A two-year study (2015 and 2016) was conducted in 2015 and 2016 at Henan Agricultural Research Farm located in Zhengzhou, Henan Province, China, as shown in Figure 1 . The study site has a long-term average annual precipitation of 609.5 mm, a mean temperature of 14.7°C, and a frost-free period of 216 d. The baseline soil fertility in the 0–20 cm soil layer was as follows: soil organic matter 9.18 g kg^-1, total N 0.98 g kg^-1, alkaline-extractable N 63.5 mg kg^-1, Olsen-extractable P 22.8 mg kg^-1, and Olsen-extractable K 133.8 mg kg^-1.

This study includes four sowing date treatments: spring sowing in late April (SSA), early summer sowing in late May (SSM), summer sowing in mid-June (SSJ), and autumn sowing in early July (ASJ). Sowing occurred from late April to early July to capture the earliest and latest dates suitable for sowing corn in the study region. Each plot was 6 m long and 3.6 m wide and consisted of 6 rows. A randomized complete block design with three replications was used for this study. The weather variables for each sowing date treatment at the experimental site are presented in Table 1 . A widely grown corn cultivar, Zhengdan 958 (ZD958) was chosen for this study because of its high yield, resistance to multiple stressors, and extensive adaptability. Before sowing, plots were finely prepared with a disk plow. Corn was planted at a row spacing of 0.6 m and plant spacing of 24.7 cm. To achieve uniform plant density, all plots were thinned at the V4 stage to achieve population densities of 67,500 plants ha^-1, which was recommended by the local corn grower guide to maximize grain yield under unstressed conditions.

The fertilizer application rates were determined based on soil test recommendations. Basal fertilizers were applied at a rate of 90 kg N ha^-1, 120 kg P₂O₅ ha^-1, and 50 kg K₂O ha^-1. Furthermore, additional N fertilizer (135 kg N ha^-1) was applied at the 12th leaf stage (V12). The experimental field was first irrigated using sprinkler systems with groundwater as the water source. The sprinklers systems were placed in the middle of each row. The sprinklers spacing was 3 cm, and the flow rate was 0.32 L h⁻¹ at an operating pressure of 0.1 MPa (Tianye Inc., China). Each plot was connected to a high precision water meter (LXS-40F, Ningbo, China) and control valve. The irrigation level was determined based on the local irrigation quota. The irrigation generally occurred two or three times depending on rainfall. The maize was irrigated immediately after sowing and was irrigated again at the V12 stage, following a split nitrogen application. The total amount of irrigation water applied ranged from 50 to 75 mm based on crop requirement. The local best management practices were followed for achieving high yield. All experimental fields were well-managed, and no water or drought stress or pest damage was observed during the growing season.2.3 Weather data

Long-term (1987-2016) weather data such as daily weather data, including the average daily temperature (ADT), daily maximum temperature (T_max ), daily minimum temperature (T_min ), daily sunshine duration (DSD), and effective precipitation (EP) were obtained from the China Meteorological Data Sharing Service System (http://data.cma.cn/site/article/id/347.html) Zhengzhou station. Daily sunshine hours were the actual duration of sunlight throughout the entire day (Hu et al., 2019; Liu et al., 2010). During the study period, weather data such as temperature and precipitation were collected from an on-site research farm weather station using an HL-10 automatic weather station (Jauntering International Corporation, Taiwan), which was located approximately 500 meters away from the experimental field. The long-term (1987-2016) mean annual temperature and effective precipitation during corn growing season were 23.9°C and 406.6 mm, respectively ( Figure 2 ). Compared to the 30-year average, 2016 was considered a hot and wet year and 2015 was close to a normal year. The distribution of growing degree-days, average daily temperature and precipitation during corn growing season in 2015 and 2016 are shown in Figure 3 . The first killing frost occurred on the October 24^th, 2015; and October 23^th, 2016.

The growth and development of crops occur within a range of temperatures, defined by a base temperature (T_base ) and an upper temperature threshold (T_ut ). In most field environments, temperatures fall between T_base and T_ut , and there is a linear relationship between cumulative temperature and the growth rate. This concept, is known as “GDD” or “heat units”, is used to describe and predict crop development. GDD is a more reliable predictor of crops are development than calendar days because it is not affected by the temperature range in which the crop is grown, provided temperatures remain between T_base and T_ut . Base temperatures vary among crops (Qian et al., 2010). For warm season crop corn, we set T_base and T_ut to 10°C and 35°C, respectively. The GDD was calculated as the following (Cross and Zuber, 1972; McMaster and Wilhelm, 1997):

Where N, T_max , T_min , and T_base are number of days from sowing to harvesting, the daily maximum temperature, the daily minimum temperature, and the base temperature, respectively. If T_{min <} T_base , then T_{min =} T_base . Similarly, if T_max > T_ut , then T_{max =} T_ut .

Before silking, the selected corn ears were marked and bagged until the silks grew 2-3 cm long. Then, manual pollination was conducted from 9:00 am to 10:00 am after removing the bags to ensure uniform pollination timing. Silking date, which is defined as the date when at least 50% of plants in a plot have one or more silks emerged from the upper ear, was recorded based on daily observations. To estimate the grain filling rate, kernels were measured from pre-tagged ears in the central rows of each plot. Starting from 10 days after silking, ears from three plants in each plot were sampled at five-day intervals until physiological maturity, indicated by a black layer in the kernel. Kernels were dried at 80°C to a constant weight and weighed. The process of grain filling was fitted with the logistic equation (Richards, 1959), and the grain filling parameters of corn were calculated as follows:

A is the maximum kernel weight (g), t is the days after pollination (d), B is the coefficient at the initial stage, and K is the slope of the logistic curve. G_max and G_mean are the maximum and mean kernel growth rates during the effective grain filling period, respectively. D is the effective grain filling duration.

At the physiological maturity stage, an area of 2.4 m² from the center of each plot (two central 2-m long rows) was manually harvested, and the weight of the grains was measured. A portable moisture meter (PM8188-A, Kett Electric Laboratory, Tokyo, Japan) was used to determine the grain moisture content. Grain yield was adjusted to a moisture content of 14% and reported. To determine 1000-kernel weight, a random sample of 1000 kernels was selected and weighted.

Data of grain yield (GY), kernel number per plant (KNP), and 1000-kernel weight (TKW) were analyzed using the Proc Mixed Model of SAS (Littell et al., 1998), considering sowing date a fixed factor, block a random factor. Since a preliminary data analysis showed a significant interaction between year and sowing date, data were analyzed separately each year. When the sowing date effect was significant at the 5% level, the means comparisons (Least Significant Difference) were performed and the least squares means (e.g., lsmeans) were reported. The grain filling process in the field was simulated using CurveExpert Professional 2.6.3 software (https://www.curveexpert.net/).

Structural equation modelling (SEM) (Schermelleh-Engel et al., 2003) was performed using IBM SPSS Amos version 19 statistical software (Arbuckle, 2010). SEM was chosen to explore the direct and indirect effects of weather factors and grain filling parameters on kernel weight. The overall fit of the model was evaluated using several indicators, including the χ² statistic with the associated probability, the root mean square error of approximation (RMSEA) with the associated probability, and the Bentler-Bonett Index or Comparative Fit Index (CFI). A RMSEA p-value greater than 0.05 and a CFI value greater than 0.90 are considered to indicate a good fit of the model. The significance of each path in the model was determined by the probability level (P< 0.05).

Long term (30-year) temperature data showed that peak temperatures during corn growing seasons typically occurs from mid-June to late July ( Figure 4 ). Analysis of the number of high-temperature days (>35°C) during the silking stage indicated the likelihood of high temperature declines after early August. The silking stage of corn planted in late April (e.g., SSA) occurred in early June, while the silking stage of early summer corn planted in late May (SSM) fell in early August.

The grain filling window varied due to different sowing dates, ranging from June 20 to Oct 20 in our study. In comparison with the 30-year average (1987-2016), the average daily temperature increased by 2.1% and 7.9% in 2015 and 2016, respectively ( Table 1 ). Growing degree-days increased by 3.3% and 12.3% in 2015 and 2016, respectively. Daily sunshine duration decreased by 4.8% in 2015 and increased by 3.2% in 2016, while effective precipitation increased by 8.4% in 2016 and decreased by 6.3% in 2015. Temperature conditions in 2015 was similar to the 30-year average, while those in 2016 were characterized by high temperatures.

There were remarkable variations in GDD, T_max, and EP during grain filling across eight environments (four sowing dates and two years), and the coefficient of variation was 12.1%, 11.5% and 42.0%, respectively ( Table 1 and Figure 5 ). GDD and ADT averaged across the four sowing dates were 6.7% and 7.1% higher in 2016 than in 2015, respectively. This difference was most evident for the SSM and ASJ sowing dates. The effective precipitation average across the four sowing dates was 14.7% higher in 2016 than in 2015, with the greatest difference being observed for the SSA sowing date, which was 116.4% higher in 2016 than in 2015. Throughout the corn grain filling stage, the SSJ experienced more days of high temperatures, particularly during the early to middle stages of grain filling compared to other sowing dates ( Figure 5 ).

In each year, grain yield (P< 0.01), kernel number per plant (P< 0.01) and 1000-kernel weight (P< 0.01) were affected by sowing dates ( Figure 6 ). In 2015, GY, KNP and TKW first increased and then decreased as sowing was delayed; while in 2016, GY, KNP, and TKW decreased gradually as a result of delayed sowing. In comparison with the ASJ, the GY increased by 21.1% in SSA, 17.9% in SSM and 9.4% in SSJ, respectively; the KNP increased by 12.0% in SSA, 9.2% in SSM and 4.4% in SSJ, respectively; the TKW increased by 13.3% in SSA, 15.6% in SSM and 8.1% in SSJ, respectively. Therefore, the earlier sowing, the higher GY and the higher KNP and TKW ( Figure 6 ).

The differences in GY for various sowing dates ranged from 1.4% to 19.8% in 2015, while the differences ranged from 5.1% to 24.4% in 2016. The average GY, KNP, and TKW across all sowing dates were 8.7%, 6.1%, and 7.4% lower in 2016 than in 2015. The SSM in 2016 encountered high temperatures, with the average daily maximum temperature reaching as high as 36.8°C on the 5th day after tasseling, which was 6.4% higher than the highest temperature recorded during corn tasseling period in 2015. As a result, GY and KNP were 12.4% and 9.5% lower in 2016 than in 2015, respectively.

The GY for the spring-, summer-, and autumn-sown corn were 4.9%, 11.6%, and 10.5% higher in 2015 than in 2016, respectively. KNP for the spring-, summer-, and autumn-sown corn were 4.0%, 7.3%, and 7.7% higher in 2015 than in 2016, respectively. The TKW for the spring-, summer-, and autumn-sown corn were 5.2%, 6.7%, and 14.2% higher in 2015 than in 2016, respectively.

The grain filling parameters derived from the logistic model showed significant differences among different sowing date treatments ( Table 2 ). On average, across the four sowing dates, the grain filling duration in a normal year 2015 was 2.97 days longer than in a high temperature year 2016, while the mean grain filling rate was 0.07% higher in 2015 than in 2016. Over the two-year study period, spring-sown corn had a longer active grain filling period, with a 1.40-day increase compared to summer-sown corn and a 3.47-day increase compared to autumn-sown corn. In contrast, summer-sown corn showed a 0.8% increase in mean grain filling rate compared to spring-sown corn, and a 5.6% increase compared to autumn-sown corn.

The grain filling process, measured by the change of kernel weight (KW) over days after pollination followed a logistic model pattern ( Figure 7 ). Significant differences in KW were observed among sowing dates during the early grain filling stage, and the mean grain filling rate was 14.2% lower in 2016 than in 2015 ( Figure 7 and Table 3 ). Significant differences in KW were observed among sowing dates during the middle grain filling stage, with the weight gain following the order: ASJ< SSJ< SSA< SSM. In 2015, the mean grain filling rate was 7.4%, 4.0%, and 1.2% higher for SSM than for SSA, SSJ, and ASJ, respectively. In 2016, the mean grain filling rate was 0.8%, 4.2%, and 12.7% higher for SSA than for SSM, SSJ, and ASJ, respectively. During the late grain filling stage, although the KW for all treatments gradually increased, the value for SSM was the largest and significantly higher than other sowing dates in 2015. KW was 4.0%, 8.1%, and 11.5% higher for SSM than for SSA, SSJ, and ASJ, respectively. The highest KW was achieved for SSA in 2016, with KW being 3.1%, 6.6%, and 27.6% higher for SSA than for SSM, SSJ, and ASJ, respectively.

A structural equation modeling analysis was conducted to examine the relationship between weather factors and grain filling parameters. It was found that specific weather factors had a significant impact on grain filling parameters ( Figure 8 ). SEM explained 88% of variation in kernel weight, 96% in average daily temperature, 86% in growing degree-days, 71% in grain filling duration and 24% in mean grain filling rate. The results showed that the grain filling duration and rates had direct positive effects on kernel weight with path coefficients of 0.42 and 0.52, respectively. However, effective precipitation had direct negative effects on kernel weight with a path coefficient of -0.18. The growing degree-days were strongly associated with the grain filling duration (path coefficient: 0.85) and indirectly affected the kernel weight through the grain filling duration with the product of path coefficients of 0.36. Furthermore, the average daily temperature was strongly associated with the mean grain filling rate (path coefficient: 0.49) and had indirect positive effects on kernel weight (product of path coefficient of 0.25), primarily through the grain filling rate.

High temperatures are predicted to become more intense and more frequent, potentially creating

disadvantageous conditions during the reproductive growth stage for crops such as corn, wheat, soybean, and rice, leading to a significant reduction in grain yield (Eyshi Rezaei et al., 2015; Gabaldón-Leal et al., 2016; Lobell et al., 2013; Mastilović et al., 2017; Siebers et al., 2015). Historical climate change trend from 1961 to 2015 resulted in a simulated yield decrease of 161.2 kg ha^-1 per decade, mostly due to a significant temperature increase of 0.43°C per decade (Zhang et al., 2019). Grain yield was highly correlated with kernel number and kernel weight (Mayer et al., 2014; An et al., 2018; Villegas et al., 2015). The average 1000-kernel weight and kernel number per plant across the four sowing dates was 8.0% and 6.5% higher in 2015 than in 2016, respectively, due to the high temperatures during grain filling in 2016. Mayer et al. also reported a similar observation that kernel weight experienced a rapid decline during the grain filling period under high temperature (Mayer et al., 2014). An early termination of grain filling under high temperature was not attributed to early leaf senescence resulting from the lack of assimilate, but rather to the abortion of kernels, resulting in reduced sink activity (Mayer et al., 2014). Consequently, the average grain yield was 9.5% higher in 2015 than in 2016 due to high temperature ripening.

The weather changes over the past forty-five years at the experimental site was consistent with findings from previous studies conducted in other regions of the NCP (Tao et al., 2016a; Wang et al., 2012; Xiao and Tao, 2016; Zhang et al., 2015). Our study demonstrated that the weather change trend during the corn growth period included increased temperature, decreased cumulative sunshine durations, and reduced effective precipitation ( Table 1 ). The historical average corn yield was 5249.9 kg ha^-1 in the NCP from 1981 to 2009 (Tao et al., 2016a). Compared with this 29-year average, the

corn yield increased by 92.2% in 2015 and by 75.5% in 2016. These yield increases may be attributed to the development of corn cultivars with greater stress tolerance (Ci et al., 2011; Gouache et al., 2012) and improvements of agronomic management practices (Chen et al., 2013), such as alteration of sowing dates. Adjusting sowing dates proved to be an effective strategy for examining the influence of various weather factors on crop yield (Tsimba et al., 2013). To mitigate the impact of weather variation on corn production, optimizing sowing date is critical to maximize the utilization of temperature, thermal and water resources, thus improving yields.

Studies showed that the prolonged heat stress during corn silking reduced kernel number per plant (KNP) (Ordóñez et al., 2015; Wang et al., 2019). Specifically, Wang et al. (2019) found that exposure to high temperature of 40°C significantly reduced KNP. Another study reported that corn kernels exposed to 35°C temperature stress for either 4 or 8 days showed different abortion rates: 97% for prolonged (8 days) stress and about 23% for short-term (4 days) stress (Cheikh and Jones, 1994). This suggests that extended high temperatures during silking lead to kernel abortion, whereas shorter heat stress periods have minimal impact on KNP. In our study, we observed a 3.5% decrease in KNP in SSM than in SSA in 2016. During the silking stage of SSM in 2016, effective precipitation and cumulative sunshine duration were recorded at 52.4 mm and 41.3 hours, respectively. The combined stress of waterlogging and shading during the silking stage inhibited the growth of ovaries and silks at the tip ear, resulting in a reduction in the number of emerged silks and total kernel number per plant (Zhou et al., 2021; Ren et al., 2023).

Temperature was a critical factor that influenced post-silking dry matter accumulation and grain yield (Zhou et al., 2016). Previous studies suggested that longer grain filling duration increase yield by promoting kernel dry matter accumulation (Liu et al., 2010; Zhu et al., 2018). The average active grain filling duration across the four sowing dates was 7.2% longer in 2015 than in 2016, while the mean grain filling rate was 0.7% higher in 2016 than in 2015. Our findings showed that the high temperatures during grain filling periods shortened the effective grain filling periods while increasing mean grain filling rates. However, the increased grain filling rate was insufficient to compensate for the yield losses caused by the shortened grain filling duration. As a result, high temperatures during grain filling shortened the active grain filling period and greatly reduced kernel weight and grain yield.

Given the harmful effects of high temperatures on crop yields, researchers suggested various adaptation strategies to mitigate the reduction in corn production. These strategies include adjusting sowing dates, selecting appropriate cultivars, and irrigation (Lv et al., 2020; Tack et al., 2017), while research showed that adjusting sowing date did not mitigate the heat stress around silking (Gao et al., 2021). Sowing date had varying effects on the weather conditions during the growth period, including changes in solar radiation and temperature (Ji et al., 2024), which can affect the process of grain filling of corn (Yan et al., 2017). Altering sowing dates may be an appropriate agronomic measure to avoid creating an unfavourable environment for crops (Opsi et al., 2013) and better adapt to changing climate conditions (Bonelli et al., 2016; Iizumi et al., 2014). Early sowing tended to minimize terminal high temperature (Paudel et al., 2023). Studies showed that delaying the sowing date led to a 1% decrease in grain yield per day, which was attributed to low temperatures during crop vegetative growth, elevated temperatures during the grain filling period, and a short duration of grain filling (Shah et al., 2020).

Our results indicated that growing degree-days had weak direct positive, but strong indirect influence on kernel weight through the grain filling duration. The early summer-sown corn increased growing degree-days by 11.4%, compared with the summer-sown corn. Early sowing could increase growing degree days and promote indirect effects on kernel weight through prolonging the active grain filling duration. The contribution of grain filling stage to the final kernel weight was in the following order: middle (57.7%), early (21.1%) and late (20.1%) stages. In 2016, the early summer-sown corn exhibited a 4.1% increase in early grain filling duration, a 3.4% increase in grain filling rate for the middle grain filling duration and a 3.6% increase in grain filling rate for the late grain filling duration compared to the summer-sown corn. Thus, the early summer-sown corn increased kernel weight by prolonging the grain filling duration, especially during the early grain filling stage, and increasing grain filling rate for the middle and late grain filling durations.

SEM was used as an effective method to identify both direct and indirect relationships among various variables (Grace, 2006). In the current study, the grain filling duration and mean grain filling rates had direct and positive effects on kernel weight, which is consistent with previous research suggesting that variations in final kernel weight were associated with changes in kernel growth rate and the active grain filling period (Hatfield, 2014). Research showed that precipitation and temperature are major limiting factors of crop production in Inner Mongolia (Li et al., 2022). In our study, effective precipitation had direct and significant negative effects on kernel weight, likely due to waterlogging and shading during silking, which results in low assimilation availability, poor pollination, and reduced kernel number per ear (Gao et al., 2018). As corn was irrigated and no water stress was observed during growing season, excessive precipitation could potentially reduce grain yield. On the other hand, the impact of sunshine duration on kernel weight was minimal, as it could meet the demands of the late grain filling stage and was not a limiting factor for the final kernel weight (Hu et al., 2019).

A previous study showed that effective grain filling period was correlated with growing degree-days, and kernel growth rate was influenced by the average daily temperature, with kernel weight strongly correlated with growing degree-days (Zhou et al., 2017). In our study, growing degree-days had weak and direct positive effects on kernel weight (path coefficient>0.1, but P>0.5), while it had strong indirect effects on kernel weight through the grain filling duration. In addition, the average daily temperature had weak direct negative effects but a significant indirect positive influence on kernel weight through mean grain filling rate. The findings indicated that planting corn earlier could increase growing degree-days and higher average daily temperature, thereby strengthening the indirect effects on kernel weight through the grain filling duration and mean grain filling rate.