1. Introduction
Sustainably meeting global grain demand is a critical challenge facing humanity and has attracted significant attention from researchers and policymakers [
1,
2,
3], and recent work has discussed the additional difficulties during the COVID-19 pandemic [
4,
5,
6]. Food security, as defined by the UN’s Sustainable Development Goals (Target 2), means ensuring sufficient supplies of food and nutrients, and grain plays a crucial role in meeting these needs. Global and regional grain demand (GD) had changed significantly in the past half-century due to rapid economic growth, population increase, rural-urban migrations, aging populations, and shifts in dietary structures [
7,
8]. Economic development in some populous countries, particularly the diet transformation, is expected to increase GD, thereby placing enormous pressure on grain supplies within these countries and globally [
9,
10]. China, as the world’s most populous country and the largest food consumer, has a significant impact on global GD, with its soybean, rice, and meat consumption each accounting for approximately 30% of the world’s total consumption by weight during 2013–2019, and seafood accounting for 38% (see
Figure 1). The dominance of China in global food production and consumption has been noted by many, including [
11,
12], especially since the fifth version of the Dietary Guidelines for Chinese Residents (DGCR 2022) was published by the government in 2022. It is likely that the diets of many Chinese people will change towards DGCR 2022 [
13,
14,
15]. If so, there is a need to emphasize the role of changes in dietary structure when predicting China’s GD.
There were various projections of future GD in China (
Table S1) and around the world, and they varied widely; the projected global GD growth from 2005 to 2050 ranges from 35 percent to 110 percent [
5]. Even the near-term projections for China’s GD in 2020, varied from 480 to 741 million tons (
Table S1). This uncertainty in projections has made formulating appropriate grain supply plans challenging for policymakers [
16,
17]. It is important to make more accurate GD forecasting to provide a basis for policymaking in grain production, consumption and trade to ensure grain security, a topic that has become a major international issue.
During the past few decades, a wide range of models and data sources have been applied to GD prediction (e.g., [
18,
19,
20]). The research methods can be roughly classified into three categories [
16]. The first category is “complex mechanistic” models, which are often built on interacting economic, ecological, demographic, and/or climate sub-models [
21,
22,
23,
24]. The second category uses a simple mechanistic approach to predict demand based on a simple (Engel’s Law) relationship between income and kilocalorie consumption [
18,
25,
26,
27,
28,
29]. The third kind, phenomenological models, assume that current trends of increasing kilocalorie consumption will continue [
20,
30,
31,
32]. Different modeling methods had resulted in a variety of forecasts for global or regional GD. However, [
16] found that these differences were not due to the complexity of the models. The purely phenomenological (time-trend) model predictions fell within the same range as simple and complex mechanistic models. While complex models are flexible and include expected changes in agricultural output, food prices, and trade, they are opaque and sometimes irreproducible. Simpler models are easier to interpret and reproduce but may not be as accurate. Therefore, decisions about model complexity should be made based on the research or policy questions of interest when models have comparable forecasting accuracy.
Given this range of approaches, several studies have analyzed the reasons behind varying predictions of GD—different assumed parameters, such as conversion coefficients from animal foods to feed grain (
Table S2), price elasticities, food waste rate, etc. [
10,
33]. Other sources of variation are the projected population, age structure, and especially the dietary structure [
20]. Uncertainties in the growth of meat and dairy consumption in South and East Asia, particularly China and India, were the primary sources of uncertainty in global GD forecasts [
34]. These problems also exist when projecting China’s GD. Only a handful of China GD projections, such as [
19,
35], have emphasized the role of changes in dietary structure. However, they failed to consider food waste and the potential impacts of the COVID-19 pandemic, which had not occurred at the time of their research. Integrating the main impact factors of GD and recalibrating key parameters using the latest information would improve the projections on GD [
1,
5,
34].
We aim to improve short-term prediction of GD by integrating these main drivers of GD, especially dietary structure and demographic structure. To do this, we established a multi-factor model that simultaneously integrates population size, age structure, dietary structure, urbanization, food waste rate, and the impacts of COVID-19. Our model is in the complex mechanistic model category. When we set the values of some parameters in the model, we refer to phenomenological models, assuming that their current trends will continue in the short term. We project China’s GD under three scenarios. In scenario 1 (S1), we assume that the dietary structure would evolve towards the balanced diet guidelines in DGCR 2022 at the annual average pace during 2013–2019, adjusting for the impacts of COVID-19. The results in scenario 1 provide a prediction of China’s annual GD from 2022 to 2025, assuming no special efforts are made to shift diets or reduce food waste during this period. In scenario 2 (S2), we assume that all residents will adopt the DGCR 2022 guidelines during 2022–2025. In scenario 3 (S3), we project GD, again assuming adoption of DGCR 2022 guidelines but ignoring the effect of changing age structure. Comparing these 3 scenarios gives the impact of changing diets and population age structure on China’s future annual GD. We then conduct a sensitivity analysis to assess the impact of different coefficients, such as the ratio of people adopting DGCR 2022, food waste, urbanization rate, etc., on the reduction of GD. Such sensitivity analyses tell us the effects of changes in these factors on reducing GD, thus providing guidance on effective strategies to reduce GD.
We make two contributions to the GD literature. First, we make short-term GD predictions using a “complex mechanistic” model, integrating the main driving factors of population size and its age structure, dietary structure shifts, urbanization, food waste, and the impacts of COVID-19. We updated or calibrated key parameters in the model, such as the standard person consumption ratios, conversion coefficients from animal foods to feed grain, and population size, with the latest survey and statistical data, which were either out of date or missed in previous studies. Secondly, by simulating different scenarios, we show the impacts of different factors on GD, especially aging and shifts in dietary structures. The results identify potential ways to reduce GD in China and should be useful for the government to formulate grain supply plans and policies to effectively reduce GD and at the same time promote healthy diets.
The remainder of the paper is as follows: the second section introduces the research methodology. The third section is the model application in China, and
Section 4 presents the results. We discuss the results, model limitations, and their implications in
Section 5. The last section draws conclusions.
2. Research Methodology
Based on the classification standards of Food and Agriculture Organization of the United Nations (FAO), GD can be divided into staple food grain, feed grain, industrial grain, seed and other grain, as shown in
Figure 2. Due to the diversity of grain uses, we employed a functional decomposition analysis to first decompose food grain demand into staple food grain (denoted
c = 1) and feed grain (
c = 2). We refer to the sum of the two as food grain and is represented by
c = 3. The demand for food grain is subject to greater uncertainty than other types of grain, which is the main driver of fluctuations in GD, as noted by previous studies [
36,
37]. Consequently, we emphasize the calculation of food grain demand in the model. We considered three age groups–0–14, 15–64, 65+—that were used in our food survey, and indexed them by
i = 1, 2, 3. The types of food consumed (the diet) were represented by
j = {Cereal, tubers and beans, Livestock and poultry meat, Aquatic products, …} as listed in
Table 1. The demand for feed grain was calculated based on the conversion rate calculation method. Since the proportion of the other types of grain in China was stable during 2013–2019, we consider the other types of grain by defining a lower bound and an upper bound on their historical change trend, following the approach in [
30,
38].
We establish a multi-factor model of GD prediction, incorporating the major drivers—shifts in dietary structure, changes in population size, age structure, urbanization, and food waste rates. Since we are projecting for the near term 2022–2025, we also integrate the effect of COVID-19. There are other important factors that are known to impact GD, in particular, food prices and income. We do not explicitly include these effects but represent them via changes in diets (changes in the types of food consumed). Refs. [
10,
39,
40] showed that dietary preferences were mostly determined by prices and household incomes, and our explicit treatment of dietary structures gave an indirect treatment of these factors.
Let
denote the quantity of food type
j consumed on average by a person in age group
i in year
t, and
denote the coefficient transferring food type
j to grain requirement of type
c.
is calculated with
Equations (s1)–(s4) in the Supplementary Materials. Equation (1) gives the amount of grain
c consumed by a group
i person in year
t (
) as the sum over all the types of grain requirements transferred to food
j:
Much of the literature used the concept of a standard person and expressing the requirement of a person of age
i relative to a standard age group. According to FAOSTAT, a person aged 17–18 years old had the highest energy demand, which then falls with age. On average, the energy demand of a 60–69-year-old person is 70.3% of the peak value by age, while an 80–89-year-old person has a ratio of 49.9% [
41]. For the food consumption by differently aged persons in China, a national survey conducted by the University of Chinese Academy of Sciences (UCAS) in 2017 [
42] shows that the volume of food consumption was the highest for those aged 15–64 for all types of food except for milk and dairy products (see
Figure 3).
We, therefore, define the standard person consumption ratio for type
c grain by persons in age group
i,
, as the ratio of their average grain demand to the demand by age group
i = 2 (15–64 years old), giving a simple account for the impact of age structure on GD:
We define the number of standard persons in age group
i, gender
k, in year
t,
, as:
where
is the population of age group
i, gender
k in year
t.
In Equation (4), we write the actual per capita gross demand for type
c grain by a gender
k person in urban areas in year
t (
as the grain embodied in the sum over all food types:
where
represents the per capita consumption of the
jth kind of food by an urban person of gender
k with the actual diet in year
t.
is the rate of wastage of the
jth kind of food in the consumption stage in year
t.
is the gross grain demand that includes food wasted in actual consumption.
We next consider how this gross consumption of grain
c would change over time with growth in per capita consumption and accounting for pandemic effects. The per capita grain demand in year
t + 1 of an urban person of gender
k (
is given by:
where
is the average annual growth rate of
in urban areas during the normal years without COVID-19, and
is the adjustment to the growth rate for the impacts of COVID-19.
Equation (6) gives the gross demand for grain type
c of a rural person of gender
k in year
t (
, where
represents the actual per capita demand for the
jth kind of food by a rural person The rate of food wastage in the consumption stage was used as the national average level.
the impacts of COVID-19 on the annual growth rate of rural grain demand is represented by
, and the average annual growth rate of
in rural areas in normal years is
.
Adding over the two types of grain gives the food grain demand on average by a person of gender
k in urban and rural areas in year
t,
and
, respectively.
We next set up the model for S2 based on the dietary guidelines in DGCR 2022. These guidelines are given for three levels of food consumption–low, medium, and high–which we denote by
d = {1, 2, 3}. We first define the per capita daily demand for grain
c by a person in the 15–64 age group in year
t under the guidelines. This demand for grain
c due to food type
j in grams, at level
d (
, is given by the recommended daily consumption of food type
j,
, transferred to grain
c equivalents, and adjusted for food wastage:
The annual demand for grain
c per person in the 15–64 age group, under the guidelines at level
d, in kilograms (
, is the sum over the food types and annualizing:
is thus the food grain demand for the
d = high level in the guidelines. The average food consumption in age group 15–64 is the highest among the three age groups in our survey [
35], and a male consumes more food than a female on average [
43]. Following [
44], we set the guideline grain
c demand by a male (
k = 1) in the standard age group (15–64), in both urban (
) and rural areas (
), equal to this high level (
d = 3) in Equation (12). The guideline demands for a standard age female (
k = 2) in urban and rural areas are then expressed as a coefficient,
, multiplied by the male demands in (13) and (14), respectively:
To summarize so far, in Step 2, we incorporated food waste and COVID-19 factors through the parameters and . Equations (4)–(9) correspond to S1, while Equations (10)–(14) are for S2 where diets change towards the guidelines.
We next define the national food grain demand in S1,
. This is given by the sum over urban and rural demands; urban demand is given by the food grain demand per standard person in the urban area (
in Equation (8)) multiplied by the standard population and the urbanization rate
, and rural demand is similarly defined:
The total urban standard population is the standard population at age group
i multiplied by the urbanization rate and summed over
i. Similarly, the rural standard population is the standard population at age group
i multiplied by 1 minus the urbanization rate and summed over
i. The food grain demand in S2 is derived from the diet guideline demands per standard person given in (12)–(14), multiplied by the respective urban and rural standard persons, and summed over the 3 age groups and 2 genders:
This equation takes into account the food grain demand of a standard person in an urban area with a balanced diet in year t () and that of a standard person in a rural area ().
S3 applies the traditional per capita method that does not distinguish between the different age groups to predict food grain demand under the diet guidelines. Let the urban population of gender
k in year
t be
and the total population be
. Recall that the per person food grain demand for the 15–64 age group under the guidelines are given as
and
in Equations (12)–(14). We first defined the per capita food grain demand in year
t (
) as the weighted average of the urban and rural demands, where the population weights are given by
and
, and summed over the genders:
The total food grain demand in S3 (
) is then this per capita demand multiplied by the total population:
Let
be the lower bound of the ratio of the rest of grain demand (i.e., excepting food grain) to total GD, and
be the upper bound ratio. The setting of these bounds is described later in
Section 3.1. For each scenario
s, the upper bound of the total grain demand (food grain plus the rest),
, and the lower bound,
, are then given by:
Finally, we calculate the total grain demand in scenario
s in year
t (
) as the average of the lower and upper bounds:
In the final step we compare the scenarios and decompose the changes in total grain demand over time. Recall that S1 uses the historical growth rate of consumption, S2 is the transition to the diet guideline, and S3 ignores the age structure of the population. First, we trace the diet guideline effect by calculating the difference rate between S2 and S1 as:
Then we trace the aging effect with the difference between S3 and S2 as:
We calculate the total staple food grain demand in S1 (
as the per person demands for staple food grain (
c = 1) multiplied by the standard population and summed over the age and gender groups:
The staple food grain demand in S2 (
) is staple food demand per standard person under diet guidelines multiplied by the standard populations in urban and rural areas and summed over the age groups and genders:
The feed grain demand in S1 and S2 is the total food grain demand less the staple food grain demand, respectively:
Given these demands for the total grain, staple food grain and feed grain, we now defined their differences between S1 and S2, respectively, as:
The share contributions of staple food grain and feed grain differences,
and
, to
are defined as:
We call our model of grain demand (GD) a multi-factor model since it integrates dietary structure, population size, population age structure, urbanization, food waste, and COVID-19 as the main impacting factors of GD. Unlike previous research, which focused on some of these factors, our model provides a more complete representation of the driving factors. Policymakers and grain companies can adapt this model to develop effective grain policies and market strategies. While the model is not exhaustive, it is multidimensional and integrative, allowing stakeholders to modify and incorporate their specialized knowledge to add to the trends identified in this study. That is, the model can act as a building block to guide future research efforts at a higher level of detail.
5. Discussion
We report the projections of GD made by other studies in
Table S1, noting whether each study considered the age structure of the population or urbanization. They used various methods, including regressions or estimating consumption demand functions. We see that the only study that considered population age structure, [
70] projected GD that was lower than the other studies that ignored it. This is consistent with our comparison of S2 and S3. Next, we consider the effect of projections of population size. Ref. [
38] projected that the population would peak at 1441.6 million in 2029 and forecasted GD to be 652.06 million tons in 2025, peaking at 676.2 million tons in 2035. However, we forecasted that the population had already peaked in 2021, with a size 2.0% smaller than that of [
38]. In S1 we predicted that the lower and upper bounds of China’s GD (
and
) in 2025 would be 649.77 and 666.55 million tons, respectively. Besides having a different peak population, [
38] also did not consider the age structure of the population. These are the main reasons why their peak GD (676.2 million tons) is larger than our
(666.55 million tons) in 2025, the highest value we predicted between 2022–2025.
We recognize some limits to the implementation of our model here. First, due to survey limitations, we were only able to distinguish three age groups (0–14, 15–64, and 65 and over) when calculating the standard person consumption ratios,
. If we had access to more detailed age group data, we would most likely project an even lower GD, i.e., the calculated ratio between S3 and S2,
, may have been even larger. Secondly, we considered gender in scenario S1, but in scenario S2, we did not have separate diet recommendations for males and females and were unable to account for a change in gender structure, which may lead to a slightly different GD. Thirdly, we should note that, as discussed in the review of models by [
16], they did not identify any of the complex model studies reporting attempts at model verification by discussing AIC or R2 statistics or employing cross-validation approaches due to data limitations; papers using simpler mechanistic models did report some form of model validation. Similarly, our model integrates numerous impact factors with multiple parameters; unfortunately, there is not enough time series data to apply statistical methods to estimate these parameters. For some parameters such as
,
,
, it is difficult to find even just one recent year of data. Fortunately, from a qualitative perspective, their short-term fluctuations are not significant. Some recent studies have conducted statistical analysis on the trend changes of some parameters, such as the COVID adjustment (
and food waste (
). We apply the values of
and
. that have been consistently estimated in the literature. The parameters for standard persons and food-grain conversion (
,
had been estimated with long lags and differ significantly across studies and we recalibrated them using the latest data. We assumed that they would remain on their current trend during our short-term prediction period of 2022–2025. This short-term forecasting assumption is used in many previous studies that use phenomenological (time-trend) models, including those that project out to 2050 [
16,
20]. We endeavored to be transparent with our determination of parameter values and presented the results of the sensitivity analyses, which we hope are sufficient to give a sense of the range of uncertainty.
This study provides some useful practical results for grain policymakers, grain companies, and anyone interested in enhancing grain security. First, our model allows for a more comprehensive understanding of the various factors that contribute to GD, enabling policymakers and officials to better prepare for uncertainties in the future and help target grain supply policy. Better policymaking will benefit both producers and consumers. Second, our results show that adopting the recommended diets would not lead to an increase in China’s GD but would benefit human health and the environment. Our calculation of the drop in GD provides a rigorous basis for promoting the adoption of the dietary guidelines in DGCR 2022. Third, [
71] estimated the annual food loss and waste in China were at least 120 million tons, making it a significant concern that cannot be overlooked. Our sensitivity analysis indicates that reducing consumption of food waste, particularly cereal, livestock and poultry, and milk waste, would have the most significant effects on reducing GD. This points towards the need for effective policies to reduce food waste. Fourth, we noted that [
72] estimated that global food consumption alone could add nearly 1 °C to warming by 2100, with 75% of this warming driven by high-methane foods such as ruminant meat, dairy, and rice. On the other hand, they also argued that simultaneous improvements in production practices, the universal adoption of healthy diets, and reductions in consumer- and retail-level food waste could avoid over 55% of the anticipated warming. Projections based on our integrated model could estimate the methane emissions and other greenhouse gases from China’s grain consumption and show how these emissions may be reduced by promoting a healthy diet and reducing food waste. Such research would have practical and significant implications. Finally, we calculated that the inclusion of aging effects reduces projected GD by 3.8%. This is equivalent to an annual savings of around 0.3 billion RMB in grain inventory costs. These findings are not only relevant for China but can also serve as a valuable reference for predicting GD in other countries that are facing similar demographic challenges related to aging populations.