A Methodological Framework for Assessing the Potential Performance of Maize/Soybean Intercropping Under 2050 Climate Scenarios

Abstract

Climate change and increasing uncertainty in global food trade pose major challenges to China’s food security. This study aimed to evaluate the potential of maize/soybean intercropping across China under current and projected 2050 climate scenarios. A database of 637 field experiments from 56 publications, combined with climate and soil datasets, was trained Random Forest machine learning model to predict yield potential and assess the impacts of replacing varying proportions of maize cropland with maize/soybean intercropping under 2050 climate change. Our results indicate that suitable intercropping zones will expand northwards, with the North China Plain and Northeast China showing particular promise, while land equivalent ratios are expected to remain stable or increase, especially in Southwest China. Scenario analyses suggest that converting 30–50% of maize land to intercropping could double soybean production while maintaining maize yields, and full conversion under low-demand scenarios could achieve near self-sufficiency in maize (96.8%) and exceed self-sufficiency in soybean (104.7%). These results demonstrate the strategic potential of maize/soybean intercropping to enhance national food security. Furthermore, this study provides an exploratory and novel methodological framework that integrates multi-source datasets and machine learning to assess the potential performance of maize/soybean intercropping under climate change, offering a useful reference for future research and model refinement.

1. Introduction

With the increasing frequency of climate change and extreme weather events, agricultural production faces escalating risks, particularly from altered precipitation patterns, rising temperatures, and more frequent droughts [1,2]. In China, global political instability and uncertainties in agricultural trade further underscore the strategic necessity of enhancing self-sufficiency in staple crops [3]. Maize and soybean are of particular importance, serving not only as key sources of food and feed but also as essential inputs for industry and biofuel production [4]. However, both crops are predominantly grown under monocropping systems, and the ecological uniformity of such systems renders them highly vulnerable to pest invasions, disease outbreaks, and the adverse effects of climate change [5,6,7,8]. To mitigate the vulnerabilities associated with monocropping and enhance agricultural resilience, diversified cropping systems—particularly intercropping—have been increasingly recognised as sustainable alternatives that improve both productivity and ecosystem stability.

Intercropping is the planned cultivation of multiple crop species on a plot of land to benefit from species complementarities and enhance land use efficiency [9,10]. Intercropping draws interest in research on ecological intensification of agriculture, due to its benefits in terms of land productivity [11,12,13,14], pest and disease control [15,16,17], weed control [18], efficient nutrient use [19,20], and soil quality [21,22]. The Chinese government has implemented a series of policies to promote diversified cropping, including intercropping and mixed cropping. Notably, the maize/soybean strip intercropping promotion policy was officially launched in 2022 with the aim of increasing soybean self-sufficiency [23,24]. However, the yield performance of maize/soybean intercropping is highly dependent on natural conditions, particularly climate and soil characteristics, leading to considerable regional variability [25]. It is not clear what the coupling relationship is between the yield advantage of intercropping and the natural conditions across China.

Given the considerable regional variability in maize/soybean intercropping yields due to climatic and soil factors, traditional field experiments and empirical studies alone are insufficient to fully capture the potential and limitations of intercropping systems across China. Consequently, data-driven approaches, such as machine learning, have emerged as promising tools for predicting crop yields and assessing the performance of complex cropping systems [26,27]. Climate change is a major driver of spatial and temporal yield variability; studies in the United States indicate that redistributing 57% of monoculture cropland could reduce yield losses by 16% [28], while strategic spatial reallocation of China’s cropping systems could contribute 23–40% toward achieving the nation’s 2030 Sustainable Development Goals [29]. Chen (2025) compiled literature on maize/soybean experimental sites and applied Random Forest models to evaluate potential yields under current cropping conditions [30]. However, this study did not account for the effects of climate change, and its recommendations for future cropping system adaptation were therefore restricted. Consequently, there is a clear need for research that integrates projected climate scenarios to assess the future performance and optimization of maize/soybean intercropping systems.

Despite the increasing interest in maize/soybean intercropping, significant knowledge gaps remain regarding how climate change will influence its spatial suitability and yield potential across China. Most existing studies have relied on local field experiments or historical datasets, lacking an integrated, data-driven framework that combines large-scale experimental data with future climate projections. To address this gap, the present study develops and tests an exploratory methodological framework that applies machine learning to evaluate the potential performance of maize/soybean intercropping under projected 2050 climate scenarios. Specifically, we address two key research questions: (1) Which regions of China are likely to become suitable for maize/soybean intercropping under projected 2050 climate conditions? and (2) To what extent could maize/soybean intercropping enhance maize and soybean yields under current and future climates? By addressing these questions, this study aims to enhance understanding of the climate suitability and yield potential of maize/soybean intercropping, and generate valuable insights and guide future model refinement, supporting the development of more resilient and sustainable agricultural systems across China. The proposed framework offers an exploratory yet scalable methodological framework for integrating multi-source data to assess intercropping potential under climate change.

2. Materials and Methods

2.1. Data Collection

Data for this study were obtained from the Web of Science Core Collection and the China National Knowledge Infrastructure (CNKI) Core Database, covering publications up to April 2022. In the Web of Science and CNKI, a topic-based search was conducted using the query (“intercrop*” OR “mixture” OR “mixed crop” OR “mixed*” OR “polycult*”) AND CU = China. The initial search yielded 4281 publications. After excluding non-Chinese and non-English articles, as well as review papers, 4046 publications remained for further screening. Titles and abstracts were then examined to exclude studies on agroforestry systems, non-field experiments, fish polyculture, rotation trials, and non-China field experiments, resulting in 1022 publications. Full-text review was subsequently conducted to ensure that only field experiments were included, each experiment comprised both maize/soybean intercropping and monocropping with identical management practices (e.g., NPK fertilisation, pesticide application, and crop varieties), and yield data were reported in the text, tables, figures, or supplementary materials. Following this rigorous screening process (see PRISMA in Supplementary Materials), in total, 637 data points were obtained from 56 publications (Supplementary Materials), with each data point including yields of maize and soybean under both intercropped and monocropped conditions. Data from figures and charts were extracted using Get Data Graph Digitizer 2.25: http://getdata-graph-digitizer.com/ (accessed on 12 April 2023). The spatial distribution of the data is shown in Figure 1.

2.2. Meteorological and Soil Data

Meteorological data were obtained from the international database WorldClim [31], which provides both annual and monthly climate indicators. Monthly indicators include maximum temperature, minimum temperature, mean temperature, and precipitation, while annual indicators include accumulated temperature above 10 °C and annual precipitation. Soil data were sourced from the HWSD 2.0 database [32] and comprised a range of nutrient and texture parameters, including soil organic matter content, pH, total nitrogen content, soil texture, percentage of sand, and percentage of loam.

2.3. Machine Learning Model Selection and Cross-Validation

The machine learning models were trained on a comprehensive dataset comprising location information (latitude and longitude), monthly climate variables (maximum and minimum temperatures, monthly rainfall), annual climate indicators (total rainfall and accumulated temperature above 10 °C), and soil properties (soil organic carbon, pH, total nitrogen, and clay content). Maize and soybean yields under both monocropping and intercropping systems served as the predicted value. These predictors enabled the models to capture the effects of environmental and soil factors on crop performance. Based on these predictions, we calculated the potential LER to identify intercropping advantage zones (i.e., higher LER values indicate greater intercropping advantages). Furthermore, future $pLER$ were applied to estimate yield changes in different scenario analyses.

The LER is defined as the sum of the relative yields of intercropped species compared with their respective sole crops (Equation (1)) [11].

$$LER={\displaystyle \frac{Y_1}{M_1}}+{\displaystyle \frac{Y_1}{M_2}}={pLER}_1+{pLER}_2$$

(1)

where $Y_1$ and $Y_2$ are the yields (per unit of total area of the intercrop) of species 1 and 2 in an intercrop, $M_1$ and $M_2$ are the yields of species 1 and 2 in the sole crops, and ${pLER}_1$ and ${pLER}_2$are the partial land equivalent ratios (relative yields) of species 1 and 2, respectively. $pLER$ are calculated as the ratio of the yields of a species in the intercrop and the sole crop. The LER indicates the relative land area required under sole crops to obtain the same yield of the component species as a unit area of the intercrop would under the same or comparable management. A LER greater than one indicates that a larger area is needed to produce the output quantities of species 1 and 2 with sole crops than with an intercrop. The $pLER$ represents the relative areas of the component crop species required to produce the yield obtained in a unit area of the intercrop.

Several machine learning algorithms were explored and trained, including Random Forest (RF), Generalised Linear Model Net (Glmnet), Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), and LASSO regression. Initial screening indicated that Random Forest and LASSO regression achieved higher prediction accuracy. Subsequently, cross-validation was performed, comparing model performance on the validation set using metrics such as R², mean squared error (RMSEP) (Supplementary Materials).

Based on model evaluation metrics, the Random Forest algorithm is more suitable for predicting the target variable than LASSO, as it shows higher explanatory power (R² = 0.594) and lower prediction error compared to LASSO (R² = 0.368, RMSEP = 0.218). The Root Mean Square Error of Prediction (RMSEP) is a widely used metric for evaluating the predictive performance of a model. It is calculated as the square root of the mean squared differences between observed and predicted values:

$$RMSEP=\sqrt{{\displaystyle \frac{1}{n}}{\sum }_{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2}$$

(2)

where $y_i$represents the observed values, ${\widehat{y}}_i$the predicted values, and $n$the number of observations. Unlike the Mean Squared Error (MSE), which is expressed in squared units, RMSEP is in the same unit as the original variable, making it more interpretable. Lower RMSEP values indicate better predictive accuracy, and they are commonly used in cross-validation or external validation to assess model reliability.

Table 1. Algorithms Settings and Performance.

Algorithms	Settings	RMSE	R2
RF	ntree = 500, mtry = 9	0.18	0.59
LASSO	alpha = 1 (LASSO), lambda = 0.000919	0.22	0.36
GLMNET	alpha = 0.5 (Elastic Net), lambda = 0.00176	0.22	0.38
MLR	Standard linear model (lm)	0.43	0.10
ANN	size = 5, decay = 0.01, maxit = 1000	0.26	0.36

summarizes the parameter settings and predictive performance (RMSE and R²) of all algorithms tested, showing that Random Forest achieved the highest predictive accuracy and was therefore selected as the optimal algorithm for this study. Cross-validation procedures based on the Random Forest model are illustrated in Figure 2.

2.4. Scenario Analysis

The scenario considered in this study involves replacing varying proportions of maize cropland with maize–soybean intercropping. We assume that farmers’ management practices and intercropping patterns remain unchanged, and evaluate the yield performance of maize/soybean intercropping under projected 2050 climate conditions.

For the climate change scenario, four emission pathways for 2050 from the BCC-CSM2-MR climate system model (CMIP6) were used [33]. The climate projections, available at a 2.5 min spatial resolution, included monthly variables (maximum temperature, minimum temperature, mean temperature, and precipitation) and annual indicators (accumulated temperature above 10 °C and total annual precipitation). These climate variables were integrated with soil attributes and directly input into the trained machine learning models to predict maize and soybean yield responses under future climate conditions. Based on the predicted yields, the corresponding partial Land Equivalent Ratios (pLER) and overall LER values were subsequently calculated for scenario analysis.

To more clearly assess the potential contribution of maize/soybean intercropping to China’s maize and soybean self-sufficiency, the future scenarios were analysed. In practice, replacing maize monocropping with maize/soybean intercropping is feasible because the maize growing season is longer than that of soybean. In contrast, replacing soybean monocropping with intercropping may impact the planting schedule of subsequent crops. Accordingly, three scenarios were established for comparative analysis:

• Scenario 1: 100% of national maize cropland is replaced with maize/soybean intercropping, while soybean cropland remains unchanged.
• Scenario 2: 50% of national maize cropland is replaced with maize/soybean intercropping, with the remaining 50% retained as maize monoculture, and soybean cropland unchanged.
• Scenario 3: 30% of national maize cropland is replaced with maize/soybean intercropping, with the remaining 70% retained as maize monoculture, and soybean cropland unchanged.

Maize yield = Y_maize × A × (pLER_maize × p_i) + Y_maize × A × (1 − p_i)

(3)

Soybean yield = Y_soybean × A × (pLER_soybean × p_i) + TY_soybean

(4)

In above equation, Y represents the current yield of maize or soybean per hectare (t/ha) from statistical yearbook [34], while A represents the total area allocated to maize cultivation (ha) in China. The pLER represents the partial land equivalent ratio, which indicates the proportion of monocrop yield achieved under intercropping. The pLER is calculated based on predicted future production using a machine learning model. For example, a pLER value of 0.75 for maize means that intercropped maize yields 75% of the monocropped maize yield. The p_i represents the scenario parameter, indicating the proportion of maize land replaced by maize/soybean intercropping, with values set at 1.0, 0.5, and 0.3 in this study. TY_soybean represents the total national soybean production in the current statistical yearbook. For areas not replaced, the per-unit-area yield was assumed constant. Data on provincial maize planting area, per-unit-area yield, total production, soybean planting area, per-unit-area yield, and total production were obtained from the China Statistical Yearbook [34]. Maize cropland that does not meet the minimum annual accumulated temperature requirement will continue to be cultivated with sole maize. The threshold for maize/soybean intercropping is an annual accumulated temperature above 10 °C of 1868 °C, derived from the minimum accumulated temperature observed at maize/soybean intercropping distribution sites in the database.

The apparent consumption of maize and soybean in 2022 was 296 million tonnes and 117 million tonnes, respectively, calculated as “production + imports − exports + (beginning stocks − ending stocks)”. According to the United Nations World Population Prospects 2022, China’s population was approximately 1.43 billion in 2022 and is projected to decline to 1.32 billion by 2050 [35]. Assuming no changes in population dietary patterns, the baseline demand in 2050 is estimated at 108 million tonnes for soybeans and 273 million tonnes for maize. Considering potential ±20% fluctuations in dietary demand, the future apparent consumption ranges were estimated as follows: Maize: low-demand scenario 219 million tonnes, baseline scenario 273 million tonnes, high-demand scenario 328 million tonnes; Soybean: low-demand scenario 86 million tonnes, baseline scenario 108 million tonnes, high-demand scenario 130 million tonnes. These estimates represent potential food consumption based on assumptions of population size and dietary patterns, reflecting demand uncertainty under different socio-economic and consumption scenarios.

2.5. Statistical Analysis

All statistical analyses and visualization were conducted in RStudio version 2025.05 based on R version 4.3.0 [36]. The Random Forest algorithm was implemented using the “randomForest” package [37].

3. Results

3.1. Impacts of 2050 Climate Change on Maize/Soybean LER

Future climate warming is projected to drive pronounced northward shifts in the spatial distribution of maize/soybean intercropping suitability. Relative to the current distribution (Figure 3a), the suitable area is expected to expand markedly, with the magnitude of expansion increasing under higher greenhouse gas emissions and greater warming. The most substantial northward expansion is projected in northern China, particularly across Inner Mongolia Autonomous Region, Heilongjiang province, and Xinjiang Autonomous Region (Figure 3b–e).

Under the backdrop of climate warming by 2050, the land equivalent ratio (LER) of maize/soybean intercropping is projected to remain broadly similar to, or slightly higher than, current levels, albeit with pronounced regional variation (Figure 3). The highest LER values, exceeding 1.3 on average, are projected in the Southwest and the North China Plain, while high-yield areas in Northeast China are concentrated primarily in southern Liaoning. Across multiple 2050 climate scenarios (Figure 3b–e), the Southwest, the North China Plain, and southern Liaoning province consistently demonstrated higher land equivalent ratio (LER) values relative to other regions in China. These results indicate that under future climate conditions, the high-latitude regions of northern China could emerge as suitable areas for maize/soybean intercropping, while Southwest China and the North China Plain are projected to have the highest LER values compared with other regions.

3.2. Scenario Analysis of National Maize and Soybean Self-Sufficiency

Our results show that expanding maize/soybean intercropping markedly increases soybean production while causing only slight changes in maize yields across SSP climate pathways (Figure 4). Full conversion of maize cropland to intercropping (Scenario 1) more than quadruples soybean output by 2050 (from ~20 Mt to 89–101 Mt), although maize yields decline to ~210–245 Mt depending on the climate scenario. Partial adoption (Scenarios 2 and 3) achieves a balance, with soybean yields doubling relative to current levels (40–60 Mt) while maize production shows a slight decrease compared with current levels (244–268 Mt vs. 277.2 Mt).

Projected 2050 maize/soybean intercropping scenarios show a trade-off, with partial conversion markedly boosting soybean yields while only slightly reducing maize production. Currently, maize monoculture yields 277 million tonnes, while soybean monoculture yields 20 million tonnes. Under a projected SPP1-2.6 by 2050, if 100% of maize cropland is converted to maize/soybean intercropping, maize yields are projected to decline to 210 million tonnes, while soybean yields increase to 89 million tonnes. With 50% of maize cropland replaced, maize yields are expected to reach 244 million tonnes and soybean yields 54 million tonnes, whereas a 30% replacement scenario results in maize yields of 257 million tonnes and soybean yields of 40 million tonnes. Under a more severe warming scenario of SSP 5-8.5, full maize cropland replacement reduces maize yields to 212 million tonnes, with soybean yields rising to 90 million tonnes. For 50% replacement, maize and soybean yields are projected at 245 million tonnes and 55 million tonnes, respectively, and under 30% replacement, maize yields reach 258 million tonnes while soybean production rises to 41 million tonnes, effectively doubling current soybean yields.

In particular, converting 100% of maize cropland could nearly achieve maize self-sufficiency reaches 96.8%, while soybean self-sufficiency exceeds demand at 104.7%; while converting 50% of maize cropland could result in maize self-sufficiency of approximately 89.3% of baseline demand and soybean self-sufficiency of about 50%. These projections indicate that even 30% adoption of intercropping could double national soybean production while slightly reducing maize yields.

Based on

Table 2. Current and Projected Maize Yield under 2050 SSP1-2.6 and SSP2-4.5 Scenarios (Unit: Million Tons).

		2050 SSP1-2.6 Scenario			2050 SSP2-4.5 Scenario
	Current	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3
China	277.2	210.1	243.6	257.0	211.9	244.6	257.6
Anhui	6.6	4.4	5.5	6.0	4.4	5.5	6.0
Beijing	0.3	0.2	0.3	0.3	0.2	0.3	0.3
Fujian	0.2	0.1	0.1	0.1	0.1	0.1	0.1
Gansu	6.6	4.9	5.8	6.1	4.8	5.7	6.1
Guangdong	0.6	0.4	0.5	0.6	0.4	0.5	0.5
Guangxi	2.8	1.7	2.3	2.5	1.7	2.3	2.5
Guizhou	3.0	2.1	2.6	2.7	2.1	2.6	2.7
Hainan	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Hebei	20.9	15.3	18.1	19.3	15.3	18.1	19.2
Henan	22.8	16.7	19.7	20.9	17.0	19.9	21.0
Heilongjiang	40.4	32.9	36.6	38.1	33.3	36.9	38.3
Hubei	3.1	2.0	2.5	2.8	2.0	2.6	2.8
Hunan	2.3	1.4	1.8	2.0	1.4	1.8	2.0
Jilin	32.6	24.9	28.8	30.3	25.9	29.3	30.6
Jiangsu	3.0	2.0	2.5	2.7	2.1	2.5	2.7
Jiangxi	0.2	0.1	0.2	0.2	0.1	0.2	0.2
Liaoning	19.6	15.5	17.5	18.4	15.3	17.5	18.3
Inner Mongolia	31.0	23.7	27.4	28.8	23.9	27.5	28.9
Ningxia	2.8	1.9	2.3	2.5	1.9	2.3	2.5
Qinghai	0.1	0.1	0.1	0.1	0.1	0.1	0.1
Shandong	26.3	18.6	22.5	24.0	18.8	22.6	24.1
Shanxi	10.2	7.8	9.0	9.5	7.7	9.0	9.5
Shaanxi	6.2	4.6	5.4	5.7	4.6	5.4	5.7
Shanghai	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Sichuan	10.5	8.4	9.5	9.9	8.4	9.4	9.8
Tianjin	1.2	0.9	1.1	1.1	0.9	1.1	1.1
Xizang	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Xinjiang	10.8	9.6	10.2	10.4	9.6	10.2	10.5
Yunnan	10.3	7.7	9.0	9.5	7.7	9.0	9.5
Zhejiang	0.2	0.2	0.2	0.2	0.2	0.2	0.2
Chongqing	2.6	1.8	2.2	2.3	1.8	2.2	2.3

Note: Current status is based on the 2022 National Statistical Yearbook. Scenario 1: 100% of maize cropland converted to maize/soybean intercropping; soybean cropland unchanged. Scenario 2: 50% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged. Scenario 3: 30% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged.

,

Table 3. Current and Projected Maize Yield under 2050 SSP3-7.0 and SSP5-8.5 Scenarios (Unit: Million Tons).

		2050 SSP3-7.0 Scenario			2050 SSP5-8.5 Scenario
	Current	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3
China	277.2	245.3	261.2	267.6	247.1	262.1	268.1
Anhui	6.6	4.8	5.7	6.1	5.1	5.9	6.2
Beijing	0.3	0.3	0.3	0.3	0.3	0.3	0.3
Fujian	0.2	0.1	0.1	0.1	0.1	0.1	0.1
Gansu	6.6	5.6	6.1	6.3	5.6	6.1	6.3
Guangdong	0.6	0.4	0.5	0.6	0.4	0.5	0.6
Guangxi	2.8	1.7	2.3	2.5	1.7	2.3	2.5
Guizhou	3.0	2.4	2.7	2.8	2.4	2.7	2.8
Hainan	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Hebei	20.9	18.3	19.6	20.2	18.5	19.7	20.2
Henan	22.8	19.4	21.1	21.7	20.3	21.5	22.0
Heilongjiang	40.4	38.1	39.2	39.7	38.5	39.5	39.8
Hubei	3.1	2.2	2.7	2.8	2.3	2.7	2.9
Hunan	2.3	1.4	1.8	2.0	1.4	1.8	2.0
Jilin	32.6	29.8	31.2	31.7	29.9	31.2	31.8
Jiangsu	3.0	2.4	2.7	2.8	2.5	2.7	2.8
Jiangxi	0.2	0.1	0.2	0.2	0.1	0.2	0.2
Liaoning	19.6	18.7	19.1	19.3	17.7	18.6	19.0
Inner Mongolia	31.0	27.3	29.1	29.9	27.5	29.2	29.9
Ningxia	2.8	2.2	2.5	2.6	2.2	2.5	2.6
Qinghai	0.1	0.1	0.1	0.1	0.1	0.1	0.1
Shandong	26.3	22.1	24.2	25.1	23.1	24.7	25.3
Shanxi	10.2	9.2	9.7	9.9	9.2	9.7	9.9
Shaanxi	6.2	5.3	5.8	5.9	5.4	5.8	5.9
Shanghai	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Sichuan	10.5	9.6	10.0	10.2	9.5	10.0	10.2
Tianjin	1.2	1.1	1.2	1.2	1.0	1.1	1.2
Xizang	0.0	NA	NA	NA	NA	NA	NA
Xinjiang	10.8	11.7	11.2	11.1	11.7	11.2	11.1
Yunnan	10.3	8.6	9.4	9.8	8.6	9.4	9.8
Zhejiang	0.2	0.2	0.2	0.2	0.2	0.2	0.2
Chongqing	2.6	2.0	2.3	2.4	2.0	2.3	2.4

Note: Current status is based on the 2022 National Statistical Yearbook. Scenario 1: 100% of maize cropland converted to maize/soybean intercropping; soybean cropland unchanged. Scenario 2: 50% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged. Scenario 3: 30% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged.

,

Table 4. Current and Projected Soybean Yield under 2050 SSP1-2.6 and SSP2-4.5 Scenarios (Unit: Million Tons).

		2050 SSP1-2.6 Scenario			2050 SSP2-4.5 Scenario
	Current	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3
China	20.3	89.2	53.7	40.3	90.0	54.5	40.8
Anhui	0.9	2.2	1.6	1.3	2.2	1.6	1.3
Beijing	0.0	0.1	0.1	0.0	0.1	0.1	0.0
Fujian	0.1	0.2	0.1	0.1	0.2	0.1	0.1
Gansu	0.1	1.5	0.7	0.5	1.5	0.7	0.5
Guangdong	0.1	0.3	0.2	0.2	0.3	0.2	0.2
Guangxi	0.2	0.8	0.5	0.3	0.8	0.5	0.3
Guizhou	0.3	0.8	0.5	0.4	0.8	0.5	0.4
Hainan	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Hebei	0.2	6.2	3.2	2.0	6.1	3.2	2.0
Henan	0.8	7.5	4.2	2.8	7.6	4.2	2.9
Heilongjiang	9.5	18.8	14.0	12.2	19.1	14.3	12.4
Hubei	0.4	1.1	0.7	0.6	1.1	0.7	0.6
Hunan	0.3	1.0	0.6	0.5	1.0	0.7	0.5
Jilin	0.7	8.4	4.6	3.0	8.7	4.7	3.1
Jiangsu	0.5	1.4	1.0	0.8	1.4	1.0	0.8
Jiangxi	0.3	0.3	0.3	0.3	0.3	0.3	0.3
Liaoning	0.3	5.4	2.8	1.8	5.3	2.8	1.8
Inner Mongolia	2.5	8.8	5.5	4.3	8.9	5.7	4.4
Ningxia	0.0	0.4	0.2	0.1	0.4	0.2	0.1
Qinghai	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Shandong	0.6	8.0	4.3	2.8	8.1	4.3	2.8
Shanxi	0.2	2.5	1.4	0.9	2.5	1.3	0.9
Shaanxi	0.3	1.8	1.1	0.8	1.8	1.1	0.8
Shanghai	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Sichuan	1.1	3.9	2.2	1.7	3.9	2.2	1.7
Tianjin	0.0	0.4	0.2	0.1	0.4	0.2	0.1
Xizang	0.0	NA	NA	NA	NA	NA	NA
Xinjiang	0.1	3.1	1.4	0.9	3.1	1.4	0.9
Yunnan	0.3	3.2	1.7	1.2	3.2	1.7	1.2
Zhejiang	0.2	0.3	0.3	0.2	0.3	0.3	0.2
Chongqing	0.2	0.9	0.5	0.4	0.9	0.5	0.4

Note: Current status is based on the 2022 National Statistical Yearbook. Scenario 1: 100% of maize cropland converted to maize/soybean intercropping; soybean cropland unchanged. Scenario 2: 50% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged. Scenario 3: 30% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged.

and

Table 5. Current and Projected Soybean Yield under 2050 SSP3-7.0 and SSP5-8.5 Scenarios (Unit: Million Tons).

		2050 SSP3-7.0 Scenario			2050 SSP5-8.5 Scenario
	Current	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3
China	20.3	101.3	59.9	44.1	101.3	60.0	44.1
Anhui	0.9	2.3	1.6	1.4	2.4	1.7	1.4
Beijing	0.0	0.1	0.1	0.0	0.1	0.1	0.0
Fujian	0.1	0.2	0.1	0.1	0.2	0.1	0.1
Gansu	0.1	1.8	0.8	0.5	1.8	0.8	0.5
Guangdong	0.1	0.3	0.2	0.2	0.3	0.2	0.2
Guangxi	0.2	0.8	0.5	0.3	0.8	0.5	0.3
Guizhou	0.3	0.9	0.6	0.4	0.8	0.5	0.4
Hainan	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Hebei	0.2	7.3	3.8	2.4	7.3	3.8	2.3
Henan	0.8	8.5	4.7	3.2	8.8	4.8	3.2
Heilongjiang	9.5	20.4	14.9	12.8	20.4	14.9	12.8
Hubei	0.4	1.2	0.8	0.6	1.3	0.8	0.6
Hunan	0.3	1.0	0.7	0.5	1.0	0.7	0.5
Jilin	0.7	9.9	5.3	3.5	9.8	5.3	3.4
Jiangsu	0.5	1.6	1.1	0.9	1.6	1.1	0.9
Jiangxi	0.3	0.3	0.3	0.3	0.3	0.3	0.3
Liaoning	0.3	6.4	3.3	2.1	6.3	3.3	2.1
Inner Mongolia	2.5	9.9	6.1	4.6	9.7	6.0	4.6
Ningxia	0.0	0.4	0.2	0.2	0.4	0.2	0.2
Qinghai	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Shandong	0.6	9.4	5.0	3.2	9.6	5.1	3.3
Shanxi	0.2	2.9	1.6	1.0	3.0	1.6	1.0
Shaanxi	0.3	2.1	1.2	0.8	2.1	1.2	0.9
Shanghai	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Sichuan	1.1	4.4	2.4	1.9	4.4	2.4	1.9
Tianjin	0.0	0.4	0.2	0.1	0.4	0.2	0.1
Xizang	0.0	NA	NA	NA	NA	NA	NA
Xinjiang	0.1	3.9	1.7	1.1	3.9	1.7	1.1
Yunnan	0.3	3.5	1.9	1.3	3.5	1.9	1.3
Zhejiang	0.2	0.3	0.3	0.2	0.3	0.3	0.2
Chongqing	0.2	0.9	0.6	0.4	0.9	0.6	0.4

Note: Current status is based on the 2022 National Statistical Yearbook. Scenario 1: 100% of maize cropland converted to maize/soybean intercropping; soybean cropland unchanged. Scenario 2: 50% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged. Scenario 3: 30% of maize cropland converted to maize/soybean intercropping, soybean cropland unchanged.

, our results highlight a trade-off between a slight reduction in maize production and a substantial increase in soybean output under different proportions of maize cropland converted to maize/soybean intercropping.

As presented in Table 2 and Table 3, current maize production in China stands at approximately 277 million tonnes. By 2050, under the four climate scenarios with 100% of maize cropland converted to maize/soybean intercropping (Scenario 1), national maize production is projected to decline to between 210 and 247 million tonnes, corresponding to a reduction of 11–24% relative to present levels. Under Scenario 2, in which 50% of maize cropland is converted, production is expected to remain comparatively stable at 244–262 million tonnes, representing a 5–12% decrease, whereas conversion of 30% of maize cropland (Scenario 3) would result in 257–268 million tonnes, only 3–7% below current production.

Our results indicate that converting 30–100% of maize cropland to maize–soybean intercropping could raise China’s national soybean production from the current ~20 million tonnes to 40–101 million tonnes by 2050, with production increasing alongside both the conversion rate and the degree of climate warming (Table 4 and Table 5). Current soybean production in China is estimated at approximately 20 million tonnes. By 2050, under the four climate scenarios with 100% of maize cropland converted to maize/soybean intercropping (Scenario 1), national soybean output is projected to increase to between 89 and 101 million tonnes, with the variation between these projections reflecting the effects of climate warming under the respective scenarios. When 50% of maize cropland is converted (Scenario 2), production is expected to reach 54–60 million tonnes, representing a 170–200% increase relative to current levels, whereas conversion of 30% of maize cropland (Scenario 3) would still yield 40–44 million tonnes—approximately double current national soybean production.

At the provincial scale, the greatest potential is observed in Heilongjiang, Inner Mongolia, Shandong, Jilin, Henan, and Hebei. Under Scenario 2, these provinces are projected to contribute 35.8–39.9 million tonnes, representing a 76–97% increase over the current national output (20.3 million tonnes), thereby identifying them as priority regions for future soybean expansion.

4. Discussion

This study presents an exploratory and novel methodological framework that integrates multi-source field, climate, and soil data with machine learning to evaluate the potential performance of maize/soybean intercropping under projected 2050 climate scenarios. Using this framework, our results indicate that suitable intercropping areas are expected to expand northwards, with the greatest yield gains projected in the North China Plain and Southwest China. Notably, warmer scenarios tend to enhance this advantage, emphasizing the potential of intercropping to improve crop productivity. Under the 50% maize replacement scenario with basic demand, maize self-sufficiency is projected to reach 89.7%, while soybean self-sufficiency reaches 50.9%. Simulation results further indicate that full or partial conversion of maize cropland to maize/soybean intercropping could substantially increase soybean production, effectively doubling current soybean yields under moderate replacement scenarios, while maintaining maize yields within an acceptable range. These findings highlight the strategic potential of maize/soybean intercropping for improving soybean self-sufficiency and promoting sustainable intensification in China under 2050 climate scenarios, while future research is needed to develop more accurate models for improved projection reliability.

Our study proposes a novel methodological approach that integrates multi-source field, climate, and soil data with machine learning to evaluate the potential performance of maize/soybean intercropping. While similar approaches have been widely applied for predicting monocropping yields [26,27], their application to intercropping systems remains rare. We proposed to use the partial Land Equivalent Ratio (pLER), calculated from predicted maize and soybean intercropping yields, as the main parameter to quantify the yield changes resulting from replacing monocropping land with intercropping. Compared with process-based intercropping crop models such as APSIM-strip [38] and MONICA [39], which simulate crop growth based on mechanistic understanding of physiological and biophysical processes, our machine-learning approach bypasses detailed parameterisation and directly learns patterns from multi-source datasets, enabling rapid, large-scale yield prediction across diverse environments. This allows flexible integration of heterogeneous climate, soil, and management data, making it particularly useful for exploratory scenario analysis. However, it provides less insight into underlying physiological processes and may be less robust under conditions not represented in the training data. Currently, some attempts at combining crop models with machine learning have shown promising results [40,41]. As intercropping datasets continue to expand and methodologies advance, such hybrid models may become increasingly feasible, enabling improved predictive performance with relatively lower data requirements.

Our results are consistent with previous studies showing that global warming has promoted the expansion of cultivated land at high latitudes, and intercropping systems to more efficiently utilize enhanced light and temperature resources to improve productivity. In Germany, simulations using the process-based agroecosystem model MONICA showed that winter wheat/soybean intercropping achieved higher land-use efficiency than sole cropping, particularly in warmer regions under high-emission climate change scenarios [39]. Moreover, a recent machine learning–based meta-analysis identified the Yangtze River Basin as one of the most suitable regions for maize/soybean intercropping, with suitability strongly influenced by temperature and sunshine hours [30]. These findings align with our results, showing that warmer climate conditions enhance the yield advantages of intercropping. In addition, sole maize yields in high-latitude regions of China are also expected to increase under rising temperatures [42], while in Europe, sole soybean cultivation could extend northwards to latitudes of 55–57.5° N [26] under future climate change. Together, these findings reinforce the robustness of our results and highlight the broader potential of intercropping systems to enhance yield stability and land-use efficiency under future climate change.

The yield advantage of intercropping increases under climate warming, which stands in contrast to the response of monocropping [43]. Maize/soybean intercropping contributes more to national maize and soybean self-sufficiency under climate warming (Table 1-4). Under conditions of annual rainfall above 800 mm and mean annual temperature above 15 °C, maize/soybean intercropping increases LER by 25% [44]. Sunshine duration was the primary factor contributing to the land equivalent ratio (LER) of maize/soybean intercropping [30]. This improvement is primarily attributed to higher resource-use efficiency in intercropping systems, particularly for light and soil nutrients, resulting from interactions between maize and soybean [45,46]. In addition, intercropping shows higher yield stability than both sole crops across various climate zones [47]. In climate-sensitive regions, such as Southeast Asia, farmers often practice intercropping to cope with uncertain climatic conditions [48]. This evidence indicates that intercropping not only enhances crop productivity but also provides greater resilience to future climate change compared with monocropping, making it a promising strategy for sustaining national food security.

Climate warming is expected to reshape the spatial distribution of maize/soybean intercropping systems across China. In newly suitable high-latitude regions, simultaneous intercropping systems are likely to be adopted, whereas in low-latitude regions, relay intercropping with shorter overlapping periods is expected to become more prevalent [49]. The potential maize/soybean intercropping zones expand northwards, with new areas concentrated in Northeast China and Xinjiang Autonomous Region (Figure 3b–e). In practice, these regions currently rely on single-season cropping could easily transform to intercropping [50]. Relay intercropping is mainly distributed in South, Southwest and Northwest China (Figure 1), where accumulated temperature is too much for single-season and not enough for double-season [49]. Relay cropping, a specific pattern of intercropping in which the second crop is sown before the first is harvested, enables overlapping growth periods, allowing continuous land use, optimizing soil protection [51], improving resource-use efficiency [52], and potentially enhancing system profitability [53]. Under climate warming, relay intercropping, which has shorter overlapping days between the two crops compared with current practices [54], could achieve higher yield advantages and be adopted in Southwest China [11]. In China, however, relay cropping promotion may be constrained in some regions due to the limitations of current cropping systems. For instance, in the North China Plain, where winter wheat–summer maize rotations are prevalent, replacing spring maize with maize/soybean intercropping could disrupt subsequent wheat planting [55]. Therefore, although relay cropping offers higher yield potential, local phenology and cropping practices must be considered. Hence, projected climate warming is expected to be sufficient to allow maize/soybean intercropping in these newly suitable areas, and reducing overlapping days in current relay intercropping is key to further increasing yield advantages.

Maize/soybean intercropping generally follows a “partial maize yield traded for substantial soybean gain” pattern [56]. Our scenario analyses indicate that, by 2050, 100% maize cropland converted to maize/soybean intercropping could achieve near-complete self-sufficiency for maize and soybean under a low-demand scenario (Figure 4). Even converting only 30% of maize cropland would double national soybean production relative to current levels (Figure 4). This has significant strategic implications for reducing China’s long-term reliance on soybean imports [57]. At the provincial scale, Heilongjiang, Inner Mongolia, Shandong, Jilin, Henan, and Hebei emerge as priority regions, projected to contribute 35.8–39.9 million tonnes under the 50% replacement scenario, representing a 76–97% increase over current national output. Overall, the results underscore the potential of intercropping as a practical strategy for improving food security, balancing trade-offs between staple and legume crops, and guiding regional agricultural planning under future climate scenarios.

By systematically simulating potential maize/soybean intercropping areas and productivity, this study introduces a novel methodological approach, while acknowledging several remaining limitations: (1) Database Size Limitation: The relatively low predictive performance of the Random Forest model (R² = 0.59) reflects the limited size and heterogeneity of available intercropping data [58]. Unlike monocropping systems [26,27], intercropping lacks unified national databases, and published studies remain the primary data source. This study is based on 637 records, constrained by diverse crop combinations and management practices. Among existing systems, maize/soybean intercropping provides the most comprehensive dataset in China and forms the basis of this analysis. (2) Scenario assumptions: Future simulations focused on changes in cropland area and cropping systems, without accounting for technological advances that could enhance productivity, such as high-yielding or stress-tolerant cultivars [59], precision fertilization [60], and improved field management [61]. Likewise, potential adverse factors—including increased frequency of extreme events [62], soil nutrient dynamics [63], pest and disease risks [64], and uncertainties in farmer adoption [65]—were not considered. Despite these limitations, this study represents the first application of machine learning–based scenario analysis to intercropping systems. The framework and approach provide valuable insights, and predictive performance is expected to improve with larger, more standardized datasets in the future.

5. Conclusions

This study presents the first systematic assessment of maize/soybean intercropping potential across China under projected 2050 climate scenarios, using an exploratory methodological framework that integrates multi-source datasets with machine learning. Our results suggest that suitable intercropping areas are expected to expand northwards, with the North China Plain and Southwest China showing the greatest yield gains. Scenario analyses indicate that partial or full conversion of maize cropland to intercropping could substantially increase soybean production—effectively doubling current yields under moderate replacement—while maintaining maize yields within acceptable ranges. Under low-demand scenarios, full intercropping adoption could achieve near self-sufficiency in maize and exceed self-sufficiency in soybean, demonstrating the potential of intercropping to enhance national food security.

Importantly, this study emphasizes the development of a transparent and scalable methodological framework that allows for the integration of heterogeneous datasets and provides an exploratory approach to assess intercropping potential under climate change. While the predictive capacity of the model is limited by current data availability, the framework offers a valuable reference for guiding future model refinement, informing agricultural planning, and supporting more resilient and sustainable cropping systems across China.