed to have a hybrid from each of the three maturity groups in the trial every year. Although there were no hybrids included in the trial every year, the official crop investigation and certification system in Hungary guarantees that the expected yield of the new registered hybrids included in the trial will be at least as high as that of the older cultivars.

Table 2

The experiment has four replications. Planting date was the main plot factor of the trial and was laid out according to a Latin square design with four superrows and four supercolumns, denoted by factors SR and SC, respectively (Fig. 3). Each superrow was divided into five longrows (LR), to which the five fertilizer levels were allocated. Furthermore, each main plot was divided into five columns (CL) to accommodate the five varieties tested in each year. The observational unit is the subplot (SP), located at the longrow×column intersections.

Figure 3

2.3. Statistical analysis

The treatment factors planting dates, fertilization doses can be treated as both qualitative (P, F) and quantitative (D, N) characteristics. Additionally, the factor variety (V) and year (Y) can be explored by its quantitative FAO classification (M) and time trend (T), respectively. For the latter two we do not expect that quantification can explore all variation seen within the factors. However, the aim of the current analysis is to explore the quantitative nature of all four factors. A general overview on model development is shown in Fig. 4. Two final models were developed: One model with a single response surface curve fitted across years and another model fitting separate response surface curves for good and bad years, with below average and above average yields, respectively.

Figure 4

2.3.1 Model development for single surface model

Block model

The statistical model used for analysis is developed here by first considering a single year and then extending to multiple years. To represent the field layout and allocation of treatments to experimental units, the following block model (Piepho et al. 2003) is used for a single year, using the block factors defined before

SR + SC + SR.SC + SR.LR + SR.SC.CL + SR.SC.LR.CL (1)

Here, SR.SC corresponds to main plots, and SR.SC.LR.CL corresponds to subplots. Hence, all six block factors shown in Fig. 3 are represented. All design effects in this block model are modelled as random. We extend the model to multiple year by extending each effect with the factor year (Y):

SR.Y + SC.Y + SR.SC.Y + SR.LR.Y + SR.SC.CL.Y + SR.SC.LR.CL.Y (2)

Year is denoted as the repeated factor (Piepho et al. 2004) because it indexes repeated observations on the same design unit. The design unit itself is identified by the level of the effect in (1) that is extended by factor Y in (2). For example, main plots are identified by levels of the effect SR.SC, and all observations on the same main plot are assumed to be serially correlated. Here, we use the first order autoregressive AR(1) model with all design effects in (2) except for the subplots where we additionally allow for heterogeneous variances (ARH(1)).

Treatment model

The treatment model is developed using the P, F and V factors included in the experiment. The basic treatment model for a single year is

V × F × P = V + F + P + V.F + V.P + F.P + V.F.P (3)

This model is modified to account for the random factor Y by adding a random main effect for Y and also adding all effects in (3) crossed with Y as random. Hence, the extended model is

V × F × P × Y = V + F + P + V.F + M.P + F.P + V.F.P + (4)

Y + V.Y + F.Y + P.Y + V.F.Y + V.P.Y + F.P.Y + V.F.P.Y

where random effects are listed after a colon. The full model is obtained by combining models (2) and (4).

Response surface regression

Models so far treated P, V, F and Y as qualitative factors. However, P can be quantified by day of year (D) and F by the amount of nitrogen (N). Additionally, V can be quantified by the FAO number M and Y can be quantified by the continuous variable T for calendar year. In the latter two cases, we do not expect that all differences in V and Y can be covered by M and T, respectively. Furthermore, note that preliminary inspection revealed that N shows a response with diminishing returns reaching a plateau and then dropping only slowly with further increasing N. A quadratic model in N may not represent this well. Hence, we experimented with different powers of N and decided to replace N with N^1/2. To simplify the presentation, we replaced N values by their square root but kept the symbol N to represent the factor. We fitted a second-order response surface model with all four variables D, N, M and T. Such a model is satisfying if there are no serious deviations from the response surface. To check this assumption, we first fitted a second-order response surface regression to single year data using D, M and N as quantitative regressor variables. In that first step, we fitted the treatment model given by

D + D.D + M + M.M + N + N.N + D.M + D.N + M.N + V.F.P (5)

separately for each year (Y) and assessed the lack of fit using the fixed effect V.F.P. In addition, the model comprised all random design effects in Eq. (1). The model fit was satisfactory (see Supplementary Material), hence we considered an extension of the second-order response surface model by inclusion of the continuous factor T for the calendar year. In this case, no lack-of-fit test (test of V.P.F.Y) was performed for the across-year analysis as we assumed that the trend will not explain years completely and there is for sure year-by-year variation that will be captured by the random effects involving Y in Eq. (4). Additionally, both variables Y and V were now assumed as random, while quantitative variables M and T were taken as fixed. The full model therefore included the second-order response surface regression on variables M, D, N and T as fixed effects, all effects from (4) including either V or Y as random effects, and all effects of (2) as random effects.

After fitting the response surface regression for all four quantitative variables, the fixed effects were subsequently pruned by discarding non-significant effects, observing the marginality principle. Thus, we started by inspecting the highest-order interaction and removed it if it was not significant (at α = 0.05), in which case we moved on to the nearest lower-order interactions to proceed with the next tests, etc. A summary of the covariates used in the regression analyses is presented in Table SM1.

2.3.2 Model development to fit separate response surface curves for good and bad years

For the final model developed above we inspected best linear unbiased predictions (BLUPs) for the random deviations of Y from trend T. Based on these BLUPs we fitted a separate response surface regression for (i) years with positive (= good years) and negative BLUP (= bad years). Again, fixed effects crossed with group were selected via backward selection.

The fitted models are reported as contour plots for two of the four variables (two out of the four variables D, M, N and T), fixing the other two at specific values. For analysis, we centered M at 350, T at 2010 and D at 120. This linear shift is intended to numerically stabilize the regression analysis. It does not affect slope estimates but does shift the intercept. The fitted values are not affected.

The statistical analysis were performed using ASReml 4.2 standalone for analysis and PROC RSREG in SAS for graphics.

Note, that in exploring possible temporal trends the effect of a total of 58 annual and monthly environmental factors (see examples in Fig. SM1) was also investigated. These factors were included as covariates in the single response surface model. As detailed in the Discussion section, weather related covariates (Table SM2.) were not used in the final model, since year factor (Y) has been shown to be a reasonably good integrator that aggregates the impact of weather factors and their variations with sufficient statistical confidence.

Though there are many reasons why results from a LTE can be useful for providing useful information on promising measures to adapt to a changing climate; yet, there are also limitations in view of various aspects: (i) shifts in future seasonality (i.e. shifts in rainfall patterns but also frost risk patterns, etc.) that are part of climate change projections, need to be considered when deriving potentially promising adaptation options from experiments conducted historical (past) weather conditions; (ii) effects of elevated atmospheric CO₂ concentration: although these have to be considered most for crops of the C3 photosynthetic type (like wheat or barley) (e.g. Lobell and Gourdji 2012; Rötter et al., 1999) elevated CO₂ also has beneficial effects on C4 crops like maize, in particular in improving their water use efficiency under drought conditions (e.g. Durand et al., 2018) as is also the case for C3 crops ( O'Leary et al., 2015). To overcome these shortcomings crop growth simulation was used for extrapolating the patterns detected in the LTE results for the future. Crop modeling has the potential to help us understand the relative influence of environmental factors (such as climate and soil) and genotype and management on the outcomes of long-term experiments. For example, Dobermann et al. (2000) explored this in their research.

2.4. Crop model simulations

The potential of planting date as a mitigation option was examined in depth using the Biome-BGCMuSo biogeochemical model. The Biome-BGCMuSo model is a general-purpose, process-based model that simulates the full carbon, nitrogen, and water budget of terrestrial ecosystems (Fodor et al., 2021; Hidy et al., 2022). Biome-BGCMuSo is a branch of the well-known Biome-BGC model, which was first developed by Running and Hunt (1993). The Biome-BGC model underwent significant enhancements and expansions in numerous aspects relative to its original formulation. The developments addressed a number of key areas, including soil processes, the introduction of management options, and the quantification of disturbance effects on plant physiology, as well as numerous other processes (Hidy et al., 2016). Additionally, the model has been enhanced to better simulate the effects of various stress factors, including drought, nitrogen, and heat stress. In cropland simulations, beyond meteorological, soil and crop input data, detailed management information (including the timing and amount of applied fertilizer, planting date, harvest date, residue management) is required for the simulations. Locally measured meteorological and soil data were used as model inputs, while observed plant related data (yield, maximum leaf area index (LAI_max), flowering date and harvest index) from the LTE were used for model calibration. The highly efficient Conditional Interval Reduction Method (CIRM) method (Hollós et al., 2022) was used for calibrating the length of the vegetative period, length of the reproductive period, and specific leaf area parameters, as well as for fine tuning of the partition of biomass into root/stem/leaf/kernel in the vegetative and in the reproductive period parameters. CIRM is a machine learning approach using decision tree-based white box approximations. CIRM effectively employs variables for which only limited data are available as constraints, enabling the identification of parameter space regions where simulations remain realistic (e.g., ensuring that LAI_max falls within a predefined interval based on observations). Yield data was used to minimize the difference between the observation and the simulation during calibration, while the rest of the observations (Table 3) was used as constraints to ensure realistic simulations with the calibrated model parameters. In line with the agromanagement options applied by the majority of Hungarian farmers in the study period, only data from treatments that met the following conditions were used in the calibration: N level: 120–180 kg/ha; hybrid FAO number: 300–400; planting date (DOY): between 105 and 120.

Table 3

Range of observed plant phenotypic data supporting crop model calibration used as constraints in the CIRM method. Flowering dates are given in DOY.
Observed feature	Minimum	Maximum	Observation period
Flowering date	176	192	2001–2022
Maximum of leaf area index	2.9	4.2	2001–2004 and 2017–2022
Harvest index	0.48	0.55	2005–2017

Table 3

With a minor change, we applied the method suggested by Ojeda et al. (2018) to assess model performance both in calibration and validation. The following statistical indicators were used: concordance correlation coefficient (CCC) defined by Lin (1989), mean absolute error (MAE) defined e.g. in Willmott et al. (2005) and mean signed error (MSE, also known as bias) which is also defined in Lin (1989) We decided to use MAE instead of root mean square error as the former has some advantages over the latter: MAE is a more natural and unambiguous measure of average error (Willmott et al., 2005).

During validation, simulations were carried out for the 2001–2020 period for a 10×10 km resolution grid covering the area of Hungary with 1014 cells. For each cell, soil and weather data were retrieved from the DOSoReMI (Pásztor et al., 2020) and the FORESEE (Kern et al., 2024) databases, respectively. For all simulations 150 kg/ha/year N fertilizer level and April 25 (DOY = 115) as planting date were used uniformly. Simulated yields were aggregated on NUTS-3 (county) level (EuroStat, 2024) and compared to the observed yield data retrieved from the database of the Hungarian Statistical Office.

Observed weather data were then substituted with data from climate projections (Fig. 5) produced by 5 different Global and Regional Climate Model (GCM-RCM) combinations driven by the RCP4.5 and RCP 8.5 scenarios (van Vuuren et al., 2011) for the 2041–2060 and for the 2081–2100 periods. A total of 20 simulations were carried out for the future, for each grid cells: 10 with the same agro-management options that were used during validation (BAU – Business As Usual) and 10 simulations with 3 weeks earlier planting (3WEP). The ten BAU and the ten 3WEP simulations were aggregated separately, and the results for the future time windows and the baseline period (2001–2020) were compared by using color-coded maps

Figure 5

3.1 Trends of climatic indicators

At the long-term experiment site, significant trends were found for the temperature related (Tmean and nrHotD) indicators as well as for Vapor Pressure Deficit. The expected number of hot days in the flowering period more than doubled and the mean temperature rose by more than 2°C during the 30 years of the study period (Fig. SM1). For all indicators for which a significant trend was identified at the Martonvásár site, the same trends were observed across the entire region, and those were significant for a considerable proportion of the area: 100%, 64.1% and 90.8% in case of Tmean, nrHotD and VPD, respectively. In light of this, it can be reasonably concluded that the climatic changes responsible for the observed effects at Martonvásár are likely to have a similar impact in the whole region under study. The subsequent modelling results serve to corroborate this conclusion (see section 3.5). There were no significant changes in the amount or distribution of precipitation for the whole year or for shorter periods within the year. Heat stress and atmospheric drought appear to be responsible for the adverse changes in yield levels.

3.2 Single response surface model

Our main fitted model, obtained after model selection, is reported in Table 4. Quadratic terms are significant for factors N, M and D, and the regression coefficients are negative. For time, only linear terms are significant, including the interactions M.T and D.T. The presence of these interactions means that the optimal sowing dates (D) as well as the optimal maturity class (M) change over time.

Table 4

Selected single response surface regression model for yield (Mg ha^− 1) with parameter estimates, standard errors and t-tests. The terms N, D, M and T represent quantitative variables fertilizer dose, planting day, FAO classification and time trend.
Effect^§	Estimate	Standard error	t-value	p-value
Intercept	6031.13	446.14	13.52	< 0.0001
N	435.61	38.1982	11.40	< 0.0001
N.N	-17.8776	2.4380	-7.33	< 0.0001
D	-23.4818	5.6154	-4.18	< 0.0001
D.D	-1.7090	0.3883	-4.40	< 0.0001
M	8.8217	3.7738	2.34	0.0194
M.M	-0.06767	0.02784	-2.43	0.0151
T	-47.0981	40.8646	-1.15	0.2491
N.D	-2.2058	0.2406	-9.17	< 0.0001
D.T	-1.1054	0.5460	-2.02	0.0429
M.T	-0.3239	0.1420	-2.28	0.0226
The terms N, D, M and T represent the square root of fertilizer dose (kg ha^− 1), planting time (DOY), FAO number and time trend. D, M and T were centered at 120, 350 and 2010, respectively.

Table 4

3.3 Time series analysis

The first and most important result of the developed model is that it predicts a clear decline in yield levels irrespective of the nutrition level, the maturity group and of the planting date (Fig. 6). Even with adequate nutrient supply, hybrid and planting date selection yield levels of over 10 tons per hectare in the early 1990s have fallen well below 9 tons in three decades (Fig. SM2). The shape of the iso-lines shows that the yield levels of late varieties decline at a much more intense rate than that of the early hybrids (Fig. 6B): compare the change of around 3 Mg ha^− 1 for the late varieties (FAO > 500) with the practically constant yield levels for the early hybrids (FAO < 300), over three decades.

The data analysis resulted in the following temporal trends for the optimum nitrogen fertilization level (N), the optimum FAO number (M) and for the optimum planting date (D).

$$\:{N}_{opt}\left(T\right)={(12.1831-0.0617\bullet\:(D-120\left)\right)}^{2}$$

$$\:{M}_{opt}\left(T\right)=-2.3932\bullet\:(T-2010)+415.18$$

$$\:{D}_{opt}\left(T\right)=-0.3234\bullet\:(T-2010)-0.6453\bullet\:{N}^{0.5}+113.13$$

The optimum level of N fertilization did not change significantly over time. Its value has stagnated at around 177 and 144 kg ha^− 1 (Fig. 6a) for early and late sowing, respectively (Fig. SM2). This observation simply reflects the fact that higher yield productions require more nitrogen inputs. It is important to note that this fertilization level corresponds with the highest average yield not with the maximum income. The maximum income based N fertilization optimum is closer to 120 kg ha^− 1 as above this level the yield achieved increases only slightly with increasing fertilizer rates (Fig. SM2). For a given planting date, the time invariant optimal level of nutrient supply corresponds with smaller and smaller yields.

Regarding variety selection, the optimal FAO number, providing the highest possible yield on average, is clearly decreasing with time (Fig. 6b). Before 2000, hybrids with FAO number over 450 gave the highest yields, irrespective of the nutrition level and the planting date (Fig. SM3). Today, the medium-early maturity group hybrids (FAO number less than 400) provide the highest possible yields

A similar clear trend could be observed in the optimum sowing date during the study period (Fig. 6c). Irrespective of hybrid selection the optimum sowing date shifted 10 days earlier during the 3 decades of the experiment. The benefit of earlier sowing dates is also reflected in the level of N fertilization resulting in a 4 days of average difference between the extensive (60 kg N ha^− 1 y^− 1) and the intensive (180 kg N ha^− 1y^− 1) nutrition regimes (Fig. SM4). Earlier sowing obviously entails a higher yield potential and thus higher N application required to realize that potential.

Figure 6

3.4 Analysis of inter-annual differences

Regardless of the year type (good or bad) and the fertilization level, hybrids of the same maturity group produce the maximum yield. Hybrids with around 420 FAO number comprise the optimal maturity group (Fig. SM5). Important to note, that this is an average for the 30 years. As it was showed earlier that the FAO number of hybrids with the highest yields clearly decreased over the observation period.

In bad years, the optimal planting date is more than two weeks earlier than in the high yielding years (Fig. SM5). The difference is more pronounced in stands fertilized intensively (DOY = 82 vs DOY = 108) than in stands fertilized extensively (DOY = 89 vs DOY = 111). The optimal sowing date does not depend at all on variety selection. There is a month difference between the earliest optimum planting date, corresponding to intensively fertilized hybrids in bad years, and the latest planting date optimum (extensively fertilized hybrids in good years). The flatness of the iso-lines in the direction of the planting date axis in sub-optimal years shows that in these years the further away we are from choosing the right hybrid (Fig. SM5), the less important the sowing time. According to the contour plots the planting date related results could be summarized as follows: (1) the worse the year the earlier the optimum planting date (Fig. SM5); (2) the earlier the planting date the higher the optimum N fertilization level as more fertilizer is needed to achieve the expected higher yields (Fig. SM6); (3) the higher the N fertilization level the more sensitive the yield level to planting date especially in suboptimal years (Fig. SM6).

In bad (under average yield) years, considerably less nitrogen is needed for maximum yields. Irrespective of hybrid or sowing date selection the adequate level of N fertilization is around 50 kg ha^− 1 less in sub-optimal years (210 vs 159 kg ha^− 1) corresponding to a slightly over 3 Mg ha^− 1 yield difference between the two types of years (Fig. SM6).

Traces of climate change were detectable during the three decades of the experiment, though no significant trend was found regarding precipitation and evapotranspiration related indicators. On the other hand, the frequency of weather extremes especially in the flowering period changed considerably in the past 30 years. The number of precipitation days show a clear though not significant declining trend in this period. In the first two decades of the experiment, there was one year per decade with less than 10 rainy days in the flowering season. In the last decade there were six such years. The number of hot days more than doubled during 30 years. As the technological level of cultivation has not changed during the study period, and the yield potential of the new varieties is certainly no worse than before, climate change is most likely the main cause of the observed changes. Mainly due to the increased heat stress coupled with considerable atmospheric drought around anthesis the expectable yield levels have decreased by more than 21%. Similar results have been reported in previous national (Fodor et al., 2014) and international (Webber et al., 2018) crop modelling studies.

The two models presented in this paper fit linear and quadratic terms for time trends. In order to explore the effect of environmental factors, we additionally used annual and monthly meteorological data as covariates in the final single response surface model. A total of 58 models adding one of the 58 covariates were fitted. The added covariate was not significant in most of the cases. In case of significance, the year variance was reduced up to 40% or increase up to 4%. Furthermore, time trend was increased or decreased by up to no more than 10%. No changes on other regression coefficients including time-by-FAO classification interactions were found (results not shown). Note that covariate values vary between years only, but not between plots. Thus, it is to be expected that their inclusion in the model can only affect time trend but no effect of other factors varied in the experiment. Further note that these additional analyses do not provide any causal inference, as the environmental covariate data is purely observational.

According to Shim et al. (2017) the decrease in kernel number accounted for a much greater contribution to the yield reductions due to temperature elevation than did the decrease in individual kernel weight in maize cultivars. Partial pollination caused by heat stress seems to be the actual cause of yield reductions that cannot be mitigated with higher nitrogen fertilization doses. Increasing fertilization doses above a certain level won’t result in higher yields and certainly won’t realize higher revenues. The stagnating nitrogen fertilization optimums coupled with the decreasing yield levels mean gradually increasing production costs. The fact that the same optimum N fertilization level is sufficient to achieve lower and lower yields calls into question the view that intensification could promote sustainable development in this climatic region.

Medium-early hybrids are less affected by the environmental changes than the late hybrids because their flowering phase overlaps much less with the critical, stressful period. Three decades ago, late varieties yielded 15 percent more than medium-early varieties, but nowadays medium-early varieties yield nearly 10 percent more.

Depending on external (abiotic stress status) and internal (maturity group) factors the planting date optima may differ by a month across the years. Even on average, the optimal planting date has now shifted into the first decade of April in accordance with other studies (Marcinkowski and Piniewski, 2018; Yasin et al., 2022). As the likelihood of unfavorable years is expected to increase in the future, earlier planting, even before 1st of April, may become an effective mitigation option. Since the likelihood of extreme weather events is also expected to increase with climate change there is still the question of whether the simple 'early sowing' as a mitigation option will be feasible at all, despite the late frosts that may occur. The chance of late frosts (days with T_min < 0°C) in the 03.21–04.10 period is currently around 10% at the study area. According to 10 available climate projections for the region (Kern et al., 2023), toward the end of the century, this likelihood is estimated to decrease down to 3.6 and 1.0%, whether RCP4.5 or RCP8.5 scenarios are considered, respectively.

3.5 Crop modelling results

Performance of the calibrated Biome-BGCMuSo model in simulating maize yield is demonstrated in Fig. 7. After calibrating the selected plant specific parameters, the model was capable of estimating the observed values with a comparable efficiency reported in similar studies (Bao et al. 2017; Sándor et al. 2017; Soufizadeh et al. 2018; Diancoumba et al., 2024). Figure 7. also conveys an important message: The Biome-BGCMuSo model cannot accurately calculate county-level yields from year to year, but it can simulate average yields over longer periods with reasonable accuracy at NUTS-3 level. For the purpose of this study, this latter capability of the model is sufficient, as we only want to predict the trend of changes in average yields over longer periods.

Figure 7

Mainly due to the mid-summer heat waves and the droughty Augusts that are becoming more and more frequent, maize yields are projected to decline significantly towards the end of the century (Fig. 8.). In particular, regions with above-average yields, immediately west of the Danube and in the south-west, are expected to suffer significant yield losses of more than 25%. These results are in good agreement with previous modelling studies showing that, without mitigation strategies (BAU management), climate change is expected to have negative impacts on maize in the Carpathian basin (Webber et al., 2018). However, earlier sowing, an easy and inexpensive change in management (Minoli et al., 2022), can reduce yield losses to below 15% in almost the whole study area. In the wetter areas of the region, western and north-eastern Hungary, this mitigation option can even fully offset the negative impacts of climate change (Fig. 8.).

Figure 8

We may conclude that late hybrids seem to have no perspective in the Pannonian climatic zone.

Early sowing, shifting the planting date even into the last decade of March, will come only with a marginal chance of losing crop due to frost damages when approaching the end of the century.

Sub-optimal environmental conditions may greatly change the effect of certain agro-management factors. In bad years the differences in hybrid selection or in the level of nitrogen fertilization may result in a much greater impact on the yield than in good years (Fig. SM4-5). When planning nitrogen fertilization levels the planned planting date also should be taken into account as it clearly influences the fertilizations level optimum. Additionally, fertilization recommendations could be adjusted after a bad year to account for the considerable amount of nutrients that was not taken up, taking into account the possible immobilization. The harmonization of planting date, fertilization level and variety selection for obtaining the achievable yield is crucial especially in bad years. Generally speaking, the determination of the optimum of any of the investigated factors is only possible if the other two are taken into account. This principle should be taken into account in the next generation of plant production related advisory systems. This is the first comprehensive study that combines long-term historic weather data, high-resolution soil data, climate projection data as well as statistical and crop simulation modelling tools in order to provide reliable mitigation strategies for farmers and policy makers.

Acknowledgement We gratefully acknowledge the excellent technical assistance provided by the staff of the Centre for Agricultural Research, Crop Production Department, Martonvásár.

Authors’ contributions

Klára Pokovai: Methodology, Writing, Editing ,

Hans-Peter Piepho: Conceptualization, Data analysis, Editing

Jens Hartung: Formal analysis, Visualization, Writing

Tamás Árendás: Data curation, Supervision

Péter Bónis: Resources, Investigation

Eszter Sugár: Resources, Investigation

Roland Hollós: Data analysis, Software, Editing

Nándor Fodor: Funding acquisition, Project administration, Conceptualization, Visualization, Editing

All the authors read and approved the final manuscript.

Funding This work was supported by Széchenyi 2020 programme, the European Regional Development Fund ‘Investing in your future’, the Hungarian Government: [grant number GINOP-2.3.2-15-2016-00028] as well as by the TKP2021-NKTA-06 project that has been implemented with the support provided by the Ministry of Innovation and Technology of Hungary from the National Research, Development and Innovation Fund, financed under the [TKP2021-NKTA] funding scheme.

Data availability The datasets generated during and/or analyzed during the current study are not publicly available but are available from the authors on reasonable request.

Code availability Simulations were undertaken with the BiomeBGC-MuSo (v6.1) model. The code is available on request but permission is required to change it.

Conflict of interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Ethics approval Not applicable.

Consent to participate Not applicable.

Consent for publication Not applicable.

Allen RG, Pereira LS, Raes D et al. (1998) Crop Evapotranspiration-Guidelines for Computing Crop Water Requirements. FAO Irrigation and Drainage Paper 56. Food and Agriculture Organization of the United Nations (FAO), Rome. https://www.fao.org/4/X0490E/x0490e00.htm
Bao Y, Hoogenboom G, McClendon R et al. (2017) A comparison of the performance of the CSM-CERES-Maize and EPIC models using maize variety trial data. Agricultural Systems 150:109–119. https://doi.org/10.1016/j.agsy.2016.10.006
Bassu S, Fumagalli D, Toreti A et al. (2021) Modelling potential maize yield with climate and crop conditions around flowering. Field Crops Res 271:108226. https://doi.org/10.1016/j.fcr.2021.108226
Berti A, Dalla Marta A, Mazzoncini M, Tei F (2016) An overview on long-term agroecosystem experiments: present situation and future potential. Eur J Agron 77:236–241. https://doi.org/10.1016/j.eja.2016.01.004
Blanchy G, D’Hose T, Donmez C et al. (2024) An open-source metadataset of running European mid- and long-term agricultural field experiments. Soil Use Manag 40:e12978. https://doi.org/10.1111/sum.12978
Buhiniˇcek I, Kauˇci ́c D, Kozi ́c Z et al. (2021) Trends in Maize Grain Yields across Five Maturity Groups in a Long-Term Experiment with Changing Genotypes. Agric 11:887.
https://doi.org/10.3390/ agriculture11090887
Choruma DJ, Balkovic J, Odume ON (2019) Calibration and Validation of the EPIC Model for Maize Production in the Eastern Cape, South Africa. Agron 9:494. https://doi.org/10.3390/agronomy9090494
Cuculeanu V, Marica A, Simota C (1999) Climate change impact on agricultural crops and adaptation options in Romania. Clim Res 12:153-160. https://www.int-res.com/articles/cr/12/c012p153.pdf
Diancoumba M, Kholová J, Adam M et al. (2024) APSIM-based modeling approach to understand sorghum production environments in Mali. Agron Sustain Dev 44:25. https://doi.org/10.1007/s13593-023-00909-5
Dobermann A, Dawe D, Rötter R et al. (2000) Reversal of rice yield decline in a long-term continuous cropping experiment. Agron J 92:633-643. doi:10.2134/agronj2000.924633x
Donmez C, Schmidt M, Cilek A et al. (2023) Climate change impacts on long-term field experiments in Germany. Agric Syst 205:103578. https://doi.org/10.1016/j.agsy.2022.103578
Durand JL, Delusca K, Boote K et al. (2018) How accurately do maize crop models simulate the interactions of atmospheric CO2 concentration levels with limited water supply on water use and yield? Eur J Agron 100:67-75. doi: 10.1016/j.eja.2017.01.002
EC 2022a. Cereals in EU. https://agriculture.ec.europa.eu/farming/crop-productions-and-plant-based-products/cereals_en Accessed on 14 Dec 2022
EC 2022b. EU’s cereals consumption. https://agriculture.ec.europa.eu/farming/crop-productions-and-plant-based-products/cereals_en Accessed on 14 Dec 2022
Eckersten H, Herrmann A, Kornher A et al. (2012) Predicting silage maize yield and quality in Sweden as influenced by climate change and variability. Acta Agriculturae Scandinavica, Section B — Soil & Plant Sci 62:151–165. https://doi.org/10.1080/09064710.2011.585176
EuroStat 2024. NUTS - Nomenclature of territorial units for statistics. https://ec.europa.eu/eurostat/web/nuts/ Accessed on 19 July 2024
Fodor N, Pásztor L, Németh T (2014) Coupling the 4M crop model with national geo-databases for assessing the effects of climate change on agro-ecological characteristics of Hungary. Int J Digit Earth 7:391–410. https://doi.org/10.1080/17538947.2012.689998
Fodor N, Pásztor L, Szabó B et al. (2021). Input database related uncertainty of Biome-BGCMuSo agro-environmental model outputs. Int J Digit Earth 14(11):1582–1601. https://doi.org/10.1080/17538947.2021.1953161
Ginaldi F, Bindi M, Dalla Marta A et al. (2016) Interoperability of agronomic long term experiment databases and crop model intercomparison: the Italian experience. Eur J Agron 77:209-222. https://doi.org/10.1016/j.eja.2016.02.007.
Grosse M, Hierold W, Ahlborn MC et al. (2020) Long-term field experiments in Germany: classification and spatial representation. Soil 6:579–596. https://doi.org/10.5194/soil-6-579-2020
Hidy D, Barcza Z, Hollós R et al. (2022) Soil-related developments of the Biome-BGCMuSo v6.2 terrestrial ecosystem model. Geosci Mod Dev 15:2157–2181. https://doi.org/10.5194/gmd-15-2157-2022
Hidy D, Barcza Z, Marjanovic H et al. (2016) Terrestrial ecosystem process model Biome-BGCMuSo v4.0: summary of improvements and new modeling possibilities. Geosci Mod Dev 9:4405–4437. https://doi.org/10.5194/gmd-9-4405-2016
Hollós R, Fodor N, Merganičová K et al. (2022) Conditional interval reduction method: a possible new direction for the optimization of process based models. Environ Modell Softw 158:105556. https://doi.org/10.1016/j.envsoft.2022.105556
IUSS Working Group (2015) World Reference Base (WRB) for Soil Resources. International soil classification system for naming soils and creating legends for soil maps. World Soil Resources Reports 106, FAO, Rome.
Jugenheimer RW (1958) Hybrids maize breeding and seed production. FAO Agric Dev Pop 62:99–103.
Kern A, Dobor L, Hollós R et al. (2024) Seamlessly combined historical and projected daily meteorological datasets for impact studies in Central Europe: the FORESEE v4.0 and the FORESEE-HUN v1.0. Clim Serv. 33:100443. https://doi.org/10.1016/j.cliser.2023.100443
Kersebaum K, Boote K, Jorgenson J et al. (2015) Analysis and classification of data sets for calibration and validation of agro-ecosystem models. Environ Modell Softw 72:402–417. http://dx.doi.org/10.1016/j.envsoft.2015.05.009
KSH 2022. Harvest area of major crops in Hungary (in Hungarian) https://www.ksh.hu/docs/hun/xftp/stattukor/vet/20210601/index.html Accessed on 14 Dec 2022
Li X, Storkey J, Mead A, Shield I, Clark I, Ostler R, Roberts B, Dobermann A (2023) A new Rothamsted long‑term feld experiment for the twenty‑frst century: principles and practice. Agron Sustain Dev 43:60. https://doi.org/10.1007/s13593-023-00914-8
Liang S, Li Y, Zhang X et al. (2018) Response of crop yield and nitrogen use efficiency for wheat-maize cropping system to future climate change in northern China. Agric For Meteorol 262:310–321. https://doi.org/10.1016/j.agrformet.2018.07.019
Lobell DB, Gourdjii SM (2012) The influence of climate change on global crop productivity. Plant Physiol 160:1686−1697.doi:10.1104/pp.112.208298
Macholdt J, Piepho HP, Honermeier B et al. (2020) The effects of cropping sequence, fertilization and straw management on the yield stability of winter wheat (1986–2017) in the Broadbalk Wheat Experiment, Rothamsted, UK. J Agric Sci 158:65–79. https://doi:10.1017/S0021859620000301
Marcinkowski P, Piniewski M (2018) Effect of climate change on sowing and harvest dates of spring barley and maize in Poland. Int Agrophys 32(2):265–271. https://doi.org/10.1515/intag-2017-0015
Marton TA, Kis A, Zubor-Nemes A et al. (2020) Human Impact Promotes Sustainable Corn Production in Hungary. Sustainabilty 12:6784. https://doi.org/10.3390/su12176784.
Mereu V, Gallo A, Trabucco A et al. (2021) Modeling high-resolution climate change impacts on wheat and maize in Italy, Clim Risk Manage 33:100339. https://doi.org/10.1016/j.crm.2021.100339.
Minoli S, Jägermeyr J, Asseng S et al. (2022) Global crop yields can be lifted by timely adaptation of growing periods to climate change. Nat Commun 13:7079. https://doi.org/10.1038/s41467-022-34411-5
Moore FC, Lobell DB (2014) The Adaptation Potential of European Agriculture in Response to Climate Change. Nat Clim Chang 4:610–614. DOI: 10.1038/nclimate2228
Ojeda JJ, Volenec JJ, Brouder SM et al. (2018) Modelling stover and grain yields, and subsurface artificial drainage from long-term corn rotations using APSIM. Agric Water Manage 195:154-171. https://doi.org/10.1016/j.agwat.2017.10.010
O'Leary GJ, Christy B, Nuttall J et al. (2015) Response of wheat growth, grain yield and water use to elevated CO2 under a Free-Air CO2 Enrichment (FACE) experiment and modelling in a semi-arid environment. Glob Change Biol 21:2670-2686. doi: 10.1111/gcb.12830
Olesen JE, Trnka M, Kersebaum KC et al. (2011) Impacts and adaptation of European crop production systems to climate change. Eur J Agron 34:96–112. https://doi.org/10.1016/j.eja.2010.11.003.
Parent B, Leclere M, Lacube S et al. (2018) Maize yields over Europe may increase in spite of climate change, with an appropriate use of the genetic variability of flowering time. Proc Natl Acad Sci 115:10642–10647. https://doi.org/10.1073/pnas.1720716115.
Parker PS, Shonkwiler JS, Aurbacher J (2017) Cause and Consequence in Maize Planting Dates in Germany. J Agro Crop Sci 203:227–240. https://doi.org/10.1111/jac.12182.
Pásztor L, Laborczi A, Takács K et al. (2020) Progress in the elaboration of GSM conform DSM products and their functional utilization in Hungary. Geoderma Reg 21. ISSN 2352-0094. https://doi.org/10.1016/j.geodrs.2020.e00269
Pereyra‑Goday F, Jebari A, Takahashi T, Rovira P, Ayala W, Lee MRF, Rivero MJ, McAulife GA (2024) Carbon footprint of mixed farming crop‑livestock rotational‑based grazing beef systems using long term experimental data. Agron Sustain Dev 44:41. https://doi.org/10.1007/s13593-024-00977-1
Piepho HP, Büchse A, Emrich K (2003) A Hitchhiker's Guide to Mixed Models for Randomized Experiments. J Agron Crop Sci 189:310–322. https://doi.org/10.1046/j.1439-037X.2003.00049.x
Piepho HP, Büchse A, Richter C (2004) A Mixed Modelling Approach for Randomized Experiments with Repeated Measures. J Agron Crop Sci, 190:230–247. https://doi.org/10.1111/j.1439-037X.2004.00097.x
Pongrácz R, Bartholy J, Miklós E (2011) Analysis of projected climate change for Hungary using ensembles simulations. Appl Ecol Env Res 9:387–398. DOI: 10.15666/aeer/0904_387398
Rasmussen PE, Goulding KWT, Brown JR et al. (1998) Long-term agroecosystem experiments: assessing agricultural sustainability and global change. Sci 282:893–896. https://doi.org/10.1126/science.282.5390.893.
Reckling M, Döring TF, Bergkvist G et al. (2018) Grain legume yields are as stable as other spring crops in long-term experiments across northern Europe. Agron Sustain Dev 38:63. https://doi.org/10.1007/s13593-018-0541-3.
Running SW, Hunt ERJ (1993) Generalization of a forest ecosystem process model for other biomes, BIOME-BGC, and an application for global-scale models. In: Ehleringer, J.R. and Field, C. (eds.) Scaling Physiological Processes: Leaf to Globe. Academic Press, San Diego, 141–158.
Russelle MP, Olson RA, Hauck RD (1987) Planting date and nitrogen management interactions in irrigated maize. Field Crops Res 16:349–362. https://doi.org/10.1016/0378-4290(87)90072-4.
Sándor R, Barcza Z, Acutis M et al. (2017) Multi-model simulation of soil temperature, soil water content and biomass in Euro-Mediterranean grasslands: Uncertainties and ensemble performance. Eur J Agron 88:22–40. http://dx.doi.org/10.1016/j.eja.2016.06.006
Shim D, Lee KJ, Lee BW (2017) Response of phenology- and yield-related traits of maize to elevated temperature in a temperate region. The Crop J. 5:305–316. https://doi.org/10.1016/j.cj.2017.01.004.
Soufizadeh S, Munaro E, McLean G et al. (2018) Modelling the nitrogen dynamics of maize crops – Enhancing the APSIM maize model. Eur J Agron 100:118–131. https://doi.org/10.1016/j.eja.2017.12.007
Tsimba R, Edmeades GO, Millner JP, Kemp PD (2013) The effect of planting date on maize grain yields and yield components. Field Crops Res. 150:135–144. https://doi.org/10.1016/j.fcr.2013.05.028.
Webber H, Ewert F, Olesen JE et al. (2018) Diverging importance of drought stress for maize and winter wheat in Europe. Nat Commun 9:4249. https://doi.org/10.1038/s41467-018-06525-2.
Willmott CJ, Matsuura K (2005) Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Clim Res 30:79–82. doi:10.3354/cr030079
Yasin M, Ahmad A, Khaliq T et al. (2022) Climate change impact uncertainty assessment and adaptations for sustainable maize production using multi-crop and climate models. Environ Sci Pollut Res 29:18967–18988. https://doi.org/10.1007/s11356-021-17050-z
Zhu G, Liu Z, Qiao S et al. (2022) How could observed sowing dates contribute to maize potential yield under climate change in Northeast China based on APSIM model. Eur J Agron 136:126511. https://doi.org/10.1016/j.eja.2022.126511.
Žydelis R, Weihermüller L, Herbst M (2021) Future climate change will accelerate maize phenological development and increase yield in the Nemoral climate. Sci Tot Environ 784:147175. https://doi.org/10.1016/j.scitotenv.2021.147175.

SupplementaryMaterial2024ASD.docx

Download PDF

Reviewers agreed at journal
31 Oct, 2024
Reviewers invited by journal
31 Oct, 2024
Editor invited by journal
31 Oct, 2024
Editor assigned by journal
22 Oct, 2024
First submitted to journal
21 Oct, 2024

You are reading this latest preprint version

Climate change related lessons learned from a long-term field experiment with maize

Status:

Version 1

Abstract

Figures

1. Introduction

2. Material and Methods

2.1. Soil and climatic characteristics of the experiment area

2.2. Experimental design

2.3. Statistical analysis

2.3.1 Model development for single surface model

2.3.2 Model development to fit separate response surface curves for good and bad years

2.4. Crop model simulations

3. Result and Discussion

3.1 Trends of climatic indicators

3.2 Single response surface model

3.3 Time series analysis

3.4 Analysis of inter-annual differences

3.5 Crop modelling results

4. Conclusion

Declarations

References

Supplementary Files

Status:

Version 1