A long-term, analysis-ready dataset of rice phenology in middle and lower reaches of the Yangtze River Plain of China from 1981 to 2024

Abstract

Background

Quantifying the impact of climate change on rice production, a cornerstone of global food security, requires long-term, spatially explicit, and readily analyzable phenological data. While the Middle and Lower Yangtze River Plain (MLYRP) is China's primary rice bowl, a critical gap persists: the lack of a publicly available, multi-decadal, station-based dataset that provides not only raw phenological dates but also curated statistical trends for the entire growth cycle, tailored for immediate reuse by the research community.

The dataset

To bridge this gap, we present “MLYRP-rice phenology data”, a foundational dataset characterizing rice phenology dynamics across the MLYRP from 1981 to 2024. Derived from rigorously quality-controlled records of 95 national agrometeorological stations, this dataset is architected for reuse, offering two synergistic products: (1) A visualization collection comprising 2,432 high-resolution graphs that intuitively depict the probability distribution and temporal trends of 11 phenological stages and 3 growth phase lengths via kernel density estimation and linear regression. (2) A machine-actionable statistical core, structured as an Excel spreadsheet, which provides the complete results of the linear trend analysis (including slope, intercept, R², F-value, and p-value) for every station and variable.

Value and potential for reuse

This dataset transcends a mere data compilation by delivering an analysis-ready product. It is designed as a community benchmark to directly facilitate crop model validation, climate-impact meta-analyses, remote sensing product verification, and data-driven agricultural adaptation planning. By providing pre-computed, statistically robust trends, “MLYRP-rice phenology data” establishes an indispensable empirical foundation for detecting and attributing climate impacts on one of the world's most critical agricultural systems.

Availability

The dataset is open-access and available at https://doi.org/10.57760/sciencedb.25297.

1. Introduction

Rice serves as a staple food for more than 50% of the global population and is fundamental to food security, particularly in Asia [1]. However, its productivity is highly vulnerable to climate change [2]. Crop phenology—the timing of key developmental stages—serves as a primary integrator of environmental conditions, directly governing photosynthetic capacity, biomass accumulation, and ultimate yield formation [3]. Consequently, long-term, high-fidelity phenological records are not merely observational data; they are essential empirical constraints for validating and improving the next generation of Earth System Models (ESMs) and process-based crop models. They provide the "fingerprint'' of past system behavior against which model skill is measured and future projections are calibrated, serving as a critical tool to reduce the "phenological uncertainty'' in climate impact assessments.

Phenology represents one of the most sensitive biological indicators of crop responses to climatic variations. The past four decades has witnessed accelerated global warming and increasing frequency of extreme weather events [2,4,5,6]. Examining long-term trends in the progression of rice phenological stages and its duration during this period enables mechanistic quantification of climatic drivers (including temperature, precipitation, and solar radiation) on crop phenology, thereby establishing an empirical foundation for comprehensive impact assessments of climate change on agricultural systems [6]. Furthermore, analyzing long-term variations in rice phenology and its duration facilitates optimization of agricultural production zoning and varietal adaptation. This provides data-driven support for redefining cultivation suitability domains, breeding stress-resilient cultivars with climate-synchronized growth cycles, and enhancing resource-use efficiency [7]. The integration of multi-decadal crop phenology datasets enables construction of “pheno-climatic-yield” nexus models, which project risks of phenological phase shifts under future climate scenarios. Such predictive capacity allows proactive formulation of adaptive interventions (e.g., optimized sowing windows and precision field management), thereby generating targeted policy solutions for smart agriculture and ensuring sustained yield enhancement.

In China, rice cultivation spans vast areas, divided into 6 major rice cultivation regions and 16 sub-regions [8,9]. The Middle and Lower Yangtze River Plain (MLYRP) accounts for over one-third of China's total rice cultivation area [10]. Through the synergistic integration of natural endowments, historical legacy, and technological innovations, this region has evolved into not merely China's "strategic agricultural nexus" but also serves as a living heritage of global paddy civilization and a critical paradigm for sustainable agricultural intensification [11]. High rice yield acts as a cornerstone for food security in China and globally [12], holding significant strategic importance for ensuring national food security. Changes in rice developmental stages directly impact photosynthesis, thereby influencing grain quality and yield [13]. Studying these developmental shifts is critical for rational agricultural planning, growth regulation, prevention of agricultural meteorological disasters, and yield prediction [3]. Variations in crop developmental stages—such as earlier or delayed phenology, shortened or extended growth periods—affect biomass accumulation time, ultimately threatening crop productivity and food security [[14], [15], [16]].

Understanding climate-phenology interactions here has profound implications for regional and global food security. Despite its importance, the research community has been hampered by a persistent data bottleneck. Existing phenological resources are often characterized by one or more critical limitations: (i) temporal fragmentation, lacking the continuous, >40-year span needed to discern robust climate-driven trends; (ii) phenological incompleteness, focusing on a few key stages (e.g., heading, maturity) while omitting the full developmental sequence from sowing to maturity required for comprehensive model calibration.

Our dataset, spanning 95 stations over 44 years with full statistical outputs, is engineered specifically to dismantle this barrier. By providing pre-computed, machine-readable trend statistics for every station and phenological stage, we address the need for analysis-ready data products that enable immediate reuse without extensive preprocessing. We move beyond presenting raw observations to delivering a curated, analysis-ready data product. Our primary innovation lies in bundling intuitive visualizations with a machine-readable repository of pre-computed statistical trends for every phenological metric at each station. This dual offering transforms the dataset from a static archive into a dynamic, foundational resource. It is purpose-built to serve as a community-wide benchmark for validating and improving process-based crop and land surface models, for conducting efficient large-scale meta-analyses, and for informing localized adaptation strategies with station-specific evidence. This paper details the dataset's compilation, rigorous quality control, structure, and its multifaceted potential for reuse across the Earth system sciences.

Unlike previous efforts, the primary innovation of this dataset lies in transforming raw observations into a readily analyzable data product. This is achieved through two key offerings: 1) a curated collection of high-resolution visualizations depicting spatiotemporal patterns, and 2) a machine-readable Excel table containing all linear trend parameters, which elevates the dataset from a mere collection of figures to a foundational resource for quantitative analysis. The outcomes establish a robust theoretical framework for understanding climate-crop interactions while providing empirical support for optimizing cultivation management strategies. This investigation delivers a dataset that transcends previous efforts by offering an unprecedented, ready-to-use resource of spatially explicit, data-driven insights, thereby directly addressing the data gap and empowering the research community with a foundational tool for climate adaptation research.

2. Data and Methods

2.1. Study region and agrometeorological station network

Our dataset includes the records collected from 1981 to 2024 of 95 agrometeorological observation stations in the MLYRP, including Jiangsu, Zhejiang, Anhui, Jiangxi, Hubei, Hunan provinces, and Shanghai Municipalities. These stations were maintained by the Chinese Meteorological Administration to investigate local crop production and agricultural meteorological disasters, as well as provide suggestions for local farmers to manage agricultural meteorological disasters. There were 1 station in Shanghai Municipalities, 8 stations in Jiangsu province, 15 stations in Zhejiang province, 11 stations in Anhui province, 16 stations in Jiangxi province, 21 stations in Hubei province, 23 stations in Hunan province (Fig. 1).

Fig. 1.

Spatial distribution of 95 agrometeorological observation stations included in the dataset.

All stations are located in a rice primary production region, and represent the typical cropping system in the MLYRP. They are geographically and climatologically different, and have good records of crop data during 1981–2024. These stations provide a unique, multi-decadal record of standardized observations, offering an unparalleled opportunity to analyze long term phenological trends.

2.2. Dataset construction workflow

The dataset construction followed a structured pipeline (Fig. 2) designed to ensure robustness and reproducibility:

Fig. 2.

Flowchart of the dataset construction process for rice phenology in the MLYRP.

(1) Data collection: Historical data and electronic data were collected from official archives and official website of Chinese Meteorological Administration. The detailed phenology records from 1981 to 2024 included the dates of rice sowing (SO), emergence (EM), trefoil (TS), transplanting (TR), regreening (RE), tillering (TI), jointing (JO), booting (BO), heading (HE), milking (MI) and maturity (MA).

(2) Data processing quality control: A multi-tiered quality control protocol was implemented:

Firstly, the collected data are categorized by year, site, crop, and phenology.

Then, A multi-step quality control protocol was implemented to identify and handle potential anomalies and outliers:

Systemic Factor Screening: Station metadata and historical records were consulted to identify periods affected by non-climatic factors (e.g., station relocations, changes in observation protocols). Data from these periods were flagged for expert validation.

Statistical Thresholding: For each phenological stage at each individual station, observations falling beyond three standard deviations (±3σ) from the long-term mean were automatically flagged as potential outliers.

Expert Validation: Automatically flagged values were not removed arbitrarily. Instead, they were rigorously examined by agrometeorological experts. This involved cross-referencing with adjacent stations to check for spatial coherence (suggesting a climatic extreme) and consulting original logbooks for notations. Only values conclusively deemed erroneous after this contextual analysis were corrected or excluded, preserving legitimate climatic extremes.

Finally, the validated dates were converted to day of year (DOY), calculation of growth period length (GLs), and statistical analysis (e.g., mean and standard deviation computation).

We assigned January 1 of each calendar year a reference value of 1, subsequently transforming the date of the universal period of each phenology period to the day of year (DOY). According to the rice growth and development characteristics of each stage, the whole growth period length (GLs) was divided into 3 stages, namely, sowing to jointing (SO-JO), jointing to maturity (JO-MA), and sowing to maturity (SO-MA).

(3) Data analysis and product generation

Kernel density estimation: We applied kernel density estimation (KDE), a nonparametric probability density estimation technique, to analyze rice phenological records across agrometeorological observation stations. The longitudinal dataset (44 years) was subjected to comprehensive statistical characterization, including computation of central tendency measures (mean) and dispersion parameters (standard deviation). These analytical outputs-integrating KDE-derived density functions with descriptive statistics-formed the foundation for generating scientifically rigorous visual representations in our dataset compilation. This methodology enabled robust estimation of population characteristics from sample observations, outperforming conventional histogram-based methods through its continuous density estimation independent of arbitrary bin-width selection [17,18]. The basic expression of KDE function is:

$$f\left(x\right)=\frac{1}{nh}\sum _{i=1}^{n}K\left(\frac{x-x_i}{h}\right)$$ (1)

$$h=1.059·\delta ·n^{-\frac{1}{5}}$$ (2)

where $x_1,x_2,\dots ,x_n$ are the DOY samples for a given phenological stage at a station; $n$ is the sample size; $K$ is the kernel function; $h$is the bandwidth; $\delta$ is the standard deviation; and $f\left(x\right)$ is the estimated probability density of DOY $x$. For the final published KDE products, $h$was automatically selected using Scott’s rule, the default in OriginLab 2024. Silverman’s rule of thumb was used only in the sensitivity analysis reported in Section 3.2 and was not used to generate the published figures.

Univariate linear regression: Phenology trend analysis was conducted through first-order linear regression modeling, with temporal patterns in DOY and GLs quantified using slope coefficients. This method was selected to provide a clear, robust, and interpretable estimate of the overall net change over the 44-year study period, which is a community-standard approach for establishing a baseline understanding in climatological and phenological studies. The regression parameters were estimated via ordinary least squares (OLS) optimization, expressed as:

$$Y=kx+b$$ (3)

Where Y denotes either DOY or GLs measurements, and x represents calendar year; b represents the y-intercept; k represents the annual rate of change. Decadal trends were derived as 10×k, with negative coefficients indicating phenology date advancement/GLs contraction and positive values signifying phenology date delay or GLs extension. Two tailed t-test analysis was used to determine the statistical significance of the change trend. The significance level for all regression analyses was set at α = 0.05. The regression analysis revealed p-values < 0.01 for key variables.

2.2.1. Final product curation

Kernel density estimation (KDE) and linear regression were systematically executed in OriginPro (Version 2024). The primary purpose of these analyses was to generate the standardized graphical and numerical outputs that constitute the core of this dataset. The outputs—2,432 high-resolution KDE and trend graphs (JPEG) and the comprehensive statistical spreadsheet (XLSX)—were organized into a structured repository.

All resulting statistical parameters from the linear regression (including province, station ID, station name, crop types, phenology period, slope (variation trends), intercept, R², model mean square, F-value, and p-value) for every phenological stage and growth period length at each station were compiled into a comprehensive Excel spreadsheet, forming a key part of the published data product.

2.3. Innovation in data curation framework

Our data curation methodology introduces three key innovations that distinguish this dataset from previous efforts:

Computational Reproducibility Framework: All analytical procedures—from quality control to statistical derivation—are implemented through standardized computational workflows in OriginPro, ensuring complete methodological transparency and reproducibility.

Multi-Scale Validation Protocol: We implement a novel validation approach combining (1) automated statistical outlier detection (±3σ thresholds), (2) spatial coherence analysis across adjacent stations, and (3) expert domain knowledge integration, creating a robust quality assurance framework for long-term observational data.

FAIR Compliance Implementation: The dataset architecture explicitly embeds FAIR principles through persistent identifiers (DOI), rich metadata schemas, machine-readable statistical derivatives, and open-access licensing (CC-BY), enabling both human and machine actionable data reuse.

3. Technical Validation

To ensure the dataset meets the highest standards of reliability and fitness for reuse, we implemented a comprehensive, multi-layered validation framework. This framework assesses the entire data pipeline, from the foundational raw observations to the final derived data products (statistical trends and visualizations).

3.1. Foundational data integrity and quality control

The integrity of this dataset is rooted in China's nationally standardized agricultural meteorological observation network, which provides the underlying raw data.

• •Observational Standardization: All phenological data were collected in strict adherence to the national "Specification for Agrometeorological Observation'' [19]. Observations were generally conducted every two days during the growing season, with daily observations during critical stages such as heading and flowering to ensure accurate recording of these short-duration events. This process was carried out by a professional workforce of over 100 certified technicians, 82% of whom possess more than a decade of field experience, ensuring high consistency and expert judgment across the network.
• •
  Systematic Quality Assurance Protocol: A rigorous, multi-step quality control (QC) pipeline was applied to the raw observational records:

  • ∗
    Automated Checks: Data underwent automated screening for completeness, duplication, valid station IDs, temporal coherence, and value range plausibility.
  • ∗
    Expert-Driven Validation: Potential outliers flagged by statistical thresholds (e.g., values beyond ±3σ from the long-term mean) were not automatically discarded. Instead, they underwent expert arbitration, which involved cross-referencing with adjacent stations for spatial coherence and consulting original logbooks to distinguish genuine climatic extremes from errors.
  • ∗
    Gap-filling and Correction: Anomalies and missing values were addressed using a hierarchical protocol. First, all flagged records were checked against original paper logbooks and official electronic archives. Second, for isolated missing values at a station, the value was estimated from the same station’s adjacent years (previous and following years) when the cropping system and cultivar group were unchanged; for phenological dates, the mean of the nearest available years was used. Third, for consecutive missing years, we referred to neighboring stations with the same rice cropping system and similar agroclimatic conditions to evaluate spatial coherence, and the final value was determined by expert agronomic judgment. Fourth, outliers beyond ±3σ were not automatically removed; they were retained as genuine climatic extremes when supported by neighboring stations and logbooks, and were corrected or excluded only when conclusively identified as errors. Missing values were rare; the exact proportion varied slightly among stations and phenological stages, but remained negligible. Sensitivity checks indicated that the filled values did not materially change the phenological trends.

A summary of the standardized phenological criteria used for observation is provided in Table 1, which formed the basis for consistent data collection.

Table 1.

Basis for determining rice phenology.

Phenology	Observation basis
Sowing (SO)	Actual sowing date.
Emergence (EM)	Emergence of the first prophyll from the coleoptile.
Trefoil (TS)	The third fully expanded true leaf emerges from the leaf sheath of the second true leaf
Transplanting (TR)	Actual transplanting date.
Regreening (RE)	After transplanting, the leaves turn green, the heart leaves re-expand or new leaves emerge (with water droplets appearing at the leaf tips in the morning). When gently lifting the plant by hand, resistance is felt, indicating that the roots have penetrated the soil and the plant has reached the regreening period standard.
Tillering (TI)	The leaf tips of newly formed tillers emerge from the leaf sheath, with the exposed portion measuring approximately 0.5-1.0 cm in length.
Jointing (JO)	The elongation of basal stem nodes marks the beginning of jointing, forming a stem with distinct nodes. The jointing height (measured from the uppermost root node) is 1.0 cm for early rice, 1.5 cm for medium rice, and 2.0 cm for late rice. In early rice, panicle differentiation initiates before jointing with first internode elongation; medium rice starts panicle differentiation synchronously with jointing, characterized by first internode stabilization and second internode elongation; late rice begins panicle differentiation post-jointing, exhibiting stabilized first and second internodes with third internode elongation.
Booting (BO)	The flag leaf is completely emerged from the leaf sheath, with the flag leaf collar becoming visible.
Heading (HE)	The panicle tip emerges from the flag leaf sheath. Some panicles protrude in a curved form from the sheath. When such curved emergence occurs extensively, it may be attributed to meteorological influences and should be duly annotated.
Milking (MI)	The grains at the panicle tip have attained normal size, with husks filled with milky fluid, exhibiting green coloration.
Maturity (MA)	Over 80% of grains on indica rice panicles and over 90% on japonica rice panicles exhibit their characteristic varietal coloration, meeting the physiological maturity standard.

3.2. Validation of derived data products

The robustness of the core data products—the linear trend statistics and kernel density estimation (KDE) visualizations—was rigorously tested.

• •
  Validation of Trend Statistics (Linear Regression): The ordinary least squares (OLS) linear regression used to generate decadal trends was subjected to two key validation steps:

  • ∗
    Residual Analysis: An analysis of residuals confirmed that the key OLS assumptions (normality, independence, and homoscedasticity) were not substantially violated, supporting the validity of the fitted models.
  • ∗
    Robustness to Outliers (Leave-One-Out Cross-Validation): To ensure the derived trends were not unduly influenced by individual extreme years, we employed a leave-one-out cross-validation (LOOCV) approach. This process systematically recalculated the linear trend after excluding each year from the 44-year series. The results demonstrated exceptional stability, with slope coefficients varying by less than ±0.1 days per decade across all iterations, confirming the robustness of the reported trends.
• •Sensitivity Analysis of Phenological Distributions (KDE): The KDE plots, which visualize the probability distribution of phenological timing, are sensitive to the selection of bandwidth. For the final published KDE products, Scott’s rule was used. To verify that the observed patterns were genuine features of the data and not artifacts of the algorithm, we conducted a sensitivity analysis. We recomputed all densities using both Scott’s rule and Silverman’s rule of thumb. The resulting probability density functions showed consistent modal peaks and distribution shapes across both methods, confirming the reliability of the visualized patterns. The final published KDE figures and statistics are based on Scott’s rule.

3.3. Intrinsic data quality and uncertainty assessment

• •
  Uncertainty Sources and Mitigation: We acknowledge and have mitigated several potential sources of observational uncertainty:

  • ∗
    Human Judgment: Minimized through standardized training and expert arbitration.
  • ∗
    Environmental Interference: Addressed via expert annotation and contextual validation.
  • ∗
    Instrumental Limitations: Mitigated by tool calibration and the use of relative, observer-defined phenological criteria.
  • ∗
    Historical Data Gaps: Filled using statistically and experientially informed imputation methods.
• •Intrinsic Consistency as a Quality Metric: The statistical properties of the final dataset itself provide compelling evidence of its high quality. The widespread high statistical significance (p < 0.01 for 78% of phenological stages) and substantial explanatory power (R² > 0.5 for key developmental transitions like transplanting and tillering) observed across the extensive station network are indicative of an internally consistent and climatically sensitive dataset. Such strong, coherent signals would be unlikely to arise from a noisy or unreliable data source.

In summary, this comprehensive technical validation—spanning foundational QC, product robustness checks, and intrinsic consistency assessment—affirms that the spatiotemporal patterns and statistical derivatives presented in this dataset are robust and statistically sound. This provides end-users with a high-confidence foundation for subsequent scientific analysis and decision-making.

4. Data Composition

This curated repository contains two main data products:

• (1)Visualization Collection (∼2.82 GB): A curated set of 2,432 georeferenced JPEG images (1200 dpi) providing KDE plots and linear trend fits for all 14 variables (11 phenology dates + 3 growth lengths) at all 95 stations.
• (2)Statistical Core (∼1 MB): A machine-readable Microsoft Excel file (MLYRP_Rice_Phenology.xlsx) containing the complete results of the univariate linear regression analysis. The spreadsheet is organized with separate sheets for phenology dates and growth lengths, with columns for: Province, Station ID, Station Name, Crop Type, Phenology Period, Slope (change/year), Intercept, R², Model Mean Square, F-value, and P-value.

No field photographs or manually captured images were used. All graphical products were generated programmatically in OriginPro 2024 from the quality-controlled DOY and GLs data. KDE plots were produced using Scott’s rule; linear trend plots were produced using ordinary least squares regression. A uniform template was applied to all figures, and each graph was exported as a 1200 dpi JPEG without additional manual editing. The storage structure of the dataset is shown in Fig. 3. The first level folder is named "MLYRP-rice phenology data'', which includes secondary folders named after 7 provinces or municipalities and Excel spreadsheet. Each secondary folders contains a tertiary folder named after observation station and single-cropping rice or double-cropping late rice. Each tertiary folders contain 16 jpg diagrams, including 1 probability density diagrams of phenology, 11 linear fitting diagrams of DOY of phenology, 1 probability density diagrams of growth period length and 3 linear fitting diagrams of growth period length.

Fig. 3.

The storage structure of the dataset.

Representative data products are demonstrated in Fig. 4, showcasing the integrated visualization framework.

Fig. 4.

Example dataset: variation characteristics of rice phenology and its lengths at lengshuitan agrometeorological observation station, Hunan Province.

Note: SO: sowing, EM: emergence, TS: trefoil, TR: transplanting, RE: regreening, TI: tillering, JO: jointing, BO: booting, HE: heading, MI: milking, MA: maturity.

Fig 4 showed that the kernel density plots indicated the maximum probability density for SO, EM, TS, TR, RE, TI, JO, BO, HE, MI, and MA occurred on day 90, day 95, day 106, day 120, day 125, day 135, day 151, day 166, day 174, day 187 and day 201, respectively. The GLs SO–JO, JO–MA, and SO–MA exhibited maximum probability density at 61 days, 50 days, and 111 days, respectively. The linear regression results revealed the following trends: SO, EM, TS, TR, RE, TI, JO, BO, HE, MI, and MA exhibit significant advancement with rates of −2.2 d decade⁻¹(R²=0.428, P<0.01), −2.9 d decade⁻¹(R²=0.526, P<0.01), −2.8 d decade⁻¹(R²=0.521, P<0.01), −4.5 d decade⁻¹(R²=0.694, P<0.01), −4.3 d decade⁻¹(R²=0.644, P<0.01), −6.0 d decade⁻¹(R²=0.730, P<0.01), −0.4 d decade⁻¹(R²=0.008, P<0.01), −2.9 d decade⁻¹(R²=0.375, P<0.01), −3.0 d decade⁻¹(R²=0.342, P<0.01), −1.3 d decade⁻¹(R²=0.090, P<0.01), and −2.1 d decade⁻¹(R²=0.202, P<0.01), respectively. SO–JO extended at rates of 2.3 d decade⁻¹(R²=0.206, P<0.01), JO–MA shorten at rates of −1.9 d decade⁻¹(R²=0.220, P<0.01), SO–MA showed no significant trends.

5. Dataset Values and Potential for Reuse

This dataset provides an analysis-ready resource of rice phenology trends across China's primary rice-producing region. By delivering pre-computed statistical derivatives alongside high-resolution visualizations, it lowers the barrier for diverse applications. Researchers can use the station-level trend statistics to benchmark crop and land surface models, conduct meta-analyses of climate impacts across geographical and management gradients, and validate long-term phenological trends derived from remote sensing products. The structured format enables efficient data extraction for large-scale synthesis, while the visualization collection facilitates intuitive exploration of spatiotemporal patterns. For policymakers and agricultural extension services, the dataset offers evidence-based insights for adaptation planning—such as optimizing planting calendars and selecting climate-adapted cultivars based on observed rates of phenological change. Educators can leverage the integrated framework as a teaching tool for climate impact assessment and data analysis. Ultimately, this dataset serves as a foundational community resource for advancing research on climate-crop interactions and supporting data-informed decisions to enhance agricultural resilience and food security.

6. Conclusion

This Data Note presents a transformative resource that advances agricultural data science through its FAIR-compliant architecture and computational readiness. By moving beyond traditional observational datasets to provide both raw data derivatives and analytical outputs, we establish a new paradigm for reusable agricultural data products. The dataset's innovation lies not only in its unprecedented temporal span and spatial coverage but also in its structural design that enables immediate computational utility. We anticipate this resource will serve as a critical benchmark for climate impact studies, crop model development, and adaptive management strategies, ultimately contributing to enhanced food security in a changing climate. Future work will focus on developing automated data integration pipelines that leverage this dataset's machine-readable structure for real-time climate adaptation decision support.

Funding

This work was supported by the National Key R&D Program of China under Grant [2024YFD2301301].

Generative AI Use Disclosure Statement

During the preparation of this work, the author(s) used DeepSeek (specifically, the DeepSeek Chat and DeepSeek-R1 tools) solely for the purpose of language polishing and proofreading. This tool was used to identify and correct grammatical errors, typos, and to improve sentence fluency. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.