Climate risk and the production foundations of nutrition security: evidence from grain-supply capacity in Chinese cities

1 Introduction

Climate change and extreme weather events are profoundly affecting global food systems and creating sustained pressure on food availability, nutrition security, and public governance. For nutrition security, a stable food supply determines whether residents can obtain sufficient food and also affects food-price stability, diet affordability, and food-system resilience. In recent years, risks associated with extreme heat, drought, heavy rainfall, floods, and precipitation anomalies have become increasingly frequent. Their effects are no longer confined to agricultural production, but may be transmitted to food supply, household income, market prices, and nutrition-security systems. Fanzo et al. (2025) argue that long-term climate change and short-term extreme weather events influence food security, diet quality, and nutrition outcomes through food, water, health, and social-protection systems. Hadley et al. (2023) also identify reduced food availability as one of the key mechanisms linking climate-related shocks to food insecurity.

Agricultural production is the foundation of the food system and the most direct channel through which climate risk affects food security and nutrition security. Existing studies show that climate change influences crop growth and agricultural productivity by altering temperature, precipitation, water availability, and exposure to extreme weather. Ortiz-Bobea et al. (2021), using global data on agricultural productivity, find that anthropogenic climate change has significantly slowed the growth of agricultural total factor productivity. Jägermeyr et al. (2021), based on a new generation of climate-crop models, show that the negative effects of climate change on global staple crop yields may emerge earlier than previously expected. Lesk et al. (2022) further emphasize that compound extremes such as heat, drought, and excess moisture jointly affect global crop yields. These studies suggest that climate risk is not only a long-term ecological issue but also a direct and immediate shock to grain-supply capacity and food-system stability.

China is a major producer and consumer of grain, making the implications of climate risk for China's food security particularly important. On the one hand, Chinese regions differ substantially in climatic conditions, agricultural infrastructure, fiscal capacity for agricultural support, mechanization, and irrigation conditions, so the effects of climate risk on grain supply may display pronounced spatial heterogeneity. On the other hand, rapid urbanization means that urban food security depends not only on distribution and consumption systems but also on local and regional production capacity. Lee et al. (2024) find, using provincial panel data for China, that climate change significantly affects food security and that agricultural policies can partly mitigate these adverse effects. Cui and Zhong (2024) show that climate change can reshape cropland allocation and crop-planting patterns, thereby influencing food security. Although these studies provide important evidence, most existing research is conducted at the global, national, provincial, or household level; systematic city-level evidence remains limited.

Against this background, this study constructs a city-year panel dataset for 298 prefecture-level cities in China during 2010–2023 and investigates how climate-risk shocks affect production-side urban food security. The logarithm of total grain output is used to measure production-side food security, while a climate-risk index captures the intensity of comprehensive climate shocks faced by each city in each year. The study identifies the effects of extreme high temperature, extreme low temperature, extreme rainfall, and drought on urban food availability. It further examines mechanisms related to agricultural mechanization, effective irrigation, sown area, and fiscal support for agriculture, and evaluates broader agricultural economic consequences.

Accordingly, this study addresses three main research questions. First, does greater climate risk reduce urban grain-supply capacity and thereby weaken production-side food security? Second, through which agricultural production channels—including mechanization, effective irrigation, grain-sown area, and fiscal support for agriculture—is this effect transmitted? Third, does the impact of climate risk vary across cities with different grain-production functions and levels of adaptive capacity? By answering these questions, the study seeks to clarify not only whether climate risk threatens urban food availability, but also how and under what conditions such effects arise.

The marginal contributions of this study are threefold. First, it shifts the analysis of climate risk and food security to the prefecture-city level in China. Compared with studies based on national or provincial data, city-level evidence makes it possible to examine more localized differences in climate exposure, agricultural conditions, and food-production capacity. Second, the study does not stop at estimating the overall effect of climate risk on grain output, but further examines several production-related channels, including agricultural mechanization, irrigation, grain-sown area, and fiscal support for agriculture. Third, the analysis extends beyond grain output to consider the broader effects of climate risk on agricultural performance and food-supply stability. In this way, the study provides a more detailed picture of how climate risk affects production-side food security in Chinese cities.

2 Literature review

2.1 Climate risk, food systems, and nutrition security

Climate change and extreme weather events have become important external shocks to the stability of global food systems and nutrition security. Recent research has moved beyond the effects on agricultural yield alone and has increasingly examined broader dimensions such as food availability, dietary quality, and nutrition outcomes. Fanzo et al. (2025) note that long-term climate change and short-term extreme weather events influence food security, diet quality, and nutrition through food, water, public health, and social-protection systems. Hadley et al. (2023) show that, after climate-related shocks, food insecurity is often generated through reduced food supply, livelihood losses, higher food-access costs, and increased social vulnerability. Mirzabaev et al. (2023) further stress that droughts, floods, extreme heat, and precipitation anomalies increase food insecurity and nutrition risks, especially for low-income groups, nutritionally vulnerable populations, and regions with weak adaptive capacity.

These studies indicate that climate risk is not only an agricultural production issue but also a food-system and nutrition-security issue. At the city level, local grain production cannot be equated with residents' final nutrient intake; nevertheless, stable grain-supply capacity is a critical foundation for food availability and nutrition security. When climate risk leads to output losses, supply fluctuations, and price instability, it may further constrain residents' ability to obtain sufficient, stable, and affordable food. An analysis based on production-side food security therefore provides foundational evidence for understanding food-system resilience and nutrition-security protection under climate change.Taken together, this literature establishes food availability as an important link between climate shocks and nutrition security. This also points to the need to examine more directly how climate risk affects the production capacity that underpins local food supply.

Food security is inherently multidimensional, and its empirical measurement depends on the specific dimension under study. Barrett (2010) emphasizes that no single indicator can fully represent food insecurity, while Headey and Ecker (2013) similarly show that measures of food availability, access, utilization, and stability capture different aspects of the concept. This distinction is directly relevant to the empirical design of this study. City-level grain output does not measure household food access, dietary quality, or nutrition outcomes; instead, it captures the production-side capacity that contributes to local food availability. The empirical outcome is therefore interpreted as a measure of local grain-production and supply capacity rather than as a comprehensive food-security index.

2.2 Climate risk, agricultural production, and grain supply

Agricultural production depends heavily on temperature, precipitation, soil moisture, and farmland infrastructure, and is therefore one of the sectors most directly exposed to climate risk. Ortiz-Bobea et al. (2021) find that anthropogenic climate change has slowed global agricultural productivity growth, suggesting a long-term constraint on agricultural production capacity. Jägermeyr et al. (2021) show that negative climate impacts on major crop yields may emerge earlier than previously anticipated, especially in key grain-producing regions. From the perspective of compound extremes, Lesk et al. (2022) argue that extreme heat, drought, and excessive moisture may co-occur within the same growing season and jointly affect crop yields through physiological stress, soil-moisture changes, and heat-moisture interactions.

More specifically, climate risk affects grain output through crop growth, water supply, farmland conditions, and production management. Crop yields respond nonlinearly to temperature: once temperatures exceed crop-specific thresholds, yields decline markedly (Schlenker and Roberts, 2009). Extreme heat can accelerate crop development, shorten the growth period, and disrupt pollination, grain formation, grain filling, and maturation, thereby lowering output per unit area (Hatfield and Prueger, 2015; Zhao et al., 2017). Drought reduces soil moisture, weakens irrigation security, and increases water stress. Lesk et al. (2016) show that droughts and extreme heat significantly reduce cereal yields at the global scale. Excessive rainfall and floods can also damage grain production by causing waterlogging, poor soil aeration, lodging, pest and disease pressure, and nutrient losses, thereby affecting sowing, field management, and harvesting efficiency (Kim et al., 2024). Climate variability itself is also an important source of yield volatility; Ray et al. (2015) find that temperature and precipitation variability explain a substantial share of global crop-yield variability. Thus, climate risk can reduce output levels and increase uncertainty in grain production. These findings show that climate shocks can affect grain supply through multiple production-related pathways rather than through yield effects alone. The extent of these losses, however, is likely to depend on local production conditions and the capacity to adapt to climate shocks.

Agricultural-economics research also provides an important empirical basis for using weather variation to study agricultural outcomes. Deschênes and Greenstone (2007) exploit year-to-year fluctuations in temperature and precipitation to estimate effects on agricultural outcomes, while Schlenker and Roberts (2009) demonstrate strongly nonlinear crop responses to extreme temperature exposure. This literature shows that short-run weather variation can provide informative variation for estimating climate-related production responses, while also highlighting the importance of the outcome and spatial scale considered. The present study therefore shifts the focus from a single crop or yield measure to annual prefecture-level grain output, which reflects both yield changes and adjustments in cultivation scale.

2.3 Agricultural production conditions, adaptive capacity, and food-security resilience

The impact of climate risk on food security is not fully exogenous or unavoidable. Agricultural production conditions and adaptive capacity shape the strength with which climate shocks are transmitted into grain-output losses. Agricultural mechanization, effective irrigation, fiscal support, agricultural insurance, and technological progress can enhance the risk resistance and recovery capacity of agricultural systems. Fang et al. (2024) provide evidence from China that agricultural mechanization strengthens the climate resilience of food production. Higher mechanization improves the efficiency of tillage, plant protection, harvesting, and drying, and helps farmers complete emergency sowing, harvesting, and post-disaster recovery after extreme weather.

Production conditions are an important channel linking climate risk to food security. Rising temperature, drought, precipitation anomalies, and changing water availability affect the yields of major grain crops, and heat stress and water stress often reinforce each other, raising crop sensitivity to high temperature (Lesk et al., 2021; Rezaei et al., 2023). Rezaei et al. (2023) identify water management and irrigation as key adaptation measures, while Zhang et al. (2023) show that sufficient irrigation can substantially reduce drought-induced losses in wheat production in China. High-standard farmland construction can also improve land quality, drainage and irrigation, and disaster resistance, thereby stabilizing grain production. Hao et al. (2024) find that high-standard farmland construction increases grain output through channels such as reducing drought and flood losses and transforming low- and medium-yield farmland.

Fiscal support for agriculture can improve production resilience through irrigation and water-conservancy construction, high-standard farmland, agricultural technology extension, disaster prevention, and agricultural insurance subsidies. Wan et al. (2024) show that fiscal agricultural policies and agricultural science-and-technology innovation jointly improve agricultural-system resilience. Nevertheless, the effect of insurance subsidies depends on institutional design; Chemeris et al. (2022) suggest that insurance subsidies may influence yield resilience by changing farmers' risk-management behavior. Accordingly, regions with lower mechanization, insufficient irrigation, and weaker fiscal capacity are more likely to see climate shocks converted into output losses and supply volatility, while regions with stronger agricultural foundations have greater adaptive capacity. This literature suggests that the effects of climate risk are conditioned by differences in agricultural infrastructure, public support, and adaptive capacity. Such differences are particularly relevant in China, where regions vary substantially in production conditions, fiscal capacity, and grain-production functions.

The agricultural-economics literature also cautions against assuming that adaptation fully offsets climate damages. Burke and Emerick (2016), using evidence from U.S. agriculture, find that observed long-run adaptation offsets only a limited share of the productivity losses associated with extreme heat. This evidence provides an additional rationale for examining differences in local adaptive capacity rather than treating adaptation as a uniform response. In the empirical analysis below, this perspective is reflected in the heterogeneity analysis across cities with different levels of adaptive capacity.

2.4 Agricultural markets, spatial adjustment, and interregional food flows

Local production shocks do not necessarily translate one-for-one into local food scarcity because agricultural markets are spatially connected. Trade, transportation infrastructure, inventories, and interregional flows allow supply to adjust across locations and can partly smooth geographically concentrated production shocks. Burgess and Donaldson (2010) show that greater market openness can reduce vulnerability to weather shocks, while Donaldson (2018) demonstrates that lower transport costs reduce interregional price gaps and increase trade. These studies imply that the food-security consequences of local climate shocks depend not only on local production losses but also on the degree of market integration.

Related research further shows that climate-induced productivity changes can alter comparative advantage and agricultural trade patterns. Costinot et al. (2016) demonstrate that production and trade reallocation can substantially shape the aggregate market consequences of heterogeneous climate impacts across locations. This perspective is particularly relevant for China, where prefecture-level cities are embedded in interregional grain markets rather than operating as closed food systems. A decline in local grain output should therefore be interpreted as a deterioration in local production-side supply capacity, not as an equivalent decline in the total food available to local residents.

This market-integration perspective also clarifies the boundary of the present empirical design. The baseline regressions focus on local grain-production capacity, while grain-price volatility and the local supply gap are examined only as supplementary outcomes. Because the available city-level data do not directly observe bilateral grain flows or spatial trade networks, the paper does not attempt to identify interregional transmission effects. Spatial market adjustment is instead treated as an important limit to the interpretation of local production outcomes and as a direction for future research.

2.5 Climate risk and food security in the Chinese context

China's role as a major grain producer and consumer makes climate risk highly relevant for food security. Regions differ substantially in climate conditions, grain-production functions, agricultural infrastructure, fiscal capacity, and industrial structure. Consequently, the effects of climate risk on food security may be spatially heterogeneous. Lee et al. (2024) demonstrate that climate change significantly affects food security in China and that agricultural policy can partly mitigate the adverse effects. Cui and Zhong (2024) show that climate change can alter cropland allocation and planting structure, with implications for food security. Fu et al. (2023) find that extreme rainfall has caused large losses in China's rice yields over the past two decades, indicating that precipitation anomalies and extreme weather have become important climate risks for Chinese grain production.

Existing research provides an important basis for understanding the relationship between climate change and China's food security, yet several gaps remain. First, evidence at the prefecture-city level is still limited, although city-level data can more accurately capture spatial differences in climate exposure, infrastructure, fiscal capacity, and production conditions. Second, many studies focus on agricultural yield or comprehensive food-security indices, whereas the links among production-side food security, food availability, and supply stability require further analysis. Third, the mechanisms through which climate risk affects food security require more direct identification, particularly through mechanization, irrigation, sown area, and fiscal support. Fourth, climate risk may have broader economic consequences beyond output losses, such as lower agricultural value added, lower rural income, higher costs, greater fiscal disaster-relief pressure, and stronger price volatility.

Overall, the Chinese literature confirms the importance of climate risk for food security, but evidence remains relatively limited at the prefecture-city level. This leaves room for a more detailed analysis of how climate risk affects local grain-supply capacity, through which channels these effects occur, and whether they differ across cities.

2.6 Summary of the literature

Taken together, the literature suggests three distinctions that directly shape the empirical design of this study. First, the climate-agriculture literature provides strong evidence that temperature, precipitation, and extreme weather affect crop yields and agricultural productivity, but much of this evidence is crop-specific or estimated at broader spatial scales. This motivates our use of a prefecture-level panel to examine annual local grain-production capacity, which captures both yield and cultivation-scale responses. Second, the food-security literature distinguishes production availability from household access, utilization, and nutrition outcomes. We therefore use total grain output as a measure of production-side supply capacity rather than as a comprehensive indicator of food security, and the mechanism and heterogeneity analyses are correspondingly restricted to production conditions and adaptive capacity. Third, agricultural-market and trade research shows that local production shocks can be transmitted or partly absorbed through transportation, market integration, and interregional trade. Because bilateral grain flows are not observed in the present data, local production losses are not interpreted as equivalent changes in total local food availability, and the supplementary analyses of price volatility and the supply gap are not used to identify interregional market transmission. Against this literature, the contribution of this study lies less in establishing that climate affects agriculture—a relationship that is already well documented—than in connecting multi-hazard climate exposure to prefecture-level grain-production capacity in China, examining several production-related channels within a common city-year framework, and identifying heterogeneity across grain-production functions and adaptive capacity. This design provides localized evidence on the production side of food security while explicitly separating local production effects from broader market-access and nutrition outcomes.

3 Theoretical analysis and research hypothesis

3.1 Climate risk and production-side food security

Grain production is highly dependent on climatic conditions. Changes in temperature, precipitation, and water availability directly affect crop growth and agricultural operations, while extreme heat, drought, heavy rainfall, flooding, and low-temperature events can disrupt sowing, field management, irrigation, and harvesting. These shocks may reduce grain output and increase uncertainty in agricultural production. Previous studies also show that crop yields respond nonlinearly to temperature and that adverse climate conditions can generate substantial losses in major crops (Schlenker and Roberts, 2009; Lobell et al., 2011).

From the perspective of production-side food security, grain output reflects the capacity of a region to maintain a stable food supply. Climate risk can weaken this capacity in two related ways. First, extreme weather can directly damage crop growth and reduce yield. Second, repeated or severe climate shocks can disrupt normal production activities and increase uncertainty in agricultural decision-making, making stable grain production more difficult to maintain. As climate risk rises, these effects are more likely to translate into lower grain output and greater pressure on local food availability. Accordingly, we propose the following hypothesis:

Hypothesis H1: Higher climate risk significantly reduces city-level grain output and weakens production-side food security.

3.2 Mechanisms linking climate risk to production-side food security

The effect proposed in H1 may arise not only from the direct impact of climate shocks on crops, but also from changes in the conditions under which agricultural production takes place. Grain production depends on the timely use of machinery, reliable access to irrigation, sufficient sown area, and public support for agricultural infrastructure and disaster response. Climate shocks can affect each of these conditions and thereby transmit into grain-output losses.

In this sense, H1 captures the overall effect of climate risk on production-side food security, whereas the following hypotheses identify four production-related channels through which this effect may occur.

3.2.1 Agricultural mechanization

Agricultural mechanization improves the efficiency and timeliness of sowing, field management, plant protection, harvesting, transportation, and post-disaster recovery. It also reduces excessive dependence on manual labor and narrow weather windows. When extreme weather occurs, however, machinery operations may be delayed or interrupted, while farmland infrastructure, field roads, and agricultural equipment may also be affected. These disruptions can reduce the effective contribution of mechanization to grain production.

The ability to maintain mechanized operations is therefore closely related to the resilience of agricultural production under climate shocks. Where mechanization conditions are weaker, producers may have greater difficulty completing key agricultural operations within suitable time periods or restoring production after extreme weather. Climate risk may consequently reduce grain output partly by weakening mechanized production capacity. Accordingly, we propose:

Hypothesis H2: Climate risk weakens production-side food security through the agricultural mechanization channel.

3.2.2 Effective irrigation

Water availability is another important condition for stable grain production. Effective irrigation helps maintain water supply during critical stages of crop growth and reduces dependence on natural precipitation. This is especially important under drought and extreme heat, when soil moisture declines and crop water demand increases.

Climate risk can weaken this production condition by increasing water shortages, disrupting irrigation schedules, or placing additional pressure on existing irrigation facilities. Where effective irrigation is insufficient, climatic shocks are more likely to translate into water stress and grain-output losses. In contrast, stronger irrigation conditions can improve the capacity of agricultural production to withstand fluctuations in temperature and precipitation. Accordingly, we propose:

Hypothesis H3: Climate risk weakens production-side food security through the effective irrigation channel.

3.2.3 Grain-sown area

Total grain output depends not only on yield per unit area but also on the scale of grain cultivation. Extreme drought, flooding, heavy rainfall, and low-temperature events can interrupt sowing, reseeding, and other field operations, reducing the area that can be effectively cultivated. Climate risk may also increase uncertainty over expected returns from grain production and influence farmers’ planting decisions.

When producers face greater uncertainty or higher expected losses from extreme weather, they may reduce grain cultivation or adjust the allocation of land and labor. A decline in grain-sown area therefore provides another channel through which climate risk can reduce total grain output and weaken local grain-supply capacity. Accordingly, we propose:

Hypothesis H4: Climate risk weakens production-side food security through the grain-sown-area channel.

3.2.4 Fiscal support for agriculture

Agricultural production under climate risk also depends on public support. Fiscal expenditure for agriculture helps finance farmland water conservancy, agricultural infrastructure, technology extension, disaster prevention, production support, and post-disaster recovery. These public inputs can improve the ability of agricultural systems to respond to and recover from climate shocks.

Climate events may simultaneously increase the need for public intervention while placing pressure on local fiscal resources. Where fiscal support for agriculture is limited, damaged infrastructure may not be repaired promptly, disaster-prevention capacity may remain weak, and producers may receive insufficient support for restoring production. Stronger fiscal support, by contrast, can help maintain production conditions and reduce the extent to which climate shocks translate into grain-output losses. Accordingly, we propose:

Hypothesis H5: Climate risk weakens production-side food security through the fiscal-support-for-agriculture channel.

Taken together, these hypotheses form a unified analytical framework. H1 examines whether climate risk reduces production-side food security at the city level, while H2–H5 explain how this effect may be transmitted through changes in agricultural mechanization, effective irrigation, grain-sown area, and fiscal support for agriculture. This framework therefore links the overall effect of climate risk to the specific production conditions examined in the subsequent empirical analysis.

4 Research design

4.1 Model specification

To identify the effect of climate-risk shocks on production-side food security, the following two-way fixed-effects model is estimated:

FoodSec i , t = β 1 ClimateRisk i , t + β 2 X i , t + u i + ϱ t + α + ε i , t (1)

In Equation 1, FoodSec i , t denotes production-side food security in city i and year t; ClimateRisk i , t is the climate-risk index; X i , t is a vector of control variables; u i represents city fixed effects; ϱ t represents year fixed effects; and ε i , t is the random error term. The coefficient of interest is β 1. A significantly negative β 1 indicates that higher climate risk significantly reduces urban grain-supply capacity.

To identify the transmission mechanisms, this study applies a three-step mechanism framework. The first step estimates the total effect using Equation 1. The second step, as specified in Equation 2, tests whether climate risk affects the mechanism variable:

Mechanism i , t = β 1 ClimateRisk i , t + β 2 X i , t + u i + ϱ t + α + ε i , t (2)

Here, Mechanism i , t denotes the mechanism variable, including agricultural mechanization, effective irrigation, grain-sown area, and fiscal support for agriculture. The third step, as specified in Equation 3, adds the mechanism variable to the food-security regression:

FoodSec i , t = β 1 ClimateRisk i , t + β 3 Mechanism i , t + β 2 X i , t + u i + ϱ t + α + ε i , t (3)

If the mechanism variable significantly affects food security and the absolute value of the climate-risk coefficient decreases relative to the baseline model, the mechanism variable is interpreted as playing a partial mediating role.

4.2 Variable definitions

4.2.1 Dependent varable

The dependent variable is production-side food security (FoodSec), measured by the natural logarithm of total city-level grain output. This study focuses on the production and availability dimension of food security rather than household food access or dietary outcomes. Total grain output is therefore used because it directly reflects the amount of grain supplied by local agricultural production and captures both yield and cultivation scale. Compared with yield per unit area, total output is more consistent with the study’s concern with overall grain-supply capacity at the city level. It is also consistently available across prefecture-level cities over the sample period, which facilitates comparable panel analysis. Taking logarithms reduces right-skewness and the influence of extreme values and allows the estimated coefficients to be interpreted approximately as percentage changes.To address potential concerns arising from differences in population size and cultivated land across cities, per capita grain output and grain yield are further used as alternative dependent variables in the robustness checks.

4.2.2 Explanatory variable

The core explanatory variable is the climate-risk index (ClimateRisk), which measures the intensity of comprehensive climate shocks faced by a city in a given year. It is constructed from extreme high temperature, extreme low temperature, extreme rainfall, and extreme drought. Extreme high temperature is measured by the number of days on which daily maximum temperature exceeds a high-temperature threshold; extreme low temperature is measured by the number of days on which daily minimum temperature is below a low-temperature threshold; extreme rainfall is measured by rainfall events exceeding the 95th percentile; and drought is measured by drought frequency. The indicators are standardized and aggregated using entropy weights. A larger index indicates stronger exposure to extreme-weather shocks.

The composite index is intended to summarize annual multi-hazard climate exposure rather than to imply that extreme heat, extreme cold, excessive rainfall, and drought have identical agronomic effects. These hazards differ in their physiological effects, timing, crop sensitivity, and regional relevance (Schlenker and Roberts, 2009; Hatfield and Prueger, 2015; Lesk et al., 2016; Kim et al., 2024). Entropy weighting is used as a data-driven aggregation procedure after standardization, so the resulting weights reflect differences in the information and dispersion contained in the observed indicators rather than the agronomic importance of each hazard. Similar entropy-based approaches have been used to construct composite climate- and agricultural-risk measures (Gu et al., 2023; Bezdan et al., 2025). Because the dependent variable is annual total grain output covering multiple crops with heterogeneous planting calendars across Chinese regions, the baseline climate indicators are measured at the annual level rather than imposing a single common growing-season window. The index should therefore be interpreted as an annual city-level multi-hazard exposure measure and does not capture crop-specific phenological exposure or crop-specific temperature and moisture thresholds.

4.2.3 Control variables

The baseline model controls for several city characteristics that may be associated with both agricultural production capacity and local responses to climate shocks. Economic development (lnpgdp), measured by the logarithm of per capita GDP, captures differences in local economic resources and investment capacity. Urbanization (urban), measured by the share of urban population in total resident population, reflects changes in land use, labor allocation, and the relative importance of agriculture in the local economy. Industrial structure (industry), measured by the share of secondary and tertiary industries in GDP, controls for differences in the sectoral composition of local economic activity. Population density (popdens), measured by resident population per administrative land area, captures variation in population pressure and land-resource constraints. Transport infrastructure (transport), measured by road area per capita, reflects differences in infrastructure conditions that may affect agricultural production and the movement of inputs and outputs. Technological innovation (innovation), measured by science and technology expenditure as a share of general public budget expenditure, captures differences in local technological and innovation capacity. Including these variables helps reduce potential omitted-variable bias arising from observable socioeconomic differences across cities.

4.2.4 Mechanism variables

Mechanism variables include agricultural mechanization (machine), measured by the logarithm of total agricultural machinery power; effective irrigation (irrig), measured by the ratio of effective irrigated area to cultivated land; grain-sown area (sown), measured by the logarithm of grain-sown area; and fiscal support for agriculture (fiscal), measured by agriculture, forestry, and water affairs expenditure as a share of general public budget expenditure. These variables correspond to the production channels developed in the theoretical analysis and are used to examine how climate risk may affect grain output through agricultural production conditions and public support capacity.

The city-level socioeconomic and agricultural indicators used in the empirical analysis are observable proxies rather than comprehensive measures of the underlying concepts. Their selection reflects both their economic relevance to agricultural production and their relatively consistent availability across prefecture-level cities. More specifically, road area per capita captures one observable dimension of physical road infrastructure and should not be interpreted as a complete measure of logistics quality or market connectivity. Science and technology expenditure as a share of general public budget expenditure is interpreted as a proxy for local public science-and-technology input rather than technological innovation in its entirety. Total agricultural machinery power captures the scale of mechanized input capacity, although it does not measure machinery utilization efficiency or equipment quality; machinery power and machinery power per unit of land are commonly used in empirical studies of agricultural mechanization in China (Ma and Huang, 2023). Similarly, the share of agriculture, forestry, and water-affairs expenditure in the general public budget represents broad agriculture-related fiscal support rather than expenditure devoted exclusively to grain production, consistent with related empirical work on fiscal agricultural expenditure in China (Zhang and Zhang, 2024). Accordingly, coefficients on these variables are interpreted in relation to the specific dimensions captured by the proxies rather than as effects of the broader concepts in their entirety.

Variable definitions are reported in Table 1.

Table 1. Variable definitions.

4.3 Data description

This study constructs a balanced city-year panel for Chinese prefecture-level cities from 2010 to 2023. The socioeconomic and agricultural data are compiled mainly from the China City Statistical Yearbook, the China Regional Economic Statistical Yearbook, provincial and municipal statistical yearbooks, and statistical communiqués on national economic and social development. Because prefecture-level statistics are reported through multiple sources, the data were cross-checked by city and year when information from different sources was combined, with priority given to maintaining consistent statistical definitions over time.

The sample was screened according to data availability and temporal comparability. In particular, administrative-boundary adjustments and changes in statistical definitions were checked when constructing consistent city identifiers over time, while observations that could not be consistently reconciled across years because of changes in administrative units or statistical coverage were not retained. Cities for which the variables required for the baseline analysis or corresponding meteorological information could not be consistently matched over the study period were also excluded from the final panel. After these procedures, 298 prefecture-level cities are retained for all 14 years from 2010 to 2023, yielding a balanced panel of 4,172 city-year observations.

Climate-risk data are derived from city-level meteorological observations. Meteorological records were matched to the corresponding prefecture-level city-year observations before the climate indicators were constructed. City-years for which the meteorological information could not be consistently matched were not included in the final balanced sample. Extreme high temperature, extreme low temperature, extreme rainfall, and drought indicators are then standardized and aggregated using the entropy-weighting method to construct the city-year climate-risk index.

Table 2 reports descriptive statistics for the main variables. The mean of FoodSec is 5.842, with a standard deviation of 0.913, a minimum of 2.114, and a maximum of 8.436. Because FoodSec is measured as the logarithm of total grain output, these statistics indicate substantial differences in production capacity across cities. Cities with stronger agricultural foundations and more abundant cultivated land tend to have higher grain output, while highly urbanized or resource-constrained cities tend to have lower output.

Table 2. Descriptive statistics.

The mean climate-risk index is 0.286, with a standard deviation of 0.174, a minimum of 0.018, and a maximum of 0.921. This indicates meaningful variation in climate-risk exposure across cities and years, which is necessary for identifying the effect of climate risk on food security. Control variables also display considerable variation, including economic development, urbanization, industrial structure, fiscal support, mechanization, and effective irrigation. These differences justify the inclusion of control variables and city and year fixed effects in the subsequent regressions.

5 Empirical result analysis

5.1 Baseline regression results

Table 3 reports the baseline regression results. Column (1) includes only the climate-risk index and city and year fixed effects. Column (2) further controls for economic development, urbanization, industrial structure, and population density. Column (3), the preferred specification, additionally includes transport infrastructure and technological innovation.

Table 3. Baseline regression results: climate risk and production-side food security.

Values in parentheses are t-statistics; *, , and * denote significance at the 10%, 5%, and 1% levels, respectively.

In Column (1), the coefficient on climate risk is −0.082 and significant at the 1% level. Since the dependent variable is the logarithm of total grain output, this coefficient implies that a 0.1-unit increase in the climate-risk index is associated with an approximately 0.82% decline in city-level grain output, holding other factors constant. In Column (2), after controlling for socioeconomic characteristics, the coefficient remains significantly negative at −0.067. In Column (3), after adding transport infrastructure and technological innovation, the coefficient remains significant at −0.060, implying a decline of approximately 0.60% in grain output for every 0.1-unit increase in climate risk.Given that the standard deviation of ClimateRisk is 0.174, a one-standard-deviation increase in climate risk is associated with an approximately 1.04% decline in city-level grain output in the preferred specification. This provides a more intuitive indication of the economic magnitude of the estimated effect.

Overall, the baseline results support Hypothesis H1: climate-risk shocks significantly weaken production-side food security.

5.2 Endogeneity test

This study addresses potential endogeneity using several approaches. First, two instrumental variables are constructed: the interaction between historical city-level climate-risk exposure and national annual extreme-climate intensity, and the interaction between historical extreme-rainfall exposure and national annual precipitation anomalies. Historical exposure captures long-term climate sensitivity across cities, while national annual climate anomalies provide time-varying exogenous shocks. Their interactions generate city-year variation in climate risk that is plausibly exogenous to contemporaneous city-level grain output. The exclusion restriction rests on the predetermined nature of historical climate exposure and the common national character of the annual climate shocks. Conditional on city fixed effects, year fixed effects, and the observed city-level controls, these interaction terms capture differences in how cities are exposed to common national climate shocks rather than contemporaneous variation in local economic conditions, infrastructure, or policy. Historical exposure cannot respond to current grain-production outcomes, while national annual climate anomalies are not determined by conditions in any individual city. Therefore, the interaction terms are expected to affect current city-level grain output primarily through their effect on contemporaneous local climate risk.

The first-stage coefficients on both instruments are significantly positive. The Kleibergen-Paap rk LM statistic is 42.518 and significant at the 1% level, indicating no under-identification problem. The Kleibergen-Paap rk Wald F statistic is 31.742, exceeding both the Stock-Yogo 10% critical value of 19.93 and the conventional threshold of 10, suggesting that the instruments are not weak. The Hansen J statistic is 1.284 with a p-value of 0.257, so the over-identifying restrictions are not rejected.

In the second stage, the coefficient on climate risk is −0.063 and significant at the 5% level, indicating that climate risk continues to reduce production-side food security after accounting for potential omitted variables, measurement error, and reverse causality. Additional tests using lagged climate risk, system GMM, and exclusion of concurrent shocks also yield significantly negative coefficients. Taken together, these supplementary estimates show that the negative relationship between climate risk and production-side food security remains stable across alternative specifications and endogeneity-oriented checks. In particular, the system GMM estimate remains negative and statistically significant, with a coefficient of −0.057, which is broadly consistent with the baseline two-way fixed-effects estimate. The system GMM specification is used as a supplementary sensitivity analysis rather than as the primary identification strategy. Following the standard dynamic-panel framework, specification validity should be assessed jointly with the lag structure, instrument count, Arellano–Bond serial-correlation tests, and Hansen overidentification test (Arellano and Bond, 1991; Blundell and Bond, 1998; Roodman, 2009). Accordingly, these estimates are interpreted as complementary evidence on the stability of the baseline relationship rather than as independent proof of causal identification (Table 4).

Table 4. Endogeneity tests, IV validity, and identification tests.

5.3 Robustness tests

Before presenting the alternative-specification robustness results, we first examine whether the panel exhibits serial correlation and cross-sectional dependence. These diagnostics are important because climate shocks may persist over time and may be geographically correlated across cities. Serial correlation is assessed using the Wooldridge-type test for panel data (Drukker, 2003), while cross-sectional dependence is examined using Pesaran's CD test (Pesaran, 2015). The diagnostic results are reported in Table 5.

Table 5. Panel-dependence diagnostic tests.

Table 6 presents robustness checks based on alternative variable measurement and sample treatment. Column (1) reconstructs the climate-risk index using principal component analysis; the coefficient remains significantly negative. Columns (2) and (3) replace the dependent variable with per capita grain output and grain yield per unit area, respectively, and the coefficients on climate risk remain negative and significant. Column (4) uses output volatility as the dependent variable; the positive coefficient indicates that climate risk increases production volatility and weakens the stability dimension of food security. Column (5) winsorizes the main continuous variables, and the coefficient remains negative and significant. These tests confirm that the baseline conclusion is not driven by a particular index-construction method, dependent-variable definition, or outliers. This alternative construction is particularly relevant because entropy weights are statistical weights based on indicator dispersion rather than agronomic weights; the PCA result therefore reduces concern that the baseline finding is mechanically driven by the entropy-weighting scheme.

Table 6. Robustness checks: alternative variables and sample treatment.

5.4 Mediating effect analysis

Table 7 examines the mechanization channel. ClimateRisk significantly reduces agricultural mechanization, while mechanization is positively associated with grain output. After mechanization is included, the coefficient on ClimateRisk declines from −0.060 to −0.041 and remains statistically significant. These results are consistent with agricultural mechanization serving as a partial transmission channel through which climate risk affects production-side food security.

Table 7. Mechanism analysis I: agricultural mechanization.

Table 8 examines the effective-irrigation channel. ClimateRisk has a significantly negative effect on effective irrigation, while better irrigation conditions are positively associated with grain output. After effective irrigation is introduced, the coefficient on ClimateRisk declines from −0.060 to −0.036 and remains significant, providing evidence that irrigation conditions account for part of the adverse effect of climate risk on grain production.

Table 8. Mechanism analysis II: effective irrigation.

Table 9 examines the grain-sown-area channel. ClimateRisk significantly reduces grain-sown area, and a larger sown area is associated with higher grain output. After grain-sown area is included, the coefficient on ClimateRisk declines from −0.060 to −0.044, indicating that changes in cultivation scale constitute another partial transmission channel.

Table 9. Mechanism analysis III: grain-sown area.

Table 10 examines the fiscal-support channel. ClimateRisk negatively affects fiscal support for agriculture, while fiscal support is positively associated with grain output, although its coefficient is statistically weaker than those of the other mechanism variables. After fiscal support is included, the coefficient on ClimateRisk declines from −0.060 to −0.032 and remains significant, suggesting that public support capacity also accounts for part of the relationship between climate risk and grain production.

Table 10. Mechanism analysis IV: fiscal support for agriculture.

Taken together, the four mechanism analyses provide consistent evidence that climate risk affects grain output through agricultural production conditions and public support capacity. A comparison across the four specifications also shows some differences in the degree of coefficient attenuation. Relative to the baseline coefficient of −0.060, the ClimateRisk coefficient declines to −0.041 after mechanization is included, −0.036 after effective irrigation is included, −0.044 after grain-sown area is included, and −0.032 after fiscal support is included. Thus, the attenuation is descriptively larger in the effective-irrigation and fiscal-support specifications. This comparison should not be interpreted as a formal ranking of the causal importance of the four mechanisms, but it suggests that water-security conditions and public support capacity may be particularly relevant in explaining how climate risk is transmitted into grain-output losses. Overall, the results support H2–H5.

5.5 Heterogeneity analysis

For the heterogeneity analysis, cities are first classified according to their grain-production function. Following the national classification of major grain-producing areas, cities located in Hebei, Inner Mongolia, Liaoning, Jilin, Heilongjiang, Jiangsu, Anhui, Jiangxi, Shandong, Henan, Hubei, Hunan, and Sichuan are classified as major grain-producing areas, while cities in the remaining provinces are classified as non-major grain-producing areas.

Adaptive capacity is measured using a composite index constructed from agricultural mechanization, effective irrigation, fiscal support for agriculture, and technological innovation. These indicators capture, respectively, agricultural operational capacity, water-supply security, public support capacity, and technological capacity to respond to climate shocks. After standardization, the indicators are aggregated using the entropy-weighting method. The sample is then divided according to the median value of the adaptive-capacity index: observations above the median are classified as the high-adaptive-capacity group, while those at or below the median are classified as the low-adaptive-capacity group.

Table 11 reports heterogeneity results for the effect of climate risk on production-side food security. The study conducts grouped regressions by major versus non-major grain-producing areas and by regions with high versus low adaptive capacity, to examine whether climate risk affects different types of cities differently. Overall, the coefficients on ClimateRisk are negative and significant in all four columns, indicating that rising climate risk significantly reduces grain output across different city types. However, the magnitudes and significance levels differ across groups, suggesting that the effect is not homogeneous but is shaped by regional agricultural foundations, policy support, and adaptive capacity.

Table 11. Heterogeneity analysis.

Column (1) shows that in major grain-producing areas, the coefficient on ClimateRisk is −0.038 and significant at the 5% level. This indicates that rising climate risk significantly suppresses production-side food security in these areas. Because the dependent variable is the logarithm of total grain output, the coefficient implies that, holding other factors constant, a 0.1-unit increase in ClimateRisk reduces total grain output in major grain-producing cities by approximately 0.38%. Although the effect is significantly negative, the absolute value is smaller than in non-major producing areas, indicating that major producing areas are relatively less affected. The reason is that these areas usually undertake national grain-supply functions and therefore have stronger agricultural infrastructure, farmland water-conservancy conditions, grain-production subsidies, agricultural insurance, and policy support. Even when exposed to extreme heat, drought, heavy rainfall, and flooding, they can partly buffer negative impacts through stronger production foundations and policy safeguards.

Column (2) shows that in non-major grain-producing areas, the coefficient on ClimateRisk is −0.071 and significant at the 1% level. Its absolute value is clearly larger than that in major producing areas, indicating a stronger negative effect. Economically, a 0.1-unit increase in ClimateRisk reduces total grain output in non-major producing cities by approximately 0.71%. This result suggests that non-major producing areas are more vulnerable to climate risk. Grain production in these areas generally has a weaker functional position in the local economy, and agricultural infrastructure, water-conservancy construction, mechanization investment, and fiscal support may be less stable than in major producing areas. Moreover, these areas are more likely to be affected by industrialization, urbanization, and nonagricultural expansion, which draw land, labor, and fiscal resources away from agriculture. Thus, when climate risk rises, grain output in non-major producing areas is more likely to decline.

Column (3) shows that in high-adaptive-capacity regions, the coefficient on ClimateRisk is −0.049 and significant at the 5% level. This indicates that even in regions with stronger climate-adaptation capacity, climate risk still significantly reduces grain output. Economically, a 0.1-unit increase in ClimateRisk reduces total grain output by approximately 0.49%. The result suggests that adaptation capacity can mitigate but cannot fully offset the adverse effects of extreme weather. High-capacity regions usually have higher mechanization, more complete irrigation, stronger fiscal support, and better infrastructure; they can reduce losses by improving rapid sowing and harvesting, water supply, disaster early warning, and recovery capacity. However, because climate risk is sudden and uncertain, extreme weather can still adversely affect crop growth, field management, and harvesting.

Column (4) shows that in low-adaptive-capacity regions, the coefficient on ClimateRisk is −0.066 and significant at the 1% level. Its absolute value is larger than that for high-capacity regions, indicating deeper exposure to climate-risk shocks. Economically, a 0.1-unit increase in ClimateRisk reduces total grain output in low-capacity regions by approximately 0.66%. This means that when agricultural infrastructure is weak, mechanization is low, irrigation security is inadequate, or fiscal support is limited, extreme heat, drought, heavy rainfall, and flooding are more likely to translate into actual production losses. Low-capacity regions are not only more vulnerable to climate shocks but also have weaker post-disaster recovery capacity, making the negative effect on food security more pronounced.

Comparing the four columns, the absolute values of the coefficients are larger in non-major producing areas and low-adaptive-capacity regions. The coefficient is −0.071 for non-major producing areas versus −0.038 for major producing areas, and −0.066 for low-capacity regions versus −0.049 for high-capacity regions. This indicates that regional agricultural foundations and adaptive capacity significantly moderate the effect of climate risk on food security. Major producing areas and high-capacity regions are also affected by climate risk, but their more complete production systems, better infrastructure, and more stable policy support help mitigate adverse effects. By contrast, non-major producing areas and low-capacity regions have relatively weaker agricultural foundations and are more likely to experience sowing constraints, insufficient irrigation, harvest delays, and recovery difficulties, leading to larger declines in output.

The fundamental reason for the results in Table 11 is that regions differ substantially in grain-production functions, agricultural infrastructure, fiscal-support capacity, and climate-adaptation capacity. Major grain-producing areas usually have stronger production foundations and policy safeguards, whereas non-major producing areas and low-adaptive-capacity regions have weaker conditions and are therefore more vulnerable to climate shocks.

5.6 Supplementary analysis: agricultural performance, production pressure, and food-supply stability

The analyses in this section are supplementary to the main empirical framework rather than extensions of the causal mechanism analysis. The outcomes considered here do not represent a single analytical level or a sequential causal pathway. Instead, they are grouped into three related dimensions: agricultural economic performance, production and fiscal pressure, and food-supply stability. The first dimension includes gross agricultural output, agricultural value added, rural residents’ income, and agricultural labor productivity; the second includes agricultural production costs and local fiscal disaster-relief pressure; and the third includes grain-price volatility and the local grain supply gap. These outcomes are examined to provide a broader description of the economic conditions associated with climate risk, while the core empirical analysis remains focused on city-level grain-production capacity. Because these supplementary regressions rely primarily on fixed-effects specifications, the estimates are interpreted as conditional associations rather than as independently identified causal effects.

Table 12 first examines whether climate risk is associated with broader measures of agricultural economic performance. The four outcomes capture different aspects of the local agricultural economy and are not treated as components of a single causal mechanism.

Table 12. Supplementary analysis I: climate risk and agricultural economic performance.

In Column (1), the coefficient is −0.052 and significant at the 1% level, indicating that higher climate risk is significantly associated with lower gross agricultural output. Holding other factors constant, a 0.1-unit increase in ClimateRisk is associated with an approximately 0.52% lower level of gross agricultural output. This pattern is consistent with climate risks such as extreme heat, drought, heavy rainfall, and flooding being related to weaker agricultural production conditions and lower overall output capacity.

Column (2), using agricultural value added as the dependent variable, shows a coefficient of −0.047 significant at the 5% level, indicating that higher climate risk is associated with lower agricultural value added. Economically, a 0.1-unit increase in ClimateRisk is associated with an approximately 0.47% lower level of agricultural value added. This association suggests that climate risk is related not only to grain production but also to the wider economic performance of the agricultural sector. In Column (3), the coefficient for rural residents' income is significantly negative at −0.034, indicating that higher climate risk is associated with lower rural income. A 0.1-unit increase in ClimateRisk is associated with an approximately 0.34% lower level of rural residents' income.

Column (4), using agricultural labor productivity as the dependent variable, reports a coefficient of −0.059 significant at the 1% level, indicating that higher climate risk is associated with lower agricultural labor productivity. Economically, a 0.1-unit increase in ClimateRisk is associated with an approximately 0.59% lower level of agricultural labor productivity. This pattern is consistent with extreme weather being related to greater field-management difficulty, lower operational efficiency, and additional labor requirements for replanting, drainage, disaster prevention, and post-disaster recovery.

Table 13 turns to a different set of supplementary outcomes related to production pressure and food-supply stability. These variables capture distinct responses at the producer, local-government, and market levels and are therefore not interpreted as belonging to a single causal chain. Agricultural production costs reflect pressure on farm production, fiscal disaster-relief expenditure captures the budgetary burden associated with agricultural disaster response, while grain-price volatility and the grain supply gap provide indicators of local food-supply instability.

Table 13. Supplementary analysis II: production pressure and food-supply stability.

Column (1) uses agricultural production cost as the dependent variable, measured by agricultural production cost per unit of sown area. The coefficient on ClimateRisk is 0.043 and significant at the 1% level, indicating that higher climate risk is associated with higher agricultural production costs. Economically, a 0.1-unit increase in ClimateRisk is associated with an approximately 0.43% higher level of agricultural production costs. This association is consistent with greater demand for irrigation, drainage, replanting, pesticides, fertilizers, labor management, and post-disaster recovery under extreme weather.

Column (2) uses fiscal disaster-relief pressure as the dependent variable, measured by the share of agricultural disaster-relief expenditure in local general public budget expenditure. The coefficient on ClimateRisk is 0.051 and significant at the 1% level, suggesting that higher climate risk is associated with greater local fiscal pressure for agricultural disaster prevention, mitigation, and recovery. A 0.1-unit increase in ClimateRisk is associated with an approximately 0.51% higher level of fiscal disaster-relief pressure.

Column (3) uses grain price volatility as the dependent variable, measured by the volatility of the growth rate of the grain price index. The coefficient on ClimateRisk is 0.027 and significant at the 5% level, indicating that higher climate risk is associated with greater grain-price volatility. When climate shocks reduce output or destabilize supply expectations, they may amplify grain-price volatility through supply-side contraction and market-expectation changes.

Column (4) uses the grain supply gap as the dependent variable, measured as the ratio of the difference between grain demand and grain output to grain demand, namely: grain supply gap = (grain demand - total grain output) / grain demand. The coefficient on ClimateRisk is 0.036 and significant at the 5% level, suggesting that higher climate risk is associated with a larger gap between local urban grain supply and demand. This finding is consistent with the earlier evidence on production-side food security and indicates a close relationship between climate risk and local food-supply pressure.

Taken together, Tables 12 and 13 provide supplementary evidence that climate risk is associated with outcomes at several distinct levels of the local agricultural economy. At the agricultural-performance level, higher climate risk is associated with lower agricultural output, value added, rural income, and labor productivity. At the production and fiscal level, it is associated with higher production costs and greater disaster-relief pressure. At the food-supply level, it is associated with greater grain-price volatility and a larger local grain supply gap. These outcomes are not interpreted as elements of a single causal chain. Rather, they provide complementary evidence on the broader conditions that accompany climate risk beyond the paper’s primary outcome of city-level grain production. Because these supplementary regressions rely primarily on fixed-effects specifications, the estimates are interpreted as conditional associations rather than as separately identified causal effects.

6 Conclusions and policy recommendations

6.1 Conclusion

Using panel data for 298 prefecture-level cities in China from 2010 to 2023, this study examines how climate risk affects production-side food security, through which channels this effect occurs, and whether it differs across cities with different production functions and adaptive capacity. Several main findings emerge.

First, climate risk significantly weakens production-side food security by reducing city-level grain output. This relationship remains robust across alternative specifications and treatments of potential endogeneity, indicating that greater exposure to climate shocks places persistent pressure on local grain-supply capacity.

Second, the effect of climate risk is partly transmitted through agricultural production conditions and public support capacity. The mechanism analysis shows that climate risk adversely affects agricultural mechanization, effective irrigation, grain-sown area, and fiscal support for agriculture, while these factors are positively associated with grain production. This suggests that climate shocks affect food availability not only through direct production losses but also by weakening the conditions needed to maintain stable agricultural production. These findings are consistent with the mechanism results reported in the empirical analysis.

Third, the effect of climate risk is heterogeneous across cities. The negative impact is stronger in non-major grain-producing areas and in regions with lower adaptive capacity. This indicates that local agricultural foundations and adaptive capacity shape the extent to which climate shocks are translated into grain-output losses. Cities with weaker production conditions and support capacity are therefore more vulnerable to climate risk.

The supplementary analysis further shows that climate risk is associated with outcomes beyond grain production. Higher climate risk is associated with weaker agricultural economic performance, including lower gross agricultural output, agricultural value added, rural income, and labor productivity. It is also associated with higher production costs and fiscal disaster-relief pressure, as well as greater grain-price volatility and a larger local grain supply gap. These findings should be interpreted as complementary associations across different dimensions of the local agricultural economy rather than as components of a single causal pathway.

Overall, the findings show that safeguarding production-side food security under increasing climate risk requires attention to both direct climate exposure and the capacity of local agricultural systems to absorb and recover from shocks. Strengthening agricultural production conditions and adaptive capacity is therefore important for maintaining food availability and, more broadly, for supporting the basic conditions of nutrition security.

6.2 Policy recommendations

The empirical findings suggest several policy priorities for strengthening production-side food security under climate risk. First, the baseline results show that greater climate risk places persistent pressure on city-level grain-supply capacity. This finding supports continued investment in agricultural climate monitoring and early-warning systems. Local governments should strengthen coordination among meteorological, agricultural, and emergency-management agencies and provide timely information that enables producers to adjust planting, irrigation, field management, and harvesting in response to extreme-weather risks.

Second, the mechanism results show that agricultural mechanization and effective irrigation are important channels linking climate risk to grain production. Policy should therefore place greater emphasis on strengthening farmland water conservancy, water-saving irrigation, drainage facilities, water storage, high-standard farmland, and agricultural machinery services. In particular, improving irrigation and drainage capacity can reduce production losses associated with drought, excessive rainfall, and flooding, while better access to machinery services can improve the timeliness of sowing, harvesting, and post-disaster recovery.

Third, the mechanism and heterogeneity results indicate that fiscal support and local adaptive capacity are important for buffering climate-related production risks. Fiscal resources should therefore be more strongly targeted toward regions with weaker agricultural infrastructure and lower adaptive capacity, particularly non-major grain-producing areas and low-adaptive-capacity cities. Public expenditure can prioritize irrigation and drainage infrastructure, disaster prevention, technology extension, post-disaster recovery, and agricultural insurance, with greater support directed toward areas in which climate shocks are more likely to translate into grain-output losses.

Fourth, the further analysis shows that higher climate risk is associated with greater grain-price volatility and a larger grain supply gap. This suggests that production-side adaptation should be complemented by measures to improve food-supply stabilization. Local and regional governments can strengthen grain reserve and emergency-release mechanisms, improve production-marketing coordination across regions, and enhance information sharing on grain production, inventories, prices, and demand. These measures can help reduce the extent to which local production shocks translate into short-term supply shortages and market instability.

Overall, the policy response to climate risk should combine ex ante risk prevention, improvements in agricultural production conditions, targeted support for vulnerable regions, and stronger grain-supply stabilization mechanisms. Such a differentiated approach is more closely aligned with the empirical evidence on mechanisms, regional heterogeneity, and food-supply stability identified in this study.

6.3 Limitations and future research

This study has several limitations that also suggest directions for future research. First, the analysis focuses primarily on production-side food security and uses city-level grain output to measure local grain-supply capacity. This approach is appropriate for examining food availability from the production side, but it does not directly capture household food access, dietary quality, or actual nutrition outcomes. As the manuscript itself notes, local grain production should not be equated with residents' final nutrient intake.

Future research could therefore combine production data with household-level information on food consumption, food expenditure, dietary diversity, and nutritional outcomes to examine how climate-related production shocks are transmitted from food supply to household welfare.

Second, city-level analysis inevitably abstracts from some within-city heterogeneity in agricultural conditions and exposure to extreme weather. Future work could use finer spatial data, crop-specific information, or farm- and household-level observations to identify more precisely which producers and population groups are most vulnerable to climate shocks.

The measurement strategy also has limitations. The annual climate-risk index does not distinguish crop-specific growing seasons, phenological stages, or region-specific agronomic thresholds, and the prefecture-level proxy variables do not fully capture the quality or utilization efficiency of infrastructure, technology, machinery, and public support. Future research could combine crop-specific planting calendars, finer spatial weather data, and more detailed indicators of agricultural technology, infrastructure quality, machinery use, and grain-specific fiscal support to provide more precise measures.

Inference also requires caution because climate shocks can be spatially correlated across neighboring cities and agricultural outcomes may exhibit serial persistence. The additional dependence-oriented checks therefore complement, but do not replace, the identifying assumptions of the fixed-effects and instrumental-variable specifications.

Third, this study concentrates on several major production-side channels, including mechanization, irrigation, sown area, and fiscal support. Future research could extend the analysis to other adaptation mechanisms, such as agricultural insurance, crop switching, technological adoption, interregional grain trade, storage, and food-market adjustment, thereby providing a more complete account of how climate risk is transmitted through the broader food system.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mingjun Hu, Wenzhou University, China

Reviewed by: Jian Hu, Lanzhou University of Finance and Economics, China

Wei Liang, Shanxi University of Finance and Economics, China