Assessment of the Effect of Climate Change on the Productivity of Rainfed Maize (Zea mays L.) Using Crop Growth Model AquaCrop in Central Mexico

Abstract

Rainfed agriculture in Mexico is characterized by the threat of crop failure due to multiple stress factors, the most significant of which are adverse climatic conditions. Maize is one of the most important staple crops, used for both consumption and livelihoods. However, climate variability and change are threatening to exacerbate yield gaps and increase variability in yields from one year to the next, particularly due to changes in temperature and rainfall. Therefore, using the CMIP6 GCM ACCESS-ESM1-5 and the AquaCrop model, the impacts of climate change on maize yields were assessed for the SSP2-4.5 and SSP5-8.5 scenarios in the study area known as the Atlacomulco Rural Development District (ARDD), considering a projection towards three horizons, 2021–2040, 2041–2060, and 2061–2080, compared to a historical period (1985–2020). Through the metrics used for the validation of the AquaCrop model, it was possible to determine that the simulated values were satisfactorily adjusted to the yields measured by the Agricultural and Livestock Information System (SIAP) from 2003 to 2020 (0.7 ton ha⁻¹ < RMSE > 1.3 ton ha⁻¹; 0.1 < d-index > 0.8; 0.2% < NRMSE > 0.7%; −0.7 < EF > 0.3 and 0.4 < r > 0.8). Future climate change estimates in the ARDD indicate that the average temperatures could increase by between 1.8 and 2.8 °C, while the effective precipitation tends to decrease by up to 7.5−11% through SSP2-4.5 and SSP5-8.5, respectively, during the agricultural cycle. The model results indicate that by the year 2080, maize production in the ARDD will increase from 3.5 ton ha⁻¹ (historical) to 4.2 ton ha⁻¹ and 4.4 ton ha⁻¹, which represents an increase of 21% and 24% for SSP2.4.5 and SSP5-8.5, respectively.

1. Introduction

According to the Intergovernmental Panel on Climate Change (IPCC)’s sixth assessment report, from 1850 to 2020, the global average surface temperature increased by 1.09 °C. This increase in temperature is expected to persist in the years to come [1]. Due to rising global temperatures, severe weather events are becoming more frequent and intense, causing the ecological consequences of climate change to become increasingly severe and adversely affecting the sustainable progress of the humankind [2].

The agricultural sector is the most vulnerable environment to climate change, as it has experienced the most negative impacts with the highest recovery cost, demonstrating its low resilience to and a high degree of risk from this plausible climate threat. This situation will affect global food security [2,3]. Several extreme weather events, such as droughts, heat waves, erratic rainfall, storms, floods, and increased insect and disease infestations, have adversely affected farmers’ livelihoods [4].

The phenomenon of climate change is manifested by an increase in both daytime and nighttime temperatures, as well as variations in rainfall [5]. These alterations affect crop production, causing both abiotic stresses directly (water stress) and biotic stresses derived from insect, pest, and weed pressure indirectly. Such circumstances can have a significant impact on crop yield and quality, depending on the geographical location [6]. In the case of maize (Zea mays L.), the element of climate change greatly affects the growth and yield of the crop, resulting in a gradual decline in its productivity globally, compromising the economic viability of the existing production technologies and human food security [7].

Maize is the second most cultivated cereal globally after wheat. In 2023, a production of approximately 1.241 million tons of maize was recorded on 208 million hectares [8]. Of this amount, ca. 80% was harvested under rainfed conditions, implying the heavy dependence of this type of agriculture on rainfall as the only source of water supply [9]. For diverse uses in food, forage, industrial manufacturing, and as a source of energy, maize has become a high-value, high-demand crop whose consumption needs are expected to increase by 2050 in line with global population growth [10].

In Mexico, 73% of the area devoted to maize cultivation is under rainfed conditions, covering around 5.9 million hectares, mainly in rural areas with a high level of marginalization [11,12]. Annually, in the State of Mexico, around 430 thousand hectares are cultivated under this same production system, generating ca. 1.1 million tons of maize, of which the Atlacomulco Rural Development District (ARDD) contributes 400 thousand tons (approximately 38%), harvested on 115 thousand hectares [11,13]. Therefore, this region is considered the agricultural breadbasket of the State of Mexico, and its production directly influences food security and the local economy in central Mexico [14].

Between 1985 and 2017, the mean annual temperature in the ARDD increased by 0.93 °C, and precipitation also increased by up to 8.5% compared to the historical mean. These increases show an upward trend, mostly pronounced in the last decade [15]. The increase in ambient temperature and the variability of precipitation patterns have caused a significant impact on the phenological stages, biomass, and grain yield of the corn crop [5,16]. To assess the future impacts of climate change on a crop and to have the basis for proposing mitigation and resilience measures, it is necessary to rely on computational tools for modeling crop growth as a function of agroclimatic variables [17].

Process-based crop simulation models are important resources to help represent the understanding and knowledge of a current or future cropping system. Crop modeling systems are intended to help study crop growth and development in response to environmental variables and are able to project how changes in the environment will affect growth and yield [18,19,20]. In addition, one or more environmental conditions can be modified to anticipate the reaction of a specific crop in different environments [21].

Significant progress has been made in optimizing crop models to simulate yield due to water input, such as CERES-corn [22], CropSyst [23], STICS [24], and WOFOST [25]. These models are sufficiently sophisticated to understand the sustainability of management practices and response to environmental impacts such as climate change [7,19]. These models are mainly employed by researchers and trained users in commercial or precision agriculture, so they are considered advanced and robust, requiring a wide range of input parameters for calibration and validation, which are not readily available for the diversity of crops and sites worldwide [26]. Based on the above premise, the FAO (Food and Agriculture Organization of the United Nations) Water Unit has developed the AquaCrop model [27,28].

AquaCrop is a crop growth model that simulates how water affects the crop’s yield, whose outputs are based on soil moisture. It requires fewer input parameters, combines simplicity and accuracy, has a lower probability of error, and can be used widely across time and space [29]. Previous studies showed that AquaCrop can be a good tool for the prediction of maize crop growth and yield in rainfed agriculture [20,30,31]. In Mexico, important research on maize crop modeling has been carried out in the main irrigation districts of the country; however, research gaps have been identified in rainfed agricultural regions where, due to climate variability, maize production is more uncertain [32,33].

By summarizing the above preview and discussion, the present study was planned with the following objectives: (1) to calibrate and validate the AquaCrop model for maize cultivation under rainfed conditions in the ARDD and (2) to assess the future effect on maize crop yields caused by climate change using information from CMIP6.

2. Materials and Methods

2.1. Study Area

The study area belongs to the Atlacomulco Rural Development District (ARDD), composed of 10 municipalities: Temascalcingo, Ixtlahuaca, Atlacomulco, San Felipe del Progreso, Acambay, El Oro, Jocotitlán, Jiquipilco, Morelos, and San José del Rincón, which belong to the State of Mexico (Figure 1) [11]. These municipalities belong to the Upper Lerma River Basin, whose topography is made up of plains, steep valleys and large volcanic structures, with elevations ranging from 2230 m to 3800 m and an average altitude of 2600 m [34]. The predominant climate is subhumid temperate with summer rainfall (CW), with an average annual rainfall of 780 mm·yr⁻¹, the average annual maximum temperature of 21.5 °C, and the average minimum temperature of 5.1 °C [15].

2.2. Agricultural Data

Rainfed maize cultivation in the ARDD is developed during the annual spring–summer (P–S) agricultural cycle, which begins parallel to the onset of the first seasonal rainfall, which starts in May and culminates in October. During these months, approximately 650 mm are accumulated [35]. Maize cultivation takes between 160 to 180 days from sowing to physiological maturity, depending on the genetic variety. It is common to use native seeds and, to a lesser extent, hybrid seeds for rainfed maize cultivation [36,37].

The cultural tasks carried out in corn cultivation consist of fallowing, harrowing, sowing, weeding, and fertilization, which are carried out according to the cultivation period. In this region, chemical fertilization is common, using moderate doses of nitrogen, phosphorus, and potassium. Likewise, weed and pest control is carried out through agrochemicals [36,38]. In general, maize has a traditional and common management, which is shared by most farmers in the region. According to the World Reference Base (WRB) categorization system, five categories of reference soils predominate in the ARDD: Regosols, Cambisols, Vertizols, Umbrisols, and Phaeozems, which usually refer to a minimum development of the fertile horizon [39].

2.3. Climatic Data

Meteorological data on precipitation, maximum and minimum temperature in the ARDD from the daily record of 36 years (1985–2020) at 10 meteorological stations of the National Meteorological Service (NMS) [40] and the National Institute of Forestry, Agricultural, and Livestock Research (INIFAP) [41] were used (

Table 1. Meteorological stations located in the ARDD.

Station Code	Municipality	Longitude	Latitude	Altitude
31	Ixtlahuaca	−99.818	19.631	2545
35	San José del Rincón	−100.125	19.634	2692
15076	San Felipe del Progreso	−99.958	19.663	2564
15128	El Oro	−100.081	19.812	2601
15239	Morelos	−99.640	19.753	2831
15251	Atlacomulco	−99.874	19.798	2574
15264	Jiquipilco	−99.668	19.616	2576
36756	Jocotitlán	−99.713	19.663	2560
37673	Acambay	−99.885	19.915	2513
860124	Temascalcingo	−100.027	19.933	2379

). To control the quality of the meteorological data, a process of homogenization and filling in missing data for each station was carried out using the R-CLIMATOL package (https://www.climatol.eu/ accessed on 10 June 2024), which is described in detail by Cruz-González et al. [13]. In this research, only the analysis of the climatic information concerning the agricultural cycle (May–October) was considered.

2.4. Climate Change Scenarios

The Long Ashton Research Station Weather Generator v8.0 (LARS-WG) is a stochastic weather generation tool that performs statistical downscaling of the climate variables of maximum temperature, minimum temperature, and precipitation [42]. LARS-WG has been successfully used to simulate meteorological variables at specific locations for the present and future climate scenarios provided by general circulation models (GCMs). It was developed to investigate the effect of climate change by simulating daily time series data and is considered a basic, but flexible and computationally efficient software [43]. As reported by Semenov et al. [42], the configuration of LARS-WG in the ARDD required three processes: calibration, validation, and meteorological production of future scenarios.

LARS-WG 8.0 provides 5 models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) ensemble, which can be used in climate prediction: CNRM-CM6-1, HadGEM3-GC31-LL, MRI-ESM2-0, ACCESS-ESM1-5, and MPI-ESM1-2-LR. Of these GCMs, ACCESS-ESM1-5 was used in the present study, which was previously evaluated in the ARDD by Cruz-González et al. [13], showing a good performance in the simulation of historical precipitation and temperature, in addition to being a GCM with the largest decrease in precipitation and the largest increase in the mean temperature, resulting in a climate scenario with the driest and most erratic conditions for rainfed maize production in the ARDD. The scenarios were projected from two shared socioeconomic pathways (SSPs), SSP2-4.5 (medium forcing) and SSP5-8.5 (high forcing), considering three time horizons: near (2021–2040), medium (2041–2060), and distant (2061–2080) [1]. The research line of this study is shown in Figure 2.

2.5. Reference Evapotranspiration (ET₀)

The study for the calculation of reference evapotranspiration in the maize crop in the ARDD study area used the Ref-ET version 4.1 software (reference evapotranspiration calculator programming) [44], calculating this variable for the historical period and climate change scenarios. The Hargreaves–Samani (HS) method was selected as shown in Equation (1) [45]:

$${\mathrm{ET}}_0=0.0023 ({\mathrm{T}}_{\max }-{\mathrm{T}}_{\min }{)}^{0.5}\left({\mathrm{T}}_{\mathrm{avg} }+17.8\right) {\mathrm{R}}_{\mathrm{a}}$$

(1)

where ET₀ = reference evapotranspiration mm·day⁻¹; T_max = maximum temperature, °C; T_min = minimum temperature, °C; T_AVG = mean temperature, °C; and R_a = extraterrestrial solar radiation, mm·day⁻¹.

2.6. FAO AquaCrop Model

AquaCrop version 7.1 [17] was used in this study. To use this program for grain yield simulation in a rainfed crop (Figure 2), it is necessary to input, at a minimum, meteorological data, crop phenological development, growth-related factors, management practices, and soil properties [19,30,33]. In AquaCrop, four key variables are calculated continuously and daily using individual equations. These include the total canopy cover (CC), daily plant transpiration, accumulated aboveground biomass, and final cumulative dry yield, which are interrelated by empirical factors such as water productivity (WP) and harvest index (HI), which convert transpiration to biomass and biomass to yield, respectively [28,46]. AquaCrop uses a function of normalized water productivity and accumulated transpiration over the growing season to estimate biomass (Equation (2)).

$$\mathrm{B}=\mathrm{W}\mathrm{P}\sum {\displaystyle \frac{\mathrm{T}\mathrm{r}}{{\mathrm{E}\mathrm{T}}_0}}$$

(2)

where B = aboveground biomass (kg/m²); WP = normalized water productivity (kg/(m²⋅mm)); Tr = crop transpiration (mm); and ET₀ = reference evapotranspiration (mm). To calculate potential crop transpiration in AquaCrop, the crop coefficient (Kc) is multiplied by the reference evapotranspiration (ET₀).

Yield is calculated by adjusting the harvest index (a measure of the proportion of harvestable product relative to the total aboveground biomass) to determine crop production (Equation (3)):

$$\mathrm{Y}=\mathrm{H}{\mathrm{I}}_0\mathrm{B}$$

(3)

where Y = yield (kg/m²) and HI₀ = reference harvest index (%).

The AquaCrop model was selected in this research as it presents a water-driven growth module that allows the simulation of achievable yields as a function of water consumption under fully irrigated, deficit/limited irrigation and rainfed conditions, which is an advantage over other crop models [46]. The maize crop has a default calibration in this model [28]; however, some parameters were adjusted to rainfed conditions. Likewise, AquaCrop calculates the effective precipitation (EP) using the USDA [47] method.

2.6.1. Calibration and Validation Process: AquaCrop Model

The calibration protocol adopted in this study adheres to the guidelines described in the AquaCrop reference manual and in the FAO Irrigation and Drainage Paper No. 66, Crop Yield Response to Water. This process was carried out through trial and error iterations, that is, achieving a fit of the model to the desired conditions [46,48].

To perform the calibration of the AquaCrop model, we took into account two classes of parameters: conservative and non-conservative. In the first instance, the conservative parameters were kept constant under different crop conditions. In the present research, they were adjusted by taking as reference case studies on rainfed maize production reported by Adeboye et al. [17], Eshete et al. [18], and Sandhu & Irmak [20] (

Table 2. Calibrated conservative parameters of the AquaCrop model for rainfed maize in the ARDD.

Conservative Parameters	Default	Calibrated	Unit
Base temperature	8	7	°C
Cutoff temperature	30	27	°C
Canopy cover per seedling	6.5	6.5	cm2 plant−1
Canopy growth coefficient (CGC)	16.3	14.9	% day−1
Maximum canopy cover (CCx) (fraction soil cover)	0.96	0.7	%
Canopy decline coefficient (CDC)	11.7	11.7	% day−1
Water productivity normalized for ETo and CO2 (WP *)	33.7	33.7	(g/m2)
Reference harvest index (HIo)	48	27
Shape factor for the water stress coefficient for canopy expansion	4	2.9

). The non-conservative parameters present variations depending on the annual season, region, crop type, and crop management. In our case, these parameters were compiled from Vázquez-Carrillo et al. [36] and Martínez-Gutiérrez et al. [38], during field experiments conducted at the ARDD for the spring–summer agricultural cycle in 2016 and 2017, respectively (

Table 3. Non-conservative parameters used to set up the AquaCrop model for rainfed maize in the ARDD.

Non-Conservative Parameters	Default	Calibrated	Unit
Minimum growing temperature required for full crop transpiration	12	12	°C
Minimum effective rooting depth	0.3	0.3	m
Maximum effective rooting depth	2.3	0.65	m
Number of plants per hectare	75,000	65,000	Plant ha−1
Maximum canopy cover (CCx) (fraction soil cover)	0.96	0.7	%
Calendar days: from sowing to emergence	6	11	Day
Calendar days: from sowing to the maximum rooting depth	108	90	Day
Calendar days: from sowing to senescence	107	119	Day
Calendar days: from sowing to maturity	132	160	Day
Calendar days: from sowing to flowering	66	89	Day
Length of the flowering stage	13	23	Day
Reference harvest index (HIo)	48	27	%
Dry matter content of fresh yield	90	85	%
Minimum effective rooting depth	0.3	0.3	m
Maximum effective rooting depth	2.3	0.65	m

).

In this study, the AquaCrop model was initially run using the default crop parameters for maize. The accuracy of these parameters was assessed by comparing the model simulations with the grain yield observed in the ARDD. Subsequent to sensitivity adjustments in the program, sequenced inter-annual runs were performed for each season of the ARDD to obtain maize yields in each municipality.

To calibrate the crop model for a historical period, the yields observed in each of the municipalities of the ARDD were used, which were reported by the Agrifood and Fisheries Information Service (SIAP), considering only the data from the spring–summer maize cycle for rainfed maize, with an annual record from 2003 to 2020 [11]. Once the model was calibrated and validated, we proceeded to make the runs using climate change projections.

2.6.2. Evaluation of Model Performance

The performance of the AquaCrop simulations in reproducing grain yield was evaluated using indices reported in many AquaCrop applications. This process is an elementary step in verifying the description of real farming systems. It is advisable to employ various metrics to judge the goodness of fit of an agronomic model [49]. These indices consist of the Pearson correlation coefficient (r), root mean square error (RMSE), quadratic root mean normalized error (NRMSE), Nash–Sutcliffe model efficiency coefficient (EF), and Willmott’s index of agreement (d), which have been used to test model performance and compare simulated and observed results using different equations [49,50], as shown in

Table 4. Statistical indicators used to evaluate the simulation of the AquaCrop model and the data observed by SIAP.

No.	Statistical Indicators	Formulas	References
1	Pearson correlation coefficient	$\mathrm{r}={\displaystyle \frac{\sum \left({\mathrm{O}}_{\mathrm{i}}-\overline{{\mathrm{O}}_{\mathrm{i}}}\right)\times \left({\mathrm{S}}_{\mathrm{i}}-\overline{{\mathrm{S}}_{\mathrm{i}}}\right)}{\sqrt{\sum {(\mathrm{O}}_{\mathrm{i}}}-\overline{{\mathrm{O}}_{\mathrm{i}}}{)}^2\times (\mathrm{S}-\overline{{\mathrm{S}}_{\mathrm{i}}}{)}^2}}$	Moksony and Heged [51]
2	Root mean square error (RMSE)	$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{\sum ({\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}{)}^2}{\mathrm{n}}}}$	Jacovides and Kontoyiannis [52]
3	Quadratic root mean normalized error	$\mathrm{N}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}={\displaystyle \frac{1}{\overline{\mathrm{O}}}}\times \sqrt{{\displaystyle \frac{\sum ({\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}{)}^2}{\mathrm{n}}}}\times 100$	Bannayan and Hoogenboom [21]
4	Nash–Sutcliffe model efficiency coefficient	$\mathrm{E}\mathrm{F}=1-{\displaystyle \frac{\sum ({\mathrm{O}}_{\mathrm{i}}-{\mathrm{S}}_{\mathrm{i}}{)}^2 }{\sum ({\mathrm{O}}_{\mathrm{i}}-\overline{{\mathrm{O}}_{\mathrm{i}}}{)}^2}}$	Nash and Sutcliffe [53]
5	Willmott’s index of agreement	$\mathrm{d}=1-{\displaystyle \frac{\sum ({\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}{)}^2 }{\sum (\left|{\mathrm{S}}_{\mathrm{i}}-\overline{{\mathrm{O}}_{\mathrm{i}}}\right|+\left|{\mathrm{O}}_{\mathrm{i}}-\overline{{\mathrm{O}}_{\mathrm{i}}}\right|{)}^2}}$	Willmott [54]

Note(s): O_i, observed; S_i, simulated; n, number of observations; $\overline{\mathrm{O}}$, observed mean; and $\overline{\mathrm{S},}$ simulated mean values.

.

3. Results

3.1. Climate Change Scenario Subsection

Daily temperature and precipitation records from the ARDD for the historical period 1985–2020 were used to validate and calibrate the LARS-WG model. The effectiveness of the model in downscaling GCMs in the study area was analyzed using the Kolmogorov–Smirnov (KS) statistical test, which verified that the distributions of daily climate variables calculated from the observed and simulated data were similar. Additionally, a p-value was employed to either accept or reject the hypothesis that both datasets (observed and synthetic) originated from the same distribution.

The evaluation results shown in

Table 5. K–S test and p-value for the ARDD, seasonal distribution of the wet/dry series.

Season	Period	K–S	p	Assessment
D–J–F	Wet	0.086	1.000	Perfect
	Dry	0.138	0.971	Very good
M–A–M	Wet	0.121	0.993	Very good
	Dry	0.062	1.000	Perfect
J–J–A	Wet	0.135	0.976	Very good
	Dry	0.087	1.000	Perfect
S–O–N	Wet	0.061	1.000	Perfect
	Dry	0.030	1.000	Perfect

indicate that LARS-WG generally performs perfectly in projecting the daily distribution of the precipitation, maximum and minimum temperature variables, with only a few very good results. Therefore, the downscaling model was used more reliably in this research.

During the historical period 1985–2020, the effective precipitation (EP) of the agricultural cycle in the ARDD recorded an average of 650 mm (Figure 3). In the municipality of Temascalcingo, the lowest accumulated rainfall was recorded, with 500 mm; on the other hand, in Atlacomulco, an average EP of 710 mm was recorded. The results of the GCM ACCESS-ESM1-5 indicate that EP will decrease significantly in most municipalities of the ARDD (

Table 6. Percentage change for climate variables EP and T_avg for each municipality of the ARDD.

Station Code	Municipality	1985–2020		SSP2-4.5 (2021–2080)		SSP5-8.5 (2021–2080)
		EP	TAVG	EP	TAVG	EP	TAVG
31	Ixtlahuaca	605	14.9	−15.0	15.9	−17.5	19.2
35	San José del Rincón	680	14.5	2.0	11.8	−1.2	15.1
15076	San Felipe del Progreso	691	14.4	−13.0	11.7	−15.8	15.1
15128	El Oro	790	11.5	−3.4	16.3	−6.3	12.1
15239	Morelos	614	12.8	−4.7	25.9	−6.9	29.7
15251	Atlacomulco	703	14.2	−5.7	14.5	−8.0	17.9
15264	Jiquipilco	562	15.1	−5.8	12.4	−8.1	15.6
36756	Jocotitlán	675	15.0	−8.0	10.9	−10.2	14.1
37673	Acambay	665	11.6	−9.7	17.1	−11.5	21.2
860124	Temascalcingo	497	16.3	−17.4	14.0	−19.3	16.9

), with the exception of San José del Rincón, where a slight increase of up to 2% is identified through SSP2-4.5. The largest decreases in EP are found in Ixtlahuaca and Temascalcingo, with up to −15% for both scenarios. Figure 3 shows that the largest decreases in EP occur towards the distant horizon (2061–2080).

The historical average temperature (T_AVG) of the ARDD during the agricultural cycle oscillates around 14 °C. The municipality of El Oro reports the lowest average temperature, up to 11.4 °C, while in Temascalcingo, the average temperature is 16.3 °C (Figure 3). In this variable, both SSP scenarios project an increase in all municipalities (Table 6), presenting the highest increase in Morelos, with up to 25.9% and 29.7% in SPP2-4.5 and SSP5-8.5, respectively, with respect to the historical average of 12.8 °C. The SSP5-8.5 scenario presents the greatest impacts for this variable.

3.2. Evaluation Metrics

The AquaCrop model was validated using the corn yields observed in the ARDD by SIAP and those simulated by the program. The statistical metrics used for the model evaluation are presented in

Table 7. Statistical indices of the simulated and measured AquaCrop results for the calibration data sets for each municipality of the ARDD.

Station Code	Municipality	RMSE (ton ha−1)	d-Index	NRMSE (%)	EF	r
31	Ixtlahuaca	0.8	0.8	0.2	−0.1	0.7
35	San José del Rincón	1.3	0.7	0.3	0.1	0.7
15076	San Felipe del Progreso	1.0	0.8	0.3	−0.7	0.8
15128	El Oro	1.2	0.1	0.7	−4.1	0.4
15239	Morelos	1.1	0.5	0.4	−0.2	0.7
15251	Atlacomulco	1.2	0.4	0.3	−0.3	0.6
15264	Jiquipilco	0.8	0.7	0.2	0.1	0.6
36756	Jocotitlán	0.9	0.7	0.2	0.3	0.7
37673	Acambay	1.0	0.6	0.3	−1.4	0.6
860124	Temascalcingo	0.7	0.8	0.2	0.3	0.6

. The validation of the model was performed for each municipality based on yield data for the period 2003–2020, obtaining the following ranges for each statistical index: 0.7 ton ha⁻¹ < RMSE > 1.3 ton ha⁻¹; 0.1 < d-index > 0.8; 0.2% < NRMSE > 0.7%; −0.7 < EF > 0.3, and 0.4 < r > 0.8. The simulated and observed statistical performances were similar, indicating that the model had been calibrated satisfactorily.

3.3. Observed vs. Simulated Maize Yields

In the ARDD, the average observed yield for the period 2003–2020 was 3.5 ton ha⁻¹, while the AquaCrop simulation indicates an average yield of 3.8 ton ha⁻¹, an average difference of 300 kg ha⁻¹, which means that the simulated yields are higher than the observed yields in most municipalities (

Table 8. Average values of the measured and simulated results of the AquaCrop model in the growing seasons 2003–2020 in the ARDD, Pe_sim = percentage deviation of the simulated maize yield values from the measured values; Meas = observed yield; Sim = simulated yield.

Station Code	Municipality	Meas (ton ha−1)	Sim (ton ha−1)	Sim–Meas (ton ha−1)	Pesim (%)
31	Ixtlahuaca	3.9	4.1	0.2	5.1
35	San José del Rincón	3.6	4.4	0.8	22.2
15076	San Felipe del Progreso	3.1	3.5	0.4	12.9
15128	El Oro	2.7	3.5	0.8	29.6
15239	Morelos	3.1	3.6	0.5	16.1
15251	Atlacomulco	3.8	3.5	–0.3	–7.9
15264	Jiquipilco	3.7	3.9	0.2	5.4
36756	Jocotitlán	3.9	4.4	0.5	12.8
37673	Acambay	3.4	3.9	0.5	14.7
860124	Temascalcingo	3.9	4	0.1	2.6

). The prediction error for the simulated yields was less than 10% for 5 out of 10 records. It was observed that the minimum error (Pe_sim = 2.6%) in grain yield prediction was obtained in Temascalcingo, while the maximum error was obtained for El Oro (Pe_sim = 29.6%).

Scatter plots of the estimated and measured maize yields for the period 2003–2020 are shown in Figure 4. The regression line of the simulated yield and the observed yield aligned closely with line 1:1 (Figure 4), demonstrating acceptable consistency between the simulated and actual observations. It can be seen that the regression line is above the identity line in most municipalities, with the exception of Atlacomulco, verifying that the model overestimates the observed data. This discrepancy is due to the climatic variability presented annually for each agricultural cycle, mainly in precipitation, which leads to a period of water stress during a critical stage in the maize crop, affecting maize yields and resulting in a higher prediction error.

3.4. Maize Yields Under Climate Change Scenarios

The effect of climate change on rainfed maize production in the ARDD is projected to persist over the period 2021–2080, with an increase in crop yields (Figure 5) from 3.5 ton ha⁻¹ to 4.2 ton ha⁻¹ and 4.4 ton ha⁻¹ with the SSP2-4.5 and SSP5-8.5 scenarios. In SSP2-4.5, the municipality that projects the highest maize production is Jocotitlán, with up to 5.1 ton ha⁻¹, followed by San José del Rincón with 5.0 ton ha⁻¹, while El Oro projects the lowest yield with up to 2.5 ton ha⁻¹. On the other hand, with the SSP5-8.5 scenario, it is possible to identify San José del Rincón as having the highest yields, with up to 5.2 ton ha⁻¹, followed by the municipality of Atlacomulco with 4.9 ton-ha⁻¹, while El Oro has the lowest yields with 2.8 ton ha⁻¹.

Considering the variability of the projected yields for each municipality, a difference is observed between the SSP2-4.5 scenario and the SSP5-8.5 scenario, the latter being where the greatest impacts are perceived (Figure 5). In the municipalities of Ixtlahuaca, Jiquipilco, San Felipe del Progreso, Jocotitlan, Morelos, and Temascalcingo, higher yields are expected towards the medium horizon (2041–2060), while for Atlacomulco, San José del Rincón, El Oro and Acambay, the highest maize yields are projected towards the distant horizon (2061–2080). The lowest yields for the municipalities of Morelos, Acambay, El Oro, and Atlacomulco are identified in the near horizon (2021–2040) for both SSPs.

In Figure 6, the impact on maize yields can be identified through the different climate change scenarios and in the different time horizons. These projections indicate an increase in maize yields in most municipalities; however, decreases can be observed in the municipalities of El Oro, Acambay, and Temascalcingo, which are mostly noticeable with SSP2-4.5 during the 2021–2040 and 2041–2060 horizons. Figure 6 presents the different trends and behaviors over time of the maize yield simulations with AquaCrop for each municipality in the ARDD from 2021–2080.

4. Discussion

In the ARDD, the GCM ACCESS-ESM1-5 for the period 2021–2080 projects a decrease in effective precipitation, which is quantified at 7.5% and 11% in the SSP2-4.5 and SSP5-8.5 scenarios, respectively, compared to the historical average (650 mm·yr⁻¹). The decrease in precipitation identified in this study is similar to that reported in the State of Mexico by Estrada et al. [55], who mentioned that by the end of the 21st century, a reduction of less than 10% is possible with SSP5-8.5. Manzanilla-Quiñonez et al. [56] reported a decrease of less than 10% with respect to the historical average (1090 mm·yr⁻¹) using RCP8.5 for the period 2045–2069 in Nevado de Toluca, State of Mexico. With respect to the average temperature, an increase of 1.3 °C is projected with SSP2-4.5 and up to 1.7 °C with SSP5-8.5 with respect to the historical average temperature during the agricultural cycle (14 °C). These results are consistent with the continued increase in warming observed for the end of the 21st century by IPCC 6, which indicates a global increase in the mean temperature from 1.0 °C under SSP2-2.6 to 5.4 °C with the SSP5-8.5 scenario [1]. The results of this study are similar to those reported by Manzanilla-Quiñonez et al. [56], with a mean temperature increase of 1.87–2.03 °C under RCP8.5 during the period 2045–2069 in the Toluca Valley.

Regarding the metrics evaluated for the rainfed maize modeling in the ARDD (Table 7), in the first instance, the RMSE presented values ranging from 0.7 to 1.3 ton ha⁻¹. These values represent a low discrepancy between the observed and simulated data in the ARDD [52]; this is due to the fact that AquaCrop tends to present a less satisfactory RMSE as a result of extreme deficits or rainfed conditions, as in the present study [57]. The values of the d-index and the Pearson correlation coefficient (r) presented in this study were consistently close to 0.8, indicating a strong agreement and a linear relationship between the observed and simulated data [50,53]. The NRMSE describes an excellent simulation for all the ARDD municipalities, being less than 10% [58], with a better evaluation in the municipalities of Ixtlahuaca, Jiquipilco, Jocotitlán, and Temascalcingo, while the EF revealed values ranging from −4 to 0.3, indicating a better efficiency of the model simulation in the municipalities of Jocotitlán and Temascalcingo. According to Moriasi et al. [59], values between 0 and 1.0 indicate acceptable performance. In this case, only 4 out of the 10 records were in this range. Eshete et al. [18] determined that the EF decreases when the model identifies a water stress situation in the crop or drought conditions.

The AquaCrop model simulated maize yields acceptably in the ARDD (0.7 ton ha⁻¹ < RMSE > 1.3 ton ha⁻¹; 0.1 < d-index > 0.8; 0.2% < NRMSE > 0.7%; −0.7 < EF > 0.3, and 0.4 < r > 0.8); thus, it was possible to determine that the simulated values were satisfactorily adjusted to the yields reported by SIAP in 2003–2020. These indices were similar to those reported by Eshete et al. [18] (RMSE = 4.1%, d-index = 0.85, and EF = 0.35), obtained for maize with 25% irrigation application in Ethiopia. Mebane et al. [60] obtained the following values in rainfed maize in Pennsylvania, USA, with a cumulative annual rainfall of 761 mm: RMSE = 0.96 ton ha⁻¹, d-index = 0.9, NRMSE = 0.3%, and r = 0.9, while da Conceição et al. [56] evaluated the yield of a maize hybrid subjected to water stress in Santiago, Chile, obtaining an RMSE = 0.62 ton ha⁻¹, NRMSE = 4.41%, EF = 0.46, d-index = 0.82, and r = 0.73. Thamer et al. [61] found an RMSE = 0.84 ton ha⁻¹, d-index = 0.97, and EF = 0.92 in an arid region located in an agricultural district of Baghdad, Iraq.

In the case of the statistics where the model presented a low prediction efficiency, it was attributed to the characteristics of the observed data. Annually, in the ARDD, the yields of rainfed maize are different, with much variability, mainly because the crop depends entirely on the water input from rainfall, which in the study area occurs with erratic characteristics. This condition means that maize yields do not present a continuous pattern as the AquaCrop model tends to generate.

The comparison of the observed and simulated grain yield in the ARDD study area showed a prediction error in the range of 2.6–28.6% over the period 2003–2020. Thamer et al. [61] reported similar prediction error statistics for grain yield (2.1–5.5%) for an arid region of Iraq, Flores-Gallardo et al. [33] (0.8–16%)—in an arid area of northern Sinaloa, Mexico, Mibulo & Kiggundu [29]—under rainfed conditions in Wakiso, Uganda (2.0–3.1%), Abbas et al. [5] (1.7–6.0%)—in a semi-arid environment of Pakistan, Sandhu and Irmak [20] (2019) (1.5–17.8%)—for Nebraska, USA, and Eshete et al. [18] (3.6–8.7%)—in an irrigated deficit field in central Gondar, Ethiopia. In this regard, Adeboye et al. [17] and Zhu et al. [62] reported similar observations when comparing the observed vs. simulated maize yields; however, they pointed out that as long as the scatter points are concentrated close to the 1:1 line, it is an indicator that the model has been calibrated satisfactorily.

The impacts of climate change on maize production are documented, and hundreds of experiments have been conducted around the world, showing that the crop can show either a decrease or an increase in grain yield depending on the geographical region and climatic conditions, but most point to a negative impact [6,63]. An example of it is the finding of Kothari et al. [64] in North Texas, who reported a decrease of 8–30% by the 2050s and 40–70% by the 2080s under the RCP8.5 scenario. Araya et al. [65] projected a decrease of 18% in Garden City, Kansas, using RCP4.5 and 30% using RCP8.5 for the 2040–2069 horizon. Similarly, Lin et al. [66] in northern China showed that the average yields would decrease by 2.1% in the 2020s, by 12.9% in the 2050s, and by 22.7% in the 2080s with RCP4.5.

However, simulation results of rainfed maize yield in the ARDD for the period 2021–2080 indicate a 21% increase under SSP2-4.5 and a 24% increase with SSP5-8.5. In this regard, Umesh et al. [67] found an increase in maize yield of 28–74% under SSP2-4.5 and 34–85% under SSP5-8.5 considering the period 2025–2090 in the south of India. Abera et al. [68] reported that maize yield will increase in the highlands of Hawassa, Ethiopia; under RCP8.5, it is projected to increase by up to 50% in the short term (2010–2039), by 42% by mid-century (2040–2069), and by 35% by the end of the century (2070–2099). Johnston et al. [69] reported that simulated rainfed yields increased on average by 4.6% with the A1b scenario, by 2.4% with the A2 scenario, and by 4.0% with the B1 scenario compared to the historical average in 52 counties in South Dakota and Nebraska, USA.

For municipalities with high fluctuations and the largest yield reductions in the ARDD, such as Ixtlahuaca, San Felipe del Progreso, and Temascalcingo, these were due to annual variability and the significant decrease in projected precipitation towards the distant horizon. In contrast, in the municipalities where AquaCrop projects higher future yields, such as San José del Rincón, El Oro, and Acambay (as shown in Figure 6), the higher average temperature rise into the future is the cause.

The increase in rainfed maize yields projected in the ARDD is mainly attributed to the increase in average temperature, which in turn promotes an increase in GDD during the agricultural cycle. In this regard, Cruz-González et al. [70] identified in the ARDD study area that towards the distant horizon (2061–2080), there will be an increase in the GDD during the phenological stages of maize: germination and emergence (19% and 33% under SSP2-4.5 and SSP5-8.5, respectively), vegetative development (20% and 32%), flowering (20% and 32%), and grain filling and maturity (21% and 33%). These conditions allow for an increase in the speed of occurrence of phenological changes between stages in the maize crop in a temperate climate.

This analysis, using the entire historical temperature distribution of the agricultural cycle in the ARDD, showed temperatures ranging from 7.5 °C to 22.3 °C, while the climate change scenarios for the future period (2021–2080) indicate an increase to 9–23.2 °C and 9.5–23.6 °C with SSP2-4.5 and SSP5-8.5, respectively. The minimum temperatures reported in the ARDD are below the base temperature (10 °C), while the maximum temperature is also below the optimum crop temperature (30 °C), causing the maize development rate to be slow [71,72]. However, these thermal thresholds vary according to genotype and region, with values of 7–27 °C being the most appropriate for maize cultivars adapted to the high valleys of Mexico [73]. For maize yield, according to the literature, the greatest impact will be caused by the increase in temperature rather than by precipitation.

Although radiation and temperature define the potential productivity of the maize crop, water supply imposes an upper limit for the grain yield of the rainfed crop, considering the flowering stage (R1) as the critical period with the highest water requirement, due to the formation of the number of grains per ear, followed by grain filling and maturation [74]. Cruz-González et al. [35] determined agricultural drought stages in the phenological phases of maize in the ARDD during the period 1985–2017. Using the SPEI index, they quantified that the grain filling and maturity phase was the most susceptible, with 31% of the total dry periods. On the other hand, for the period 2041–2080, it is anticipated under SSP2-4.5 that vegetative development will be the most affected, while SSP5-8.5 suggests that the phenological phase most impacted by droughts will be grain filling and maturity, with an incidence of 26% of the total dry months.

In a rainfed farming system, the maize crop is developed under water stress conditions, and the availability of moisture is rarely sufficient for this crop, limiting its development and metabolism [30]. Moreover, all nutrients are absorbed by the plant root after their ionic form is dissolved in the soil solution; therefore, nutrient and water scarcity are the main limiting factors for the low productivity of rainfed maize [9].

From the results of this study, it was identified that effective rainfall tends to decrease towards the distant period by 7.5% and 11% under SSP2-4.5 and SSP5-8.5, respectively. Despite such a decrease in the different scenarios, the AquaCrop model identified that it would not be a limiting factor for grain production, since the effective precipitation presented in the future meets the basic water needs during the phenological stages of maize. The FAO [75] states that maize grown for dry grain production consumes 400–700 mm of water on average throughout the growth cycle, depending on climatic conditions. The projected EP in the ARDD in 2021–2080 averages 600 and 580 mm·yr⁻¹ under SSP2-4.5 and SSP5-8.5, respectively.

Although it was not an object of study in this research, it is necessary to mention some forms of adaptation to and mitigation of climate change, as it has been identified that this phenomenon will be beneficial for rainfed maize production in the future in the ARDD. However, actions could be implemented to boost yields and ensure economic sustainability and food security in central Mexico. Such strategies include (1) implementing hybrid varieties or maize breeds with higher tolerance to high temperatures, (2) increasing the rate (%) of organic residues incorporated into the soil to improve the retention of available moisture and soil structure, (3) opting for agroecological management practices with a lower presence of agrochemicals and agricultural technology, and (4) adjusting the dates of crop management practices, such as sowing.

5. Conclusions

The results showed that rainfed maize crop yields will increase with future climate change in the ARDD from 3.5 ton ha⁻¹ to 4.2 ton ha⁻¹ and 4.4 ton ha⁻¹ with the SSP2-4.5 and SSP5-8.5 scenarios. This is due to environmental factors such as the increase in the average temperature, which favors the thermal accumulation of GDD. This situation leads to an environment with greater climatic aptitudes for the establishment and development of the crop with respect to the historical climate, which presents cold conditions. On the other hand, it was identified that the effective precipitation will decrease in the future; however, even under this scenario, the water needs of maize can be satisfied.

The AquaCrop model was able to adequately simulate rainfed maize yields in the ARDD. According to the statistics RMSE, d-index, NRMSE, EF, and r, it was possible to verify that the simulated values were satisfactorily adjusted to the yields reported by SIAP from 2003–2020. The evaluation of the model allowed for reliability in projecting yields for each municipality of the study area under climate change conditions into the future.

The results analyzed in this research suggest that the FAO AquaCrop model can be used to predict rainfed agricultural production and therefore has a greater potential to guide management practices aimed at increasing food production in Central Mexico. However, the model needs to be tested with irrigation and fertilization management practices to explore its performance under these conditions.

Regarding the new opportunities and challenges in this area, it is important to mention that the perception of climate and its evolution over time is a field of study that is still developing, especially when considering the behavior of crops under changing conditions. For this reason, it is crucial to carry out initiatives to create an experimental database on crops at the national level, since, in Mexico, the main limitation is the lack of specific data by crop type. This makes it difficult to make projections such as those shown in this analysis, which are fundamental for management and decision-making in agriculture.