Skip to Content
SustainabilitySustainability
  • Article
  • Open Access

11 June 2026

17 Pages

Crop-Specific Weather–Yield Associations in Irrigation-Intensive Oasis Agriculture: Evidence from Cotton and Maize in Xinjiang, China

,
,
,
,
and
1
Institute of Agricultural Economics and Information, Xinjiang Academy of Agricultural Sciences, Urumqi 830091, China
2
Agricultural Information Institute, Chinese Academy of Agricultural Sciences, Beijing 100081, China
3
Key Laboratory of Agricultural Blockchain Application, Ministry of Agriculture and Rural Affairs, Beijing 100081, China
*
Author to whom correspondence should be addressed.
This article belongs to the Section Sustainable Agriculture

Abstract

Weather–yield relationships in arid agricultural regions are shaped jointly by temperature and precipitation exposure, irrigation conditions, crop choice, and management under water constraints. This study combines county-level cotton yield and maize grain-yield data for Xinjiang, China, from 2000 to 2020 with daily meteorological station records assigned to county-level weather exposures. We estimate two-way fixed-effects models that include temperature degree-day indicators and a quadratic precipitation term to examine crop-specific weather–yield associations in irrigation-intensive oasis agriculture. The baseline two-way fixed-effects estimates indicate that a 100 °C d increase in growing degree days is associated with a 2.85% increase in cotton yield and a 1.88% decrease in maize yield. For cotton, the baseline and common-trend specifications indicate a convex precipitation–yield association, with an estimated turning point of 141.07 mm (95% CI: 27.75–225.63 mm), while the pattern is less stable under prefecture-by-year fixed effects. Maize yield is more consistently negatively associated with growing-season heat accumulation. Post-2010 interaction terms indicate crop-differentiated changes in heat sensitivity, consistent with different temporal evolution of weather–yield associations across cotton and maize. Overall, the results show that climate-risk assessment in irrigation-intensive arid agriculture should distinguish between crop types, precipitation regimes, and the management conditions under which weather exposure is translated into yield outcomes.

1. Introduction

Climate variability and climate change affect agricultural production through changes in mean temperature, extreme temperature exposure, precipitation regimes, and the frequency of damaging weather events. A large body of crop-climate research shows that yield responses to temperature are often nonlinear: warming may benefit crop growth within a limited suitable range, whereas temperatures above crop-specific thresholds tend to reduce yields [1,2,3]. Interannual climate variability, extreme weather disasters, and warming have all been linked to crop losses or yield fluctuations, with particularly clear evidence for yield reductions associated with drought and extreme heat [4,5,6]. These risks matter for food security, farm income, and regional economic resilience, especially in regions where water availability and ecological fragility already constrain agricultural production.
The economic and agronomic meanings of weather variables are embedded in specific production systems. In rain-fed agriculture, growing-season precipitation is usually interpreted primarily as water input. In irrigation-intensive systems, however, natural rainfall interacts with artificial water supply, drainage capacity, field humidity, disease pressure, and the timing of field operations. Previous studies show that irrigation practices can modify near-surface climate and regional hydrothermal conditions [7,8], and that irrigation can affect the yield implications of heat exposure through evaporative cooling [9,10]. In cotton fields in arid Northwest China, drip irrigation under plastic film and mulching practices also alter soil hydrothermal conditions and affect cotton yield [11,12]. Canopy temperature is closely linked to water availability [13,14], and regional modeling further indicates that irrigation management can alter the spatial distribution of extreme temperature risk [15]. The same precipitation or temperature variable can therefore represent different agronomic conditions across production environments.
Existing empirical studies of climate and crop yields commonly use interannual temperature and precipitation variation in spatial panels to identify weather effects [16,17,18], and global crop studies also show that temperature and precipitation exposure can explain yield variation [19]. This approach is influential because short-run weather fluctuations are usually strongly exogenous to farmers’ contemporaneous decisions and can be matched directly to observed output. At the same time, weather variables often capture the net effect of multiple latent channels operating together [20]. Panel estimates primarily identify empirical associations between weather variation and yield. Attributing these associations to adaptation, technology diffusion, or policy channels requires additional mechanism identification [21]. Moreover, the matching and scale of meteorological data and climate model outputs also affect the economic interpretation of weather exposure indicators [22].
Arid oasis agriculture provides a particularly important setting for examining these issues. Crop production in such systems depends on stable water supply and active regulation of field heat and moisture under severe climatic constraints. Crop-specific climate sensitivity is also shaped by differences in productivity levels and management conditions [23]. In Xinjiang cotton production, temperature, disease, and agronomic management are closely interrelated [24], and regional evidence also shows that cotton yield variation is influenced by both climate and management factors [25]. These features make Xinjiang oasis agriculture a production context with independent explanatory value.
The crop dimension is equally important. Cotton and maize play different roles in Xinjiang’s agricultural system. Cotton is a high-value cash crop and is usually more closely associated with intensive field management and irrigation investment. Maize is a staple grain crop and faces different economic incentives and production constraints. Technology adoption, learning, and input adjustment are often related to expected returns, comparative advantage, and information diffusion [26,27,28]. Xinjiang agriculture is also embedded in an arid-region water allocation system, where irrigation infrastructure, basin-level water allocation, and crop management are closely tied to production sustainability [29]. These differences can cause cotton and maize to exhibit different weather–yield associations under the same county-year climate environment.
How temperature extremes and precipitation are associated with yields of different crops within irrigation-intensive oasis systems is a central question for sustainable agricultural management in arid regions. This study addresses three core questions. First, do cotton and maize exhibit different weather–yield associations within the same regional production environment? Second, is the precipitation–yield relationship for cotton linear, or does it exhibit a nonlinear pattern in which the marginal association changes over the observed range? Third, do estimated weather sensitivities change over time, and what do those changes suggest about crop-specific adjustment margins under evolving production conditions? These questions also connect to compound meteorological exposure and agricultural climate-risk assessment [30]. Answering them can help identify the climate risks that deserve priority attention in sustainable crop management in water-constrained regions.
This study uses county-level cotton yield and maize grain-yield data for Xinjiang from 2000 to 2020, matched with daily meteorological station records assigned to county-level weather exposures. We construct growing-season growing degree days, extreme heat degree days, harmful cold degree days, and precipitation indicators, and estimate crop-specific county-year fixed-effects panel models. The analysis focuses on county-scale weather–yield associations and interprets these associations within the production context of irrigation-intensive oasis agriculture. This positioning is consistent with contextualized interpretations in studies of irrigation efficiency and blue-green water accounting [31,32]. Research on water productivity and irrigation management in dry areas also shows that the yield implications of water management often depend on crop and production environment [33,34]. In other words, this study identifies systematic empirical patterns in crop sensitivity and uses them to define the management margins that should be prioritized in arid-region agricultural climate-risk research.
The study makes three main contributions. First, it compares a cash crop and a grain crop within the same arid oasis production system, showing that crop type affects how climate variation maps into yield outcomes. This contribution speaks to the literature on heterogeneity in climate impacts [23]. Second, it estimates precipitation nonlinearity and its turning point within a county panel framework, providing evidence on how the precipitation–yield relationship in managed oasis agriculture differs from that in rain-fed systems [35,36]. Third, it documents temporal changes in estimated weather sensitivity and interprets them as reduced-form evidence consistent with crop-differentiated adjustment [20,21]. Overall, the study provides context-specific evidence for climate-risk monitoring and sustainable crop management in irrigation-intensive arid agriculture.

2. Materials and Methods

2.1. Data Sources and Sample Construction

2.1.1. Study Area and Crop Selection

Xinjiang is a typical arid and semi-arid oasis agricultural region in northwestern China. Crop production is concentrated in irrigated oases, where agricultural output depends on the joint regulation of heat, water supply, field moisture, and management intensity. Cotton and maize were selected because they represent two major crop types in this production system: cotton is a high-value cash crop with intensive irrigation and field management, while maize is an important grain crop with different economic incentives and heat-risk exposure. Comparing the two crops within the same regional weather environment helps identify crop-specific weather–yield relationships rather than treating oasis agriculture as a single homogeneous production system.

2.1.2. Yield Data and Cropping System

County-level yield data for cotton and maize were obtained from the Xinjiang Statistical Yearbook and related agricultural statistical materials. Maize yield refers to maize grain yield. Yield is measured as output per unit area (t/ha). To reduce scale differences and facilitate coefficient interpretation, the baseline regression uses the natural logarithm of yield as the dependent variable:
y i t = ln y i e l d i t
where i denotes county and t denotes year. To ensure comparability, all estimations are conducted separately by crop. After excluding county-year observations with missing yield or weather variables used in the regressions, the final samples are unbalanced panels with 1203 cotton observations from 78 counties and 1339 maize observations from 96 counties; observations and county clusters are reported in the regression tables.

2.1.3. Meteorological Data

Meteorological variables were constructed from daily temperature and precipitation station records. The station records were matched to counties using station locations, administrative codes, and regional affiliation, and were then aggregated to the county-day level before growing-season aggregation. When multiple stations reported observations for the same county and day, records were averaged at the county-day level. Daily records with invalid or missing temperature or precipitation fields were treated as missing before aggregation, and county-year observations lacking the variables required for a given regression specification were excluded from that specification. The analysis was implemented by the authors using Stata 18.0 MP.

2.2. Construction of Meteorological Variables

2.2.1. Temperature Variables

Following the degree-day approach commonly used in crop-climate research, and incorporating extreme degree-day measures, this study constructs three temperature indicators to capture nonlinear and threshold effects [1,3]. Let T m e a n , d , T m a x , d , and T m i n , d denote daily mean temperature, daily maximum temperature, and daily minimum temperature on day d, respectively. The crop-specific thresholds are 12 °C and 32 °C for cotton and 10 °C and 30 °C for maize. Empirical economic studies typically link crop yields to the full temperature distribution to avoid masking extreme risks with average indicators such as annual mean temperature [1,20].
The cotton indicators are constructed as follows:
G D D d , c = max 0 , T m e a n , d − 12
E D D d , c = max 0 , T m a x , d − 32
F D D d , c = max 0,12 − T m i n , d
The maize indicators are constructed as follows:
G D D d , m = max 0 , T m e a n , d − 10
E D D d , m = max 0 , T m a x , d − 30
F D D d , m = max 0,10 − T m i n , d

2.2.2. Growing-Season Aggregation and Precipitation Variable

Daily temperature indicators are summed over the growing season to obtain annual growing-season indicators:
G D D i t = ∑ d ∈ G S G D D d
E D D i t = ∑ d ∈ G S E D D d
F D D i t = ∑ d ∈ G S F D D d
The growing season is defined as April–October for cotton and May–September for maize. Precipitation is defined as total daily precipitation over the growing season:
P r e c i p i t = ∑ d ∈ G S P R E i d t
To facilitate coefficient interpretation and avoid numerical instability caused by large-scale variables, precipitation is rescaled in units of 100 mm in the regressions.
P i t = P r e c i p i t 100 ,     P i t 2 = P i t 2
where P i t denotes growing-season precipitation measured in units of 100 mm, and P i t 2 denotes the squared precipitation term used in the regression models.

2.3. Descriptive Statistics

Table 1 summarizes the core variables. Cotton (mean = 1.60 t/ha, standard deviation = 0.40) and maize (mean = 8.33 t/ha, standard deviation = 2.65) differ substantially in yield level and volatility. High- and low-yield groups are defined separately for each crop according to whether a county’s long-run average yield is above the crop-specific median. In terms of climate exposure, maize has higher mean growing-season EDD exposure (2.72 versus 1.65) and slightly higher precipitation. In the comparison between high-yield and low-yield groups, high-yield maize counties have lower average EDD exposure and higher precipitation. The corresponding gradients are weaker for cotton. This pattern is consistent with the interpretation that cotton production conditions are more homogeneous under intensive irrigation and related field management, and it provides descriptive context for the regression analysis.
Table 1. Summary statistics for crop yields and growing-season weather variables.

2.4. Empirical Model and Estimation Strategy

2.4.1. Baseline Fixed-Effects Model

The baseline specification is a two-way fixed-effects (TWFEs) county-year panel model:
ln y i e l d i t = β 1 E D D i t + β 2 F D D i t + β 3 G D D i t + β 4 P i t + β 5 P i t 2 + μ i + λ t + ε i t
County fixed effects control for time-invariant geographic endowments, long-run institutions, soil conditions, and other county characteristics. Year fixed effects capture region-wide shocks, such as prices, macro policies, and technological progress. This identification framework is widely used in agricultural climate economics to exploit within-county interannual weather variation [16,17,18]. Standard errors are clustered at the county level to allow serial correlation and heteroskedasticity within counties [37,38,39], which is particularly important when weather variables are serially correlated and county shocks are persistent. The fixed effects design does not assume that all counties have identical soils, varieties, irrigation conditions, fertilizer use, or farm practices; rather, it absorbs time-invariant county differences while leaving time-varying management and input changes as an important interpretive limitation.

2.4.2. Precipitation Nonlinearity and Turning Point

To capture potential nonlinear effects of precipitation on yield, the model retains both the linear and squared precipitation terms and computes the turning point as:
T P m m = 100 × − β P / 2 β P 2
We use the delta method to approximate the variance and construct confidence intervals [35]. Lind and Mehlum [36] emphasize that testing U-shaped relationships in quadratic models requires examining the direction of marginal effects and the location of the turning point within the sample range. Following this logic, we interpret precipitation nonlinearity jointly using coefficient signs, the turning point, and its confidence interval.

2.4.3. Productivity-Group Heterogeneity

Because regions with different yield levels may differ in irrigation conditions, management capacity, and marginal vulnerability, we further estimate the following interaction model:
ln y i e l d i t = β 1 E D D i t + β 2 F D D i t + β 3 G D D i t + β 4 P i t + β 5 P i t 2 + θ 1 E D D i t × H i g h i + θ 2 G D D i t × H i g h i + μ i + λ t + ε i t
where High is a high-yield county dummy. High-yield counties are defined separately for each crop according to whether a county’s long-run average yield over the sample period is above the crop-specific median. The interaction specification retains the baseline weather controls and fixed effects and tests whether marginal weather elasticities vary systematically with yield level.

2.4.4. Post-2010 Changes in Weather Sensitivity

To examine temporal heterogeneity in estimated weather sensitivity, we estimate the following interaction model:
ln y i e l d i t = β 1 E D D i t + β 2 F D D i t + β 3 G D D i t + β 4 P i t + β 5 P i t 2 + ϕ 1 E D D i t × P o s t 2010 t + ϕ 2 G D D i t × P o s t 2010 t + μ i + λ t + ε i t
Post-2010 is a period indicator equal to one for 2010 and later years. This specification follows the climate-economics literature that examines temporal changes in weather–yield relationships by comparing period-specific weather sensitivities [17,18]. We use 2010 as a policy–water–technology transition anchor rather than as a statistically estimated structural break. The choice has both empirical and contextual motivations. Empirically, it separates the 2000–2020 panel into earlier and later periods of comparable length. Contextually, the years around 2010 marked a broader strengthening of Xinjiang’s regional development, water-conservancy investment, water-saving irrigation, and production-technology environment, followed by further promotion of mechanized cotton production, agronomic–mechanical integration, film-mulched drip irrigation, and standardized field management in the Northwest Inland cotton region. Accordingly, the interaction terms compare estimated weather–yield slopes between the earlier period and the later policy–water–technology transition period. We interpret them as evidence on the net temporal evolution of crop weather sensitivity in Xinjiang’s irrigation-intensive oasis agriculture, rather than as the independent causal effect of any single policy, irrigation project, or management technology.

2.5. Robustness Specifications

The study conducts two additional robustness checks. First, we add a common linear time trend to the county fixed-effects specification to control for broad temporal changes in yield levels. Second, we use prefecture-by-year fixed effects to absorb annual shocks shared by counties within the same prefecture. These specifications test the core results from the perspectives of alternative time controls and regional controls, and further characterize the temporal and spatial sources of identifying variation.

3. Results

3.1. Baseline Regression Results

Table 2 reports the baseline estimates for cotton and maize yields with temperature degree-day indicators and precipitation. Because the main patterns are similar across specifications, we use the two-way fixed-effects results as the reference. The two crops differ in both temperature and precipitation associations.
Table 2. Baseline estimates for cotton and maize yield models with degree-day and precipitation variables.
For cotton (Panel A), GDD is positive and statistically significant in the two-way fixed-effects specification. The EDD coefficient is negative but does not reach conventional levels of statistical significance. Overall, the association between cotton yield and general heat accumulation is clearer. The precipitation terms suggest a convex association for cotton in the baseline specification.
For maize (Panel B), the most stable result is the negative association between GDD and yield. FDD is negative in the county fixed-effects plus trend specification, but becomes statistically insignificant after year fixed effects are added, indicating that the cold-exposure association is sensitive to controlling for common year shocks. By contrast, the negative GDD association persists in the two-way fixed-effects model. The estimates show that maize yield is more stably negatively associated with heat accumulation in this setting.

3.2. Nonlinear Precipitation–Yield Association for Cotton

Table 3 examines precipitation nonlinearity for cotton. The linear precipitation term is negative but does not reach conventional levels of statistical significance, while the squared term is positive and statistically significant. Based on the point estimates from the quadratic specification, the cotton precipitation–yield curve changes from a negative marginal association to a positive marginal association at approximately 141.07 mm (95% CI: 27.75–225.63 mm). Following the sample-range logic emphasized by Lind and Mehlum [36], the coefficient signs, the location of the turning point, and the confidence interval jointly indicate a convex estimated curve within the observed precipitation range. Because the turning-point location remains uncertain, the 141 mm value is best interpreted as an in-sample empirical feature of the fitted nonlinear relationship rather than as a directly actionable agronomic management threshold. We therefore do not interpret the convex precipitation pattern as evidence for a specific agronomic mechanism. The result differs from the simpler water-stress interpretation often used in rain-fed settings; in Xinjiang’s irrigated oasis production context, irrigation scheduling, field humidity, solar radiation, and disease pressure provide relevant contextual considerations and useful directions for future mechanism-oriented research, but they are not separately identified by the present county-panel specification.
Table 3. Quadratic precipitation–yield estimates and turning point for cotton.

3.3. Productivity-Group Heterogeneity

Table 4 tests whether marginal climate associations differ between high-yield and low-yield counties. For both crops, the interaction terms do not show stable statistical significance overall. The results therefore support the view that differences across productivity groups mainly appear in average yield levels, while marginal climate slopes do not exhibit systematic group differences.
Table 4. Productivity-group heterogeneity in temperature sensitivity.

3.4. Temporal Changes in Weather Sensitivity

Table 5 examines whether crop-specific weather–yield associations changed across the post-2010 production period by interacting the weather variables with a post-2010 indicator. The indicator equals one for 2010 and later years. This timing is motivated by the documented strengthening of Xinjiang’s agricultural modernization during this period, including the expansion of water-saving irrigation, plastic-film and drip-irrigation systems, mechanization, and agricultural information services in oasis agriculture. The interaction estimates therefore show whether the estimated weather–yield relationships differed between the pre-2010 and post-2010 production environments. For cotton, the post-2010 interaction is positive and statistically significant for EDD and negative and statistically significant for GDD, indicating that the estimated temperature-yield associations changed in the later period. For maize, the EDD interaction is positive but not statistically significant, while the GDD interaction is negative and statistically significant, suggesting a more limited change in the estimated temperature–yield relationship. Overall, these results point to crop-specific changes in weather–yield associations under an evolving production environment, while the separate contribution of each individual policy, technology, or management practice cannot be decomposed in the present county-panel specification.
Table 5. Post-2010 interaction estimates for temperature sensitivity.

3.5. Robustness Results

Table 6 reports the robustness checks. After adding a common linear time trend, several broad patterns remain, including the positive association between cotton GDD and yield and the negative association between maize GDD and yield. After adding prefecture-by-year fixed effects, several coefficients shrink, lose precision, or change sign, indicating that annual weather shocks correlated within prefectures absorb part of the identifying variation in the baseline specification. This result further clarifies the temporal and spatial sources of variation on which the core conclusions depend.
Table 6. Estimates under alternative fixed-effects and trend controls.

4. Discussion

The results show that county-level weather exposure and yield relationships vary not only by crop but also by model specification and time period. The discussion focuses on four issues: how to interpret the cotton precipitation–yield nonlinearity, why changes in weather sensitivity differ across crops, how spatial heterogeneity and robustness results delimit the level of interpretation, and what these findings imply for climate-risk management in irrigation-intensive oasis agriculture.

4.1. Precipitation–Yield Nonlinearity in Irrigated Cotton Fields

The cotton precipitation result is the finding that most requires contextual interpretation. In the baseline specification, Table 3 suggests a convex precipitation–yield relationship for cotton, with an estimated turning point near 141 mm and a negative marginal association over much of the observed range. The confidence interval around the turning point is wide, and the precipitation pattern is less stable once prefecture-by-year fixed effects are added; these features delimit the precision and source of the result. Even with that qualification, the baseline pattern is substantively useful because it differs from the intuitive interpretation in rain-fed agriculture, where additional precipitation is often expected to relieve water stress. In Xinjiang cotton fields, precipitation enters a highly managed water regime: drip irrigation, plastic-film mulching, and field regulation jointly determine the hydrothermal conditions actually experienced by the crop. Previous modeling shows that irrigation can alter the spatial distribution of temperature extremes [15]. In systems with strong controlled water supply, additional precipitation may not be equivalent to effective water input and may instead indicate field conditions that deviate from targeted management.
This result does not imply that rainfall itself is necessarily harmful. Rather, it indicates that the county-level precipitation variable captures the net association of multiple field conditions. Rainfall may simultaneously affect soil moisture, radiation, humidity, disease pressure, and the timing of field operations, and these channels cannot be fully decomposed in a county panel. Existing cotton studies suggest that precipitation, sunshine duration, relative humidity, and canopy microclimate affect cotton growth and yield [40,41], and that temperature, disease, and agronomic management can jointly influence cotton yield [24,25]. The precipitation term in this study is therefore better interpreted as a composite field exposure shaped by water supply, humidity, radiation, and disease pressure. This finding suggests that precipitation–yield relationships in irrigation-intensive oasis agriculture should not be interpreted using the same framework as rain-fed systems.

4.2. Crop Differences in Temporal Changes in Weather Sensitivity

The interaction terms in Table 5 further show that changes in weather sensitivity did not occur synchronously across the two crops. Cotton shows weaker estimated sensitivity to extreme heat after 2010, whereas maize does not exhibit a comparable attenuation in heat sensitivity and instead shows a more negative association with GDD. This difference indicates that climate risk can be differentiated by crop structure even within the same region. The same temperature exposure may translate into different yield associations through crop-specific physiological requirements, management intensity, and input incentives.
This crop difference can be understood from the perspectives of production incentives and management inputs. In Xinjiang, changes in irrigation management, plastic-film mulching, cultivar choice, and field management improvement are closely related to expected returns. As a higher-value and more management-intensive cash crop, cotton has stronger incentives for capital-intensive and management-intensive responses [26,27,28]. Maize, as a grain crop, may face stronger constraints related to returns, service provision, and infrastructure. The results in Table 5 are therefore best interpreted as reduced-form evidence of temporal changes in crop weather sensitivity, rather than as the direct causal effect of any single technology or policy.
This point is especially important for adaptation research. Discussing “agricultural adaptation” only at the regional level can obscure adjustment differences across crops. The results point to crop-specific changes in sensitivity: cotton and maize operate in the same oasis agricultural environment, yet their estimated relationships evolve differently over time. Future research linking county panels with farm surveys, remote-sensing indicators of irrigation infrastructure, varietal replacement information, or project-level policy data would be better suited to distinguish among mechanisms related to returns, credit constraints, extension services, and institutional water management.

4.3. Spatial Heterogeneity, Productivity Groups, and Robustness Evidence

The productivity-group results in Table 4 add another layer of evidence. High-yield and low-yield counties differ clearly in average yield, but marginal weather sensitivity does not show the same systematic differentiation. This result indicates that long-run yield levels and short-run marginal responses to weather shocks do not necessarily correspond one to one. High-yield areas may have better infrastructure, management capacity, or cultivar conditions, but these advantages need not appear as statistically distinct marginal coefficients on EDD or GDD. For county panel research, this finding suggests that productivity levels can explain differences in baseline yield but may not be sufficient to explain spatial differences in weather–yield elasticities.
The robustness results in Table 6 further delimit the level of interpretation. Cotton precipitation nonlinearity is clearest in the baseline specification and the common linear trend specification. After prefecture-by-year fixed effects are added, coefficient signs and significance change, indicating that the result depends partly on common prefecture-year variation. This change should not be interpreted as a complete invalidation of the result, but rather as a change in the source of identification. The baseline model uses within-county interannual weather variation while retaining annual differences at the prefecture level; the stricter prefecture-by-year fixed-effects specification uses only residual variation among counties within the same prefecture-year. The study is therefore better suited to identifying regional-scale weather–yield relationships and crop differences than to recovering farm-level management mechanisms directly.
This robustness pattern also implies that mechanism discussion should remain consistent with the level of identification. County-level data can reveal which weather–yield relationships are systematic enough to merit further investigation, but disease pressure, field humidity, irrigation intensity, drainage capacity, and specific adoption behavior still require finer-scale data for identification. In other words, the contribution of this study is not to complete mechanism decomposition, but to identify the priority variables and spatial scales for subsequent mechanism-oriented research.

4.4. Implications for Climate-Risk Management in Oasis Agriculture

Taken together, the results point to a central conclusion: climate risk in irrigation-intensive oasis agriculture is not determined by meteorological exposure alone, but is jointly shaped by meteorological exposure, crop structure, irrigation institutions, and management intensity. Cotton precipitation nonlinearity suggests that, in systems where water supply is highly engineered, rainfall-related risk may arise through composite conditions involving field humidity, radiation, disease, and drainage. The maize heat association and temporal-change results indicate that heat-risk management for grain crops may require support strategies different from those used for cotton.
Climate-risk management should therefore not be designed only at the regional level, nor should it focus only on increasing water supply. For cotton, irrigation scheduling, drainage capacity, humidity management under plastic-film mulching, and disease monitoring are priority channels for understanding and reducing precipitation-related risk. For maize, the stable negative association with heat accumulation and the weaker evidence of later-period sensitivity attenuation suggest that risk-reducing practices may depend more on public extension, information services, and infrastructure support. This crop differentiation means that sustainable agricultural management needs to consider crop returns, management intensity, and the capacity to absorb new technologies at the same time.

5. Conclusions and Policy Implications

5.1. Research Synthesis

Using county-level panel data for Xinjiang from 2000 to 2020, this study documents four patterns in the association between crop yields and growing-season weather within a two-way fixed-effects framework. First, the temperature–yield relationship differs by crop. For cotton, GDD is positively associated with yield in the baseline two-way fixed-effects specification, while the EDD coefficient is negative but does not reach conventional levels of statistical significance. For maize, the GDD association is negative and more stable across specifications. Second, baseline and common-trend specifications suggest a convex precipitation–yield relationship for cotton, with an estimated turning point near 141 mm, implying a negative marginal precipitation association over much of the observed range; the wide confidence interval and stricter robustness checks delimit the precision and identifying source of this pattern. Third, the interaction models show that differences between high-yield and low-yield counties mainly appear in average yield levels, while marginal climate sensitivity does not show systematic group differences. Fourth, post-2010 interactions indicate crop-differentiated changes in heat sensitivity, suggesting that weather–yield relationships evolved differently across cotton and maize during the sample period.
Overall, the estimates indicate crop-specific climate responses within Xinjiang oasis agriculture. The results are consistent with the view that irrigation-intensive production systems can reshape observed precipitation–yield relationships and that changes in weather sensitivity during the sample period are crop differentiated. At the same time, the cotton precipitation pattern is clearest in the baseline specification and the common linear trend specification; the prefecture-by-year fixed-effects results further indicate that this pattern is primarily identified from the regional annual variation retained in the baseline model, rather than from residual cross-county differences within the same prefecture-year. Therefore, the contribution of this study is to identify broad patterns in crop weather sensitivity and temporal change at the county scale and to interpret these patterns within the production context of irrigation-intensive oasis agriculture.

5.2. Policy Implications

For water-scarce, irrigation-intensive systems, the results point to several management margins that deserve closer monitoring. In the baseline estimates, cotton yields are lower in wetter years over much of the observed range, indicating that irrigation scheduling, drainage capacity, and field-level moisture control are as important as water supply itself. More broadly, rainfall in oasis systems may operate through water input, humidity, radiation, disease conditions, and the timing of field operations. From this perspective, the precipitation result points to a practical management margin that standard rain-fed interpretations can easily miss.
The crop differences documented here also suggest that climate management needs to distinguish between crop types and production systems. Where private returns are higher, adjustment incentives are stronger. Where private returns are lower, the diffusion of risk-reducing practices depends more on public support. Public extension and support programs can focus on reducing information, trial, and coordination costs and supporting practices that lower-return crops may under-adopt on their own. These implications are not direct estimates of individual interventions, but they help identify concrete priorities for crop-differentiated climate-risk management.

5.3. Limitations and Future Research

This study still requires further development in three respects. First, the county-level design identifies regional-scale yield sensitivity. Farm-level management responses, spatial spillovers, and aggregation effects require finer-scale data and complementary research designs. Potential mechanisms behind the precipitation results, such as disease pressure, radiation, humidity, irrigation intensity, drainage capacity, and plastic-film management, can also be linked using finer-scale data. Second, some estimates depend on the source of identifying variation, as shown by the stricter robustness checks. Compound heat-drought, wet-heat, cold-stress, and other compound extreme-event risks provide natural extensions of the current framework. Third, the available county panel does not observe all time-varying soil conditions, varieties, fertilizer use, irrigation intensity, farm practices, or project-level policy implementation. Future research combining county panels with farm surveys, remote-sensing measures of irrigation infrastructure, varietal replacement data, and project-level management information would be better suited to distinguish among specific mechanisms while retaining the regional perspective developed in this study.

Author Contributions

Conceptualization, J.G. and L.C.; methodology, G.J., W.S. and L.C.; software, G.J. and W.S.; validation, J.Z., X.L. and L.C.; formal analysis, G.J. and W.S.; investigation, J.G.; resources, J.G. and L.C.; data curation, J.G., G.J. and W.S.; writing—original draft preparation, J.G.; writing—review and editing, J.Z., X.L. and L.C.; visualization, G.J. and W.S.; supervision, L.C.; project administration, L.C.; funding acquisition, L.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Xinjiang Academy of Agricultural Sciences through the project “Key Technologies for Intelligent Production Management and Decision Support of Major Crops in Xinjiang: Grain and Cotton”, grant number xjnkywdzc-2024001-05-02. The funder had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Institutional Review Board Statement

Not applicable.

Data Availability Statement

The county-level yield-climate panel data supporting the findings of this study are available from the corresponding author upon reasonable request. Some underlying source materials were compiled from the Xinjiang Statistical Yearbook, related agricultural statistical materials, and institutional meteorological records, and are therefore subject to source-access conditions.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Schlenker, W.; Roberts, M.J. Nonlinear temperature effects indicate severe damages to U.S. crop yields under climate change. Proc. Natl. Acad. Sci. USA 2009, 106, 15594–15598. [Google Scholar] [CrossRef] [Scilit]
  2. Lobell, D.B.; Bänziger, M.; Magorokosho, C.; Vivek, B. Nonlinear heat effects on African maize as evidenced by historical yield trials. Nat. Clim. Change 2011, 1, 42–45. [Google Scholar] [CrossRef] [Scilit]
  3. Lobell, D.B.; Hammer, G.L.; McLean, G.; Messina, C.; Roberts, M.J.; Schlenker, W. The critical role of extreme heat for maize production in the United States. Nat. Clim. Change 2013, 3, 497–501. [Google Scholar] [CrossRef] [Scilit]
  4. Lesk, C.; Rowhani, P.; Ramankutty, N. Influence of extreme weather disasters on global crop production. Nature 2016, 529, 84–87. [Google Scholar] [CrossRef] [Scilit]
  5. Ray, D.K.; Gerber, J.S.; MacDonald, G.K.; West, P.C. Climate variation explains a third of global crop yield variability. Nat. Commun. 2015, 6, 5989. [Google Scholar] [CrossRef] [Scilit]
  6. Zhao, C.; Liu, B.; Piao, S.; Wang, X.; Lobell, D.B.; Huang, Y.; Huang, M.T.; Yao, Y.T.; Bassu, S.; Ciais, P.; et al. Temperature increase reduces global yields of major crops in four independent estimates. Proc. Natl. Acad. Sci. USA 2017, 114, 9326–9331. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  7. Sacks, W.J.; Cook, B.I.; Buenning, N.; Levis, S.; Helkowski, J.H. Effects of global irrigation on the near-surface climate. Clim. Dyn. 2009, 33, 159–175. [Google Scholar] [CrossRef] [Scilit]
  8. Kang, S.; Eltahir, E.A.B. Impact of irrigation on regional climate over Eastern China. Geophys. Res. Lett. 2019, 46, 5499–5505. [Google Scholar] [CrossRef] [Scilit]
  9. Siebert, S.; Webber, H.; Zhao, G.; Ewert, F. Heat stress is overestimated in climate impact studies for irrigated agriculture. Environ. Res. Lett. 2017, 12, 054023. [Google Scholar] [CrossRef] [Scilit]
  10. Tack, J.; Barkley, A.; Hendricks, N. Irrigation offsets wheat yield reductions from warming temperatures. Environ. Res. Lett. 2017, 12, 114027. [Google Scholar] [CrossRef] [Scilit]
  11. Wang, Z.; Wu, Q.; Fan, B.; Zheng, X.; Zhang, J.; Li, W.; Guo, L. Effects of mulching biodegradable films under drip irrigation on soil hydrothermal conditions and cotton (Gossypium hirsutum L.) yield. Agric. Water Manag. 2019, 213, 477–485. [Google Scholar] [CrossRef] [Scilit]
  12. Zong, R.; Wang, Z.; Zhang, J.; Li, W. The response of photosynthetic capacity and yield of cotton to various mulching practices under drip irrigation in Northwest China. Agric. Water Manag. 2021, 249, 106814. [Google Scholar] [CrossRef] [Scilit]
  13. Kar, G.; Kumar, A. Surface energy fluxes and crop water stress index in groundnut under irrigated ecosystem. Agric. For. Meteorol. 2007, 146, 94–106. [Google Scholar] [CrossRef] [Scilit]
  14. Javadian, M.; Scott, R.L.; Woodgate, W.; Richardson, A.D.; Dannenberg, M.P.; Smith, W.K. Canopy temperature dynamics are closely aligned with ecosystem water availability across a water- to energy-limited gradient. Agric. For. Meteorol. 2024, 357, 110206. [Google Scholar] [CrossRef] [Scilit]
  15. Li, X.; Li, X.; Hua, W.; Ma, H.; Zhou, J.; Pang, X. Modeling the effects of present-day irrigation on temperature extremes over China. Front. Earth Sci. 2023, 11, 1084892. [Google Scholar] [CrossRef] [Scilit]
  16. Deschenes, O.; Greenstone, M. The economic impacts of climate change: Evidence from agricultural output and random fluctuations in weather. Am. Econ. Rev. 2007, 97, 354–385. [Google Scholar] [CrossRef] [Scilit]
  17. Burke, M.; Emerick, K. Adaptation to climate change: Evidence from U.S. agriculture. Am. Econ. J. Econ. Policy 2016, 8, 106–140. [Google Scholar] [CrossRef] [Scilit]
  18. Chen, S.; Gong, B. Response and adaptation of agriculture to climate change: Evidence from China. J. Dev. Econ. 2021, 148, 102557. [Google Scholar] [CrossRef] [Scilit]
  19. Lobell, D.B.; Schlenker, W.; Costa-Roberts, J. Climate trends and global crop production since 1980. Science 2011, 333, 616–620. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  20. Dell, M.; Jones, B.F.; Olken, B.A. What do we learn from the weather? The new climate-economy literature. J. Econ. Lit. 2014, 52, 740–798. [Google Scholar] [CrossRef] [Scilit]
  21. Angrist, J.D.; Pischke, J.-S. The credibility revolution in empirical economics: How better research design is taking the con out of econometrics. J. Econ. Perspect. 2010, 24, 3–30. [Google Scholar] [CrossRef] [Scilit]
  22. Auffhammer, M.; Hsiang, S.M.; Schlenker, W.; Sobel, A. Using weather data and climate model output in economic analyses of climate change. Rev. Environ. Econ. Policy 2013, 7, 181–198. [Google Scholar] [CrossRef] [Scilit]
  23. Ortiz-Bobea, A.; Ault, T.R.; Carrillo, C.M.; Chambers, R.G.; Lobell, D.B. Anthropogenic climate change has slowed global agricultural productivity growth. Nat. Clim. Change 2021, 11, 306–312. [Google Scholar] [CrossRef] [Scilit]
  24. Zhang, T.; Xie, Z.; Zhou, J.; Feng, H.; Zhang, T. Temperature impacts on cotton yield superposed by effects on plant growth and verticillium wilt infection in China. Int. J. Biometeorol. 2024, 68, 199–209. [Google Scholar] [CrossRef] [Scilit]
  25. Zhou, Y.; Li, F.; Xin, Q.; Li, Y.; Lin, Z. Historical variability of cotton yield and response to climate and agronomic management in Xinjiang, China. Sci. Total Environ. 2024, 912, 169327. [Google Scholar] [CrossRef] [Scilit]
  26. Hornbeck, R.; Keskin, P. The historically evolving impact of the Ogallala Aquifer: Agricultural adaptation to groundwater and drought. Am. Econ. J. Appl. Econ. 2014, 6, 190–219. [Google Scholar] [CrossRef] [Scilit]
  27. Conley, T.G.; Udry, C.R. Learning about a new technology: Pineapple in Ghana. Am. Econ. Rev. 2010, 100, 35–69. [Google Scholar] [CrossRef] [Scilit]
  28. Suri, T. Selection and comparative advantage in technology adoption. Econometrica 2011, 79, 159–209. [Google Scholar] [CrossRef] [Scilit]
  29. Thevs, N.; Peng, H.; Rozi, A.; Zerbe, S.; Abdusalih, N. Water allocation and water consumption of irrigated agriculture and natural vegetation in the Aksu-Tarim River Basin, Xinjiang, China. J. Arid Environ. 2015, 112, 87–97. [Google Scholar] [CrossRef] [Scilit]
  30. Becker, R.; Schauberger, B.; Merz, R.; Schulz, S.; Gornott, C. The vulnerability of winter wheat in Germany to air temperature, precipitation or compound extremes is shaped by soil-climate zones. Agric. For. Meteorol. 2025, 361, 110322. [Google Scholar] [CrossRef] [Scilit]
  31. Grafton, R.Q.; Williams, J.; Perry, C.J.; Molle, F.; Ringler, C.; Steduto, P.; Udall, B.; Wheeler, S.A.; Wang, Y.; Garrick, D.; et al. The paradox of irrigation efficiency. Science 2018, 361, 748–750. [Google Scholar] [CrossRef] [Scilit]
  32. Siebert, S.; Döll, P. Quantifying blue and green virtual water contents in global crop production as well as potential production losses without irrigation. J. Hydrol. 2010, 384, 198–217. [Google Scholar] [CrossRef] [Scilit]
  33. Geerts, S.; Raes, D. Deficit irrigation as an on-farm strategy to maximize crop water productivity in dry areas. Agric. Water Manag. 2009, 96, 1275–1284. [Google Scholar] [CrossRef] [Scilit]
  34. Grassini, P.; Yang, H.; Irmak, S.; Thorburn, J.; Burr, C.; Cassman, K.G. High-yield irrigated maize in the Western U.S. Corn Belt: II. Irrigation management and crop water productivity. Field Crops Res. 2011, 120, 133–141. [Google Scholar] [CrossRef] [Scilit]
  35. Oehlert, G.W. A note on the delta method. Am. Stat. 1992, 46, 27–29. [Google Scholar] [CrossRef] [Scilit]
  36. Lind, J.T.; Mehlum, H. With or without U? The appropriate test for a U-shaped relationship. Oxf. Bull. Econ. Stat. 2010, 72, 109–118. [Google Scholar] [CrossRef] [Scilit]
  37. Bertrand, M.; Duflo, E.; Mullainathan, S. How much should we trust differences-in-differences estimates? Q. J. Econ. 2004, 119, 249–275. [Google Scholar] [CrossRef] [Scilit]
  38. Cameron, A.C.; Miller, D.L. A practitioner’s guide to cluster-robust inference. J. Hum. Resour. 2015, 50, 317–372. [Google Scholar] [CrossRef] [Scilit]
  39. MacKinnon, J.G.; Nielsen, M.Ø.; Webb, M.D. Cluster-robust inference: A guide to empirical practice. J. Econom. 2023, 232, 272–299. [Google Scholar] [CrossRef] [Scilit]
  40. Li, N.; Lin, H.; Wang, T.; Li, Y.; Liu, Y.; Chen, X.; Hu, X. Impact of climate change on cotton growth and yields in Xinjiang, China. Field Crops Res. 2020, 247, 107590. [Google Scholar] [CrossRef] [Scilit]
  41. Zhang, N.; Tian, L.; Feng, L.; Xu, W.; Li, Y.; Xing, F.; Fan, Z.; Xiong, S.; Tang, J.; Li, C.; et al. Boll characteristics and yield of cotton in relation to the canopy microclimate under varying plant densities in an arid area. PeerJ 2021, 9, e12111. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.