A regional breeding strategy generation method based on corn trait big data

Abstract

The application discloses a regional breeding strategy generation method based on corn trait big data, and relates to the technical field of agricultural breeding, comprising the following steps: collecting and integrating multi-source heterogeneous corn regional test data, extracting agronomic trait data, and constructing a structured data set; according to the agricultural ecological environment characteristics and the climate zoning method, the main corn producing areas are divided into a preset number of ecological regions; a multi-dimensional data space-time analysis framework is constructed, and the structured data set is sequentially subjected to global descriptive statistical analysis, trait space-time evolution analysis, trait coordination relationship analysis, trait feature dimension reduction analysis, and ecological zoning and feature portrait analysis; and based on the space-time analysis results, customized corn breeding strategies for different ecological regions are generated. Through the innovative multi-dimensional space-time analysis framework, the space-time evolution law and complex relationship of corn traits are systematically analyzed, and scientific decision support is provided for regional breeding.

Description

Technical Field

[0001] This invention relates to the field of agricultural breeding technology, specifically to a method for generating regional breeding strategies based on big data on maize traits. Background Technology

[0002] Maize is one of China's important food crops, and its yield levels and agronomic traits show significant differences across different ecological regions. Developing targeted breeding programs is crucial for continuously improving maize yields and ensuring food security. However, current research and technology still have some limitations in maize breeding optimization, particularly in the comprehensiveness of data and the depth of analysis.

[0003] While existing agricultural management research involves big data analysis, particularly for monitoring and regulating environmental characteristics, these technologies primarily focus on real-time environmental control and health prediction. Their data sources, analytical objects, and objectives differ significantly from the needs of maize breeding optimization. Most existing methods have limited analytical scope, typically covering only a single ecological region or short-term datasets, and lack systematic research on the long-term spatiotemporal evolution of maize yield and agronomic traits across the country and multiple ecological regions. These limitations make it difficult for these methods to capture subtle long-term evolutionary trends, especially failing to reveal the synergistic and trade-off relationships among multiple traits, resulting in a lack of scientific basis for breeders' decisions regarding trait selection and breeding direction.

[0004] Furthermore, most current studies have failed to effectively utilize large-scale, long-term regional trial data to provide precise guidance for breeding decisions. The lack of a systematic, multi-dimensional analytical framework hinders in-depth analysis of the pathways and adaptive characteristics of maize variety improvement in different ecological regions. Consequently, existing breeding strategies often lack targeting and precision, resulting in relatively slow improvements in breeding efficiency. Moreover, they fail to effectively integrate macro-spatiotemporal evolution, micro-trait correlations, and regional variety structural characteristics, thus making it difficult to provide comprehensive quantitative support for regional breeding. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method for generating regional breeding strategies based on big data on maize traits, so as to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for generating regional breeding strategies based on big data on maize traits, comprising:

[0007] Collect and integrate multi-source heterogeneous maize regional experimental data, extract agronomic trait data, and construct a structured dataset;

[0008] Based on the characteristics of the agricultural ecological environment and climate zoning methods, the main corn-producing areas are divided into a predetermined number of ecological zones;

[0009] A multidimensional data spatiotemporal analysis framework is constructed, and global descriptive statistical analysis, trait spatiotemporal evolution analysis, trait synergy analysis, trait feature dimensionality reduction analysis, and ecological zoning and feature profiling analysis are performed sequentially on the structured dataset.

[0010] Based on the spatiotemporal analysis results, customized maize breeding strategies for different ecological regions are generated.

[0011] The present invention is further configured such that, based on the maize regional trial data, the extracted agronomic trait data include regional yield, growth period, plant height, ear length, thousand-grain weight, ear height, number of leaves at maturity, planting density, and number of ear rows;

[0012] The agronomic trait data are cleaned, denoised, missing value processed and standardized, and then fused according to preset key fields and stored in a relational database to form a structured dataset.

[0013] The present invention is further configured such that the ecological region includes seven major geographical regions: Northeast, North China, East China, Central China, South China, Southwest, and Northwest.

[0014] The present invention is further configured such that the global descriptive statistical analysis includes:

[0015] By reading the raw data file from the specified path and converting it into a data table for storage, the numerical agronomic trait column in the data is preprocessed, and entries that cannot be identified as numerical values ​​are converted into null values. All null values ​​are then filled with the arithmetic mean of the column.

[0016] Using preset statistical functions, a comprehensive statistical analysis is performed on all numerical agronomic trait columns to calculate global descriptive statistics for the structured dataset. These descriptive statistics include the total number of samples, arithmetic mean, standard deviation, minimum value, quantile value, and maximum value.

[0017] The data are grouped according to the preset geographic region classification variables, and box plots are generated for different regions for each agronomic trait to compare the differences in median, interquartile range and outlier distribution of each region for that agronomic trait.

[0018] The present invention is further configured such that the spatiotemporal evolution analysis of the trait includes:

[0019] For all agronomic trait columns in a structured dataset, fill missing values ​​with the arithmetic mean of all non-empty values ​​in that column;

[0020] Using provinces as the basic geographical unit, the dataset is divided into multiple data subsets. In each data subset, a univariate linear regression model is constructed with year as the independent variable and a single agronomic trait as the dependent variable. It is verified whether the year records in each data subset cover more than one year and contain data points from at least two different years. If the conditions are not met, the calculation for that data subset is terminated. If the conditions are met, a univariate linear regression model is established based on the year data, and the regression slope is calculated.

[0021] The relative annual rate of change of each agronomic trait in each province is calculated based on the regression slope and the arithmetic mean of the individual agronomic trait in the corresponding provincial data subset. If the arithmetic mean of the agronomic trait in the provincial data subset is zero, the relative annual rate of change is considered invalid.

[0022] The present invention is further configured such that the trait synergy analysis includes...

[0023] For all agronomic trait columns to be analyzed in the structured dataset, use the median of all non-empty values ​​in that column to fill in missing values;

[0024] The data is divided into corresponding time intervals by year using a preset time window. Within each time interval, the data is grouped according to province and variety name, and the arithmetic mean of all agronomic traits in each group is calculated to generate an aggregated dataset.

[0025] Provinces are mapped to geographic regions. Within each geographic region, with regional yield per acre as the target trait, Pearson correlation coefficients and significance p-values ​​between other agronomic traits and the target trait are calculated based on the aggregated dataset.

[0026] The Benjamini-Hochberg method was used to perform multiple hypothesis testing to correct the significance p-values ​​in order to control the false positive rate. The significance level was determined based on the corrected p-values, and significance was marked accordingly.

[0027] The present invention is further configured such that the trait feature dimensionality reduction analysis includes:

[0028] The data is grouped according to geographic regions, and partitioned principal component analysis is performed independently on the subset of data for each geographic region.

[0029] For data within each geographic region, Z-Score standardization is performed to ensure that all agronomic traits are listed with a uniform standard within that geographic region.

[0030] The covariance matrix or correlation coefficient matrix is ​​calculated based on the standardized agronomic trait data. The loading matrix is ​​generated by eigenvalue decomposition. The principal component score matrix is ​​obtained by multiplying the standardized agronomic trait data with the loading matrix.

[0031] Extract the score of each data sample on a preset number of principal components, and extract the loading values ​​of each original agronomic trait on a preset number of principal components. The absolute value of the loading reflects the degree of contribution of the agronomic trait to the principal components, and the sign of the loading indicates the direction of contribution.

[0032] The present invention is further configured such that the ecological zoning and feature profiling analysis includes:

[0033] Calculate the arithmetic mean of all agronomic traits in each geographic region, generate regional feature vectors, and perform Z-score normalization on the resulting regional feature vector matrix. Then, perform agglomerative hierarchical clustering based on Euclidean distance metric and Ward variance minimization join method to generate a hierarchical clustering dendrogram of agronomic traits.

[0034] Z-Score standardization was performed on the agronomic traits of all data samples. K-Means clustering was iteratively performed within the preset number of clusters. The optimal number of clusters was determined based on the principle of maximizing the silhouette coefficient. Based on the optimal number of clusters, K-Means clustering was performed to extract cluster centers as a universal trait combination pattern.

[0035] The silhouette coefficient is obtained by calculating the ratio of intra-cluster compactness to inter-cluster separation of the data sample.

[0036] Calculate the Euclidean distance between the feature vectors of each region and the trait combination patterns, and determine the pattern corresponding to the smallest Euclidean distance as the dominant trait combination pattern of the region.

[0037] The region feature vector values ​​and cluster center values ​​are normalized to a preset numerical range through linear mapping, thereby realizing dimensionless visualization of multidimensional features.

[0038] The present invention is further configured such that the customized maize breeding strategy formulates personalized breeding objectives for each region based on the evolution patterns of maize traits in each ecological region, the correlation between key traits and yield, and the synergistic effects of traits.

[0039] This invention provides a method for generating regional breeding strategies based on big data on maize traits. It involves collecting and integrating multi-source heterogeneous maize regional experimental data, extracting agronomic trait data, and constructing a structured dataset. Based on agricultural ecological environment characteristics and climate zoning methods, the main maize-producing areas are divided into a predetermined number of ecological regions. A multi-dimensional spatiotemporal analysis framework is constructed, and the structured dataset undergoes global descriptive statistical analysis, spatiotemporal evolution analysis of traits, trait synergy analysis, trait feature dimensionality reduction analysis, and ecological zoning and feature profiling analysis in sequence. Based on the spatiotemporal analysis results, customized maize breeding strategies for different ecological regions are generated. The beneficial effects include:

[0040] 1. Systematic Insights and Cross-Regional Analysis: Through the innovative Multidimensional Spatiotemporal Analysis Framework (MDSTAF), this invention breaks through the limitations of traditional research, realizing nationwide, long-term, and multidimensional analysis of maize yield and agronomic traits, providing systematic insights that combine macro and micro perspectives;

[0041] 2. Revealing complex relationships and intrinsic driving forces of traits: By employing an optimized PCA method, this invention can deeply explore and reveal the non-obvious covariation patterns and multi-trait structures of maize traits, providing a more scientific basis for variety improvement and future trend prediction.

[0042] 3. Optimize breeding decisions and resource allocation: Through a comprehensive understanding of the complex relationships between multiple agronomic traits, this invention provides breeders with more scientific multi-objective trade-off decision support. At the same time, it enables insights into regional variety structure through cluster analysis, helping breeders to more accurately formulate breeding goals and strategies for each region, thereby improving breeding efficiency and R&D results.

[0043] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0045] Figure 1 A flowchart illustrating a regional breeding strategy generation method based on big data of maize traits, as shown in an exemplary embodiment of the present invention;

[0046] Figure 2 Violin-box line distribution of maize agronomic traits within an ecological region;

[0047] Figure 3 Pearson correlation coefficient heatmap of maize agronomic traits and yield in ecological regions;

[0048] Figure 4 Principal component analysis (PCA) plot of maize agronomic traits and yield in ecological regions;

[0049] Figure 5 A hierarchical clustering dendrogram of maize agronomic traits in an ecological region;

[0050] Figure 6Radar chart comparing agronomic traits of maize varieties in different ecological regions. Detailed Implementation

[0051] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0052] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0053] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0054] A method for generating regional breeding strategies based on big data on maize traits, such as Figure 1 As shown, it includes:

[0055] Collect and integrate multi-source heterogeneous maize regional experimental data, extract agronomic trait data, and construct a structured dataset;

[0056] Based on the characteristics of the agricultural ecological environment and climate zoning methods, the main corn-producing areas are divided into a predetermined number of ecological zones;

[0057] A multidimensional data spatiotemporal analysis framework is constructed, and global descriptive statistical analysis, trait spatiotemporal evolution analysis, trait synergy analysis, trait feature dimensionality reduction analysis, and ecological zoning and feature profiling analysis are performed sequentially on the structured dataset.

[0058] Based on the spatiotemporal analysis results, customized maize breeding strategies for different ecological regions are generated.

[0059] The present invention is further configured such that, based on the maize regional trial data, the extracted agronomic trait data include regional yield, growth period, plant height, ear length, thousand-grain weight, ear height, number of leaves at maturity, planting density, and number of ear rows;

[0060] Agronomic trait data were cleaned, denoised, had missing values ​​removed, and standardized. The data was then fused according to preset key fields and stored in a relational database to form a structured dataset. Specifically, the data source for this invention is long-term, multi-ecological-regional maize regional trial data provided by the Seed Management Bureau of the Ministry of Agriculture and Rural Affairs of China, covering 25 provinces and autonomous regions in China, spanning from 1995 to 2024. This data reflects maize growth performance under different climate, soil, and irrigation conditions, ensuring the breadth and representativeness of the data. By extracting agronomic trait data from the original experimental dataset, including regional yield (RY), growth period (GP), plant height (PH), ear length (EL), thousand-grain weight (TGW), ear height (EH), number of leaves at maturity (NL), planting density (PD), and number of rows per ear (NR), important basis is provided for maize breeding decisions. The data is processed using Python. Tools such as Pandas, NumPy, and OpenPyXL were used to clean, denoise, imputate missing values, and standardize the data. Different data were merged using common fields, such as year, experimental site ID, and variety ID, to ensure data consistency and integrity. The processed data was stored in a relational database, such as MySQL, with a standardized database table structure designed to form a structured dataset with good queryability and analyzability. Through the above steps, a structured, queryable, and analyzable dataset was constructed. All data underwent preprocessing to ensure accuracy and consistency. This dataset meets the requirements of subsequent spatiotemporal analysis and trait synergy analysis, providing high-quality basic data support for the generation of subsequent breeding strategies and ensuring the scientific nature of the analysis process and the operability of the results.

[0061] The present invention is further configured such that the ecological regions include seven major geographical regions: Northeast, North China, East China, Central China, South China, Southwest, and Northwest. Specifically, based on sunlight, heat, water, soil characteristics, and topography, the main maize producing areas are preferably divided into seven major geographical regions: Northeast (NE), North China (NC), East China (EC), Central China (CC), South China (SC), Southwest China (SW), and Northwest China (NW). The maize growth characteristics of each major geographical region are affected by its climate and soil environment. Therefore, this division can effectively reflect the differences in maize planting adaptability in different ecological regions. To achieve automated mapping between experimental points and ecological regions, the present invention uses the Python programming language, combined with geospatial data processing and data analysis libraries such as GeoPandas and Pandas. In the process of automatically mapping experimental points to ecological regions, GeoPandas loads and processes geospatial data, and then Pandas accurately maps each experimental point to the corresponding ecological region based on the geographical coordinate information of the experimental point, thereby completing the regional division.

[0062] The present invention is further configured such that the global descriptive statistical analysis includes:

[0063] By reading the raw data file from the specified path and converting it into a data table for storage, the numerical agronomic trait column in the data is preprocessed, and entries that cannot be identified as numerical values ​​are converted into null values. All null values ​​are then filled with the arithmetic mean of the column.

[0064] Using preset statistical functions, a comprehensive statistical analysis is performed on all numerical agronomic trait columns to calculate global descriptive statistics for the structured dataset. These descriptive statistics include the total number of samples, arithmetic mean, standard deviation, minimum value, quantile value, and maximum value.

[0065] The data is grouped according to preset geographic region classification variables. Box plots are generated for each agronomic trait in different regions to compare the differences in median, interquartile range, and outlier distribution for that agronomic trait across regions. Specifically, to ensure data integrity and consistency, this step performs the following operations in sequence: reading the original data file from the specified path and loading it into memory, converting it into a structured data table; automatically identifying and converting non-numerical entries to null values ​​for the numerical agronomic trait columns in the data table; filling all null values ​​with the arithmetic mean of the corresponding agronomic trait column across the entire dataset to complete data imputation; calling preset statistical functions to perform a one-time global descriptive statistic calculation on all numerical agronomic trait columns in the data table, including total sample size, arithmetic mean, standard deviation, minimum value, quantile values, and maximum value, to assess the overall distribution characteristics, central tendency, and dispersion of the dataset. The quantile values ​​include the 25th quantile, median, and 75th quantile; grouping the data according to preset geographic region classification variables, generating box plots for each agronomic trait in different ecological regions, such as... Figure 2 As shown, by comparing the median, interquartile range and outlier distribution of trait values ​​in different geographical regions, the trait differences of each maize ecological region are intuitively displayed. Figure 2 This study presents the data distribution of nine agronomic traits within China's seven major maize ecological and geographical regions. It visually demonstrates the differences in various agronomic traits across different regions, helping analysts understand the characteristics of maize production in each region and providing a valid basis for subsequent optimization of breeding strategies.

[0066] The present invention is further configured such that the spatiotemporal evolution analysis of the trait includes:

[0067] For all agronomic trait columns in a structured dataset, fill missing values ​​with the arithmetic mean of all non-empty values ​​in that column;

[0068] Using provinces as the basic geographical unit, the dataset is divided into multiple data subsets. In each data subset, a univariate linear regression model is constructed with year as the independent variable and a single agronomic trait as the dependent variable. It is verified whether the year records in each data subset cover more than one year and contain data points from at least two different years. If the conditions are not met, the calculation for that data subset is terminated. If the conditions are met, a univariate linear regression model is established based on the year data, and the regression slope is calculated.

[0069] Based on the regression slope and the arithmetic mean of individual agronomic traits in the corresponding provincial data subsets, the relative annual change rate of each agronomic trait in each province is calculated. If the arithmetic mean of an agronomic trait in the provincial data subset is zero, the relative annual change rate is considered invalid. Specifically, in this embodiment, for all agronomic trait columns to be analyzed in the dataset, missing values ​​are filled using the arithmetic mean of all non-empty values ​​in that column to ensure data integrity. This ensures that all data columns are not affected by missing values ​​during spatiotemporal evolution analysis, thereby improving the reliability of the analysis results. The dataset is divided into multiple data subsets based on provinces as the basic geographical unit. For each data subset, the year is used as the independent variable. A certain agronomic trait as the dependent variable A univariate linear regression model was established to analyze the trend of this agronomic trait over time. For each data subset, a validity assessment was performed, checking whether the year records covered more than one year (i.e., the difference between the largest and smallest year was greater than or equal to 1) and ensuring that the data subset contained data points from at least two different years. If neither of these conditions was met, the analysis of that province's data subset was terminated. If both conditions were met, the regression model was continued, and the model was used if the validity conditions were satisfied. The regression slope is calculated using a univariate linear regression model in the form of "". regression slope This reflects the rate of change of the agronomic trait over the years; the regression slope is calculated. The ratio of the agronomic trait to the arithmetic mean of the data subset of the corresponding province yields the relative annual rate of change for each agronomic trait in each province. The formula for calculating the relative annual rate of change is as follows: ,in, This represents the relative annual rate of change. The regression slope, This is the arithmetic mean of the agronomic trait in the subset of data for that province; if the arithmetic mean of the trait in that province is zero, then: In this case, the relative annual rate of change is considered invalid, avoiding calculation errors caused by division by zero. By calculating the regression slope and relative annual rate of change using a regression model, the changing trends of each agronomic trait over time are quantified, eliminating the influence of trait dimensions and providing a unified standard for comparative analysis across traits and regions. Through spatiotemporal evolution analysis of maize's average annual yield and agronomic traits in various geographical regions, it was found that China's average maize yield has shown a continuous upward trend over the past 30 years, with significant regional differences. For example, the average annual maize yield growth was most significant in Xinjiang Uygur Autonomous Region and Beijing, with annual growth rates of 1.69% and 1.78% respectively, while the yield growth in Sichuan, Anhui, and other regions was relatively slow. The overall growth rate was stable, with an annual growth rate of approximately 0.98% to 0.99%. Agronomic traits also showed significant changing trends. Plant height increased in Beijing and Ningxia Hui Autonomous Region, with an annual growth rate of approximately 0.70% to 0.81%. Thousand-grain weight increased in Yunnan and Guizhou, with an annual growth rate of approximately 0.66% to 0.79%. The number of panicle rows increased in Beijing, Chongqing, and Guangxi Zhuang Autonomous Region, with an annual growth rate of approximately 0.56% to 0.62%. Planting density increased in northern provinces such as Shaanxi and Ningxia Hui Autonomous Region, with an annual growth rate of approximately 1.74% to 1.78%, while it decreased in southern provinces such as Guangdong, with an annual growth rate of approximately -0.82%. The overall growth period showed a slightly shorter trend, while the number of leaves did not change significantly.

[0070] The present invention is further configured such that the trait synergy analysis includes...

[0071] For all agronomic trait columns to be analyzed in the structured dataset, use the median of all non-empty values ​​in that column to fill in missing values;

[0072] The data is divided into corresponding time intervals by year using a preset time window. Within each time interval, the data is grouped according to province and variety name, and the arithmetic mean of all agronomic traits in each group is calculated to generate an aggregated dataset.

[0073] Provinces are mapped to geographic regions. Within each geographic region, with regional yield per acre as the target trait, Pearson correlation coefficients and significance p-values ​​between other agronomic traits and the target trait are calculated based on the aggregated dataset.

[0074] The Benjamini-Hochberg method was used to perform multiple hypothesis testing to correct for significance p-values, controlling the false positive rate. The significance level was determined based on the corrected p-values, and significance was then labeled. Specifically, trait synergy analysis was used to reflect the intrinsic association between maize yield and various agronomic traits in different ecological regions. Missing values ​​in all agronomic trait columns to be analyzed in the dataset were filled using the median of all non-empty values ​​in that column to enhance the data's robustness. To extract macro trends and smooth short-term fluctuations, spatiotemporal aggregation of the data was performed. Fixed time windows were preset to group the data into corresponding time intervals by year, such as 5 years, and within each time interval, data was further categorized by province and variety. The data is grouped by name, and the arithmetic mean of each agronomic trait within each group is calculated to generate an aggregated dataset. This aggregated dataset effectively represents the average performance of different regions and varieties within a stable time unit, laying the foundation for subsequent accurate calculation of the correlation between traits and yield. Based on preset geographical zoning rules, provinces are mapped to higher-level geographical regions, such as Northeast China and North China. This mapping helps identify differences in maize growth characteristics under different ecological environments. Within the data subset of each geographical region, using regional yield per acre as the target trait, all other agronomic traits are traversed, and the Pearson correlation coefficient between them is calculated to measure the strength of the linear relationship between the two variables. The formula for calculating the Pearson correlation coefficient is: ,in, The Pearson correlation coefficient is used. For the first The regional output per mu of a sample, For the first agronomic trait values ​​for each sample and These represent the regional regional yield value and the mean values ​​of agronomic traits, respectively. The sample size is given; within each ecological region, the Pearson correlation coefficient between yield and agronomic traits is calculated, and the results are as follows: Figure 3 As shown, thousand-grain weight was positively correlated with yield in all ecological regions, especially in the Northwest and Southwest regions, where the Pearson correlation coefficients were 0.45 and 0.58, respectively, showing the most significant correlation. Growth period and plant height had Pearson correlation coefficients of 0.56 and 0.57 in North China, and 0.55 and 0.54 in Northwest China, respectively, both significantly correlated with yield. The number of panicle rows had Pearson correlation coefficients of 0.29 and 0.28 in East and North China, respectively, indicating significant potential for yield improvement. Panicle length had Pearson correlation coefficients of -0.2 and -0.11 in South and Central China, respectively, showing a negative correlation with yield. Planting density had Pearson correlation coefficients of 0.51 and 0.45 in Southwest and Northwest China, respectively, significantly correlated with yield. To test the significance of the Pearson correlation coefficients, a null hypothesis was established... : This means that, assuming there is no significant linear relationship between the variables, the t-test statistic is used to determine whether the Pearson correlation coefficient significantly deviates from zero. The t-statistic is: ,in, To test the statistic, The degrees of freedom; based on the obtained test statistic Further calculate the significance p-value. The formula for calculating the significance p-value is: , The significance p-value; for The cumulative distribution function of the distribution is used to calculate the corresponding The probability of the value. express The cumulative distribution function of values ​​in absolute value The value at; for The absolute value of the value, because The distribution is symmetrical. We focus on the absolute value of t-values ​​to eliminate the influence of positive and negative directions, thereby calculating the corresponding significance. The value; and since the t-test is a two-tailed test, it focuses on the significance of the correlation in both positive and negative directions. Therefore, in calculating the significance... When calculating the value, it needs to be multiplied by 2 to account for both possible directions; significance. The value represents the probability of observing the current sample data or a more extreme outcome, assuming the null hypothesis holds. The smaller the value, the stronger the evidence for rejecting the null hypothesis and concluding that there is a linear correlation between the variables. Smaller values ​​indicate a stronger likelihood of a linear correlation. The value indicates that the greater the deviation of the data from the null hypothesis, the stronger the statistical basis for rejecting the null hypothesis. In the analysis of trait synergy, to control the risk of false positives caused by multiple hypothesis testing, the Benjamini-Hochberg method was used to analyze all paired comparisons. The value is used to control the false positive rate, and the adjustment is based on the corrected value. The significance of the value is determined, and the specific significance labeling rules are as follows: when the adjusted... When the value is 1, it is marked with "*" to indicate that it is statistically significant, and the null hypothesis that there is no linear correlation between the variables is rejected; when the adjusted value is 1, it is marked with "*". When the value is "**", it is marked as highly significant, indicating stronger evidence to reject the null hypothesis; when the adjusted value is "**", it is marked as "**". When the value is "***", it is marked as highly significant, indicating extremely strong evidence to reject the null hypothesis. To avoid misjudging small effects as statistically significant but lacking practical significance in large sample sizes, the effect size criterion proposed by Cohen (1988) is further introduced to classify the practical significance of the corrected significant correlation coefficients: when... When, it is a weak correlation; when When, it is moderately correlated; when When the correlation is strong, for example, the Pearson correlation coefficient between a certain agronomic trait and yield. And the adjusted The value is 0.003, which is statistically significant, but due to... The results meet the criteria for weak correlation, therefore they are not given priority in actual breeding processes. Through the above steps, this invention can systematically analyze and reveal the synergistic relationship between maize yield and various agronomic traits in different ecological regions, providing a scientific basis for subsequent breeding decisions.

[0075] The present invention is further configured such that the trait feature dimensionality reduction analysis includes:

[0076] The data is grouped according to geographic regions, and partitioned principal component analysis is performed independently on the subset of data for each geographic region.

[0077] For data within each geographic region, Z-Score standardization is performed to ensure that all agronomic traits are listed with a uniform standard within that geographic region.

[0078] The covariance matrix or correlation coefficient matrix is ​​calculated based on the standardized agronomic trait data. The loading matrix is ​​generated by eigenvalue decomposition. The principal component score matrix is ​​obtained by multiplying the standardized agronomic trait data with the loading matrix.

[0079] The scores of each data sample on a predetermined number of principal components are extracted, and the loading values ​​of each original agronomic trait on the predetermined number of principal components are also extracted. The absolute value of the loading reflects the contribution of the agronomic trait to the principal components, and the sign of the loading indicates the direction of contribution. Specifically, the trait feature dimensionality reduction analysis is achieved through partitioned principal component analysis, which aims to solve the collinearity problem among multiple agronomic traits and to refine the core dimensions that can characterize the comprehensive regional features. The structured dataset is grouped according to geographical regions, resulting in independent data subsets for each geographical region. For each geographical region's data subset, local principal component analysis is performed to ensure that the agronomic trait features of each region are effectively extracted. Based on the data subset of each geographical region, Z-Score standardization is performed on its contained agronomic trait columns to eliminate dimensional differences between agronomic traits. Based on the standardized agronomic trait data, a standardized agronomic trait data matrix is ​​constructed. Principal component analysis is performed on the agronomic trait data matrix by calculating its covariance matrix or correlation coefficient matrix, and obtaining eigenvalues ​​and corresponding eigenvectors through eigenvalue decomposition. The top eigenvalues ​​are selected in descending order. A principal component load matrix is ​​constructed. The standardized agronomic trait data matrix is ​​multiplied by the loading matrix to obtain the principal component score matrix for each sample. This maps the original high-dimensional agronomic trait features to a low-dimensional principal component space. The principal component score matrix represents the performance of each sample on each principal component. The formula for calculating the principal component score matrix is ​​as follows: , Principal component score matrix, For the standardized agronomic trait data matrix, For the front The loading matrix of each principal component; extract the loading matrix of each data sample in the first... Scores on principal components and extraction of each original agronomic trait from this prior... Loadings on each principal component reflect the contribution of each agronomic trait to the principal component. The larger the absolute value of the loading, the greater the contribution of that agronomic trait to the principal component. The formula for calculating the loading is: ,in, For the first The agronomic trait was in the first Loadings on each principal component; Let be the eigenvector components of the covariance matrix or correlation matrix, representing the th The agronomic trait was in the first The components on each principal component reflect the degree to which the feature contributes to the principal component; For the first The eigenvalues ​​of each principal component represent the variance explained by that principal component; loadings By analyzing the eigenvector components Standardization is performed to eliminate differences in the explanatory power of different principal components, ensuring that the loadings more fairly reflect the contributions of each agronomic trait to the principal components. Based on the obtained loadings, the principal component score for each data sample is calculated using the following formula: ,in, For the first The data sample at the ... Scores on principal components For the standardized first The first data sample Agronomic trait values, For the quantity of agronomic traits; in this embodiment, The value is set to 2, meaning the first two principal components, PC1 and PC2, are selected as the analysis objects. Through the above-mentioned independent dimensionality reduction processing, representative and uncorrelated comprehensive trait characteristics within each geographical region are effectively extracted, thus more accurately representing the planting characteristics of different regions; for example, ... Figure 4As shown, the principal component analysis (PCA) plots of maize agronomic traits in different ecological regions are presented. In North China, the first principal component (PC1) explained 38.57% of the variance, mainly contributed by ear height and plant height, indicating that suitable plant type is a key comprehensive factor affecting yield in North China. In Northeast China, PC1 explained 36.09% of the variance, mainly driven by growth period and leaf number, suggesting that breeding in this region should focus on the coordination between growth period and photosynthetic efficiency. In Northwest China, PC1 and PC2 together explained 55.17% of the variance, with plant height, thousand-grain weight, and ear row number having higher loadings on PC1, indicating that the synergistic effect of tall stalk, high thousand-grain weight, and high ear row number varieties can help improve maize yield in Northwest China.

[0080] The present invention is further configured such that the ecological zoning and feature profiling analysis includes:

[0081] Calculate the arithmetic mean of all agronomic traits in each geographic region, generate regional feature vectors, and perform Z-score normalization on the resulting regional feature vector matrix. Then, perform agglomerative hierarchical clustering based on Euclidean distance metric and Ward variance minimization join method to generate a hierarchical clustering dendrogram of agronomic traits.

[0082] Z-Score standardization was performed on the agronomic traits of all data samples. K-Means clustering was iteratively performed within the preset number of clusters. The optimal number of clusters was determined based on the principle of maximizing the silhouette coefficient. Based on the optimal number of clusters, K-Means clustering was performed to extract cluster centers as a universal trait combination pattern.

[0083] The silhouette coefficient is obtained by calculating the ratio of intra-cluster compactness to inter-cluster separation of the data sample.

[0084] Calculate the Euclidean distance between the feature vectors of each region and the trait combination patterns, and determine the pattern corresponding to the smallest Euclidean distance as the dominant trait combination pattern of the region.

[0085] The regional feature vector values ​​and cluster center values ​​are normalized to a preset numerical range through linear mapping to achieve dimensionless visualization of multidimensional features. Specifically, based on the aforementioned dimensionality reduction analysis, this embodiment further utilizes a combined clustering method to divide geographical regions and construct trait characteristic profiles for each region. Seven geographical regions of the maize regional experiment are selected as research objects, and the arithmetic mean of agronomic traits within each geographical region is taken to generate regional feature vectors. The regional feature vector matrix composed of all regional feature vectors is Z-score standardized to eliminate the dimensional influence between different traits. During the clustering process, Euclidean distance is used as a measure of differences between regions, and Ward's minimum variance method is used as the aggregation criterion to perform agglomerated hierarchical clustering, resulting in a hierarchical clustering tree diagram of agronomic traits reflecting the affinity between regions. The formula for calculating Euclidean distance is: ,in, For the region and region The Euclidean distance between them This indicates the overall difference between two geographical regions in terms of the mean of all agronomic traits. The smaller the value, the more similar the agronomic traits of the two geographical regions are; the larger the value, the more obvious the differences are. For the region In the The standardized value of an agronomic trait, that is, the mean of all samples in the geographic region for that agronomic trait, obtained after Z-Score standardization; For the region In the Standardized values ​​for each agronomic trait; by progressively merging similar regions, the following results are obtained: Figure 5 The hierarchical clustering dendrogram of agronomic traits in the ecological regions shown is from... Figure 5 As can be seen, the merging distance between the Northwest and North China regions is approximately 1.92, and the merging distance between the Central China and Southwest regions is approximately 3.25, forming smaller clusters. Ultimately, all ecological regions are divided into two major categories: cluster C0 (including Central China, South China, Northeast China, and Southwest China) and cluster C1 (including East China, Northeast China, and Northwest China). The South China and East China regions have the largest merging distance with other geographical regions, approximately 7.39, indicating the most significant differences in their agronomic trait structures. To extract cross-regional shared agronomic trait combination patterns, K-Means clustering analysis was used for all samples. Global cluster analysis was performed on the agronomic traits of the samples. Z-Score standardization was applied to all agronomic trait data to eliminate dimensional differences between traits, ensuring comparisons were made on the same scale and improving the accuracy of the cluster analysis. Within a preset range of cluster numbers, the K-Means clustering algorithm was iteratively executed, and the silhouette coefficient of each clustering result was calculated. The silhouette coefficient was used to evaluate the quality of the clustering results. By comparing the silhouette coefficients corresponding to different cluster numbers, the cluster number corresponding to the largest silhouette coefficient was selected as the optimal cluster number. The formula for calculating the silhouette coefficient is: ,in, For the sample The profile coefficient, Indicates sample The rationality of the current clustering method The larger the value, the larger the sample size. The better the clustering results; For the sample The average distance to other samples within the same cluster reflects the sample's... The density within the cluster; For the sample The average distance to samples from the nearest other clusters measures the sample's performance. The separation degree to other clusters is determined; after selecting the optimal number of clusters, the K-Means clustering algorithm is re-executed to obtain the corresponding cluster centers. Each cluster center represents a universal trait combination pattern, i.e., a combination of agronomic traits shared across regions; regional feature vectors are extracted for each geographic region. These regional feature vectors are composed of the arithmetic mean of agronomic traits within the geographic region and are used to reflect the overall agronomic trait characteristics of the geographic region; the Euclidean distance between each regional feature vector and each trait combination pattern is calculated. This Euclidean distance quantifies the similarity between the regional feature vector and each trait combination pattern. The smaller the distance, the stronger the similarity between the regional feature vector and each trait combination pattern. The higher the similarity of the trait combination patterns, the more likely the dominant trait combination pattern in the region will be selected. Through this matching process, a correspondence between regional trait features and globally universal patterns is established, reflecting the structural differences in agronomic traits among different ecological regions, thus establishing representative dominant trait profiles for each region. By normalizing the regional feature vector values ​​and cluster center values ​​to a preset numerical range [0.1, 0.9] using linear mapping, a dimensionless representation of multidimensional trait features is achieved. The normalized regional feature vector values ​​and dominant cluster center values ​​are plotted on a radar chart, as shown below. Figure 6 As shown, the standardized trait profiles of each ecological region are displayed. Each radar chart compares the standardized trait profile of that region with the dominant cluster center value and another cluster center value, intuitively presenting the trait structure characteristics of different ecological regions and helping to more clearly understand the comprehensive trait characteristics of each region. For example, the Northeast and Northwest regions are significantly higher than other geographical regions in traits such as growth period, plant height, thousand-grain weight, and yield, showing significant trait advantages. The Central China and East China regions show higher planting density and ear height, but lower yield and plant type related traits, presenting a compact and early-maturing structure. The trait performance of other geographical regions is between these two typical patterns. This visualization result provides an intuitive basis for the differentiated positioning of regional breeding objectives. The normalization formula is: in, The value is the normalized value; These are raw numerical values, such as observed values ​​of a certain agronomic trait; This is the minimum value of the trait in the global data, that is, the minimum observed value of the trait among all data samples; This represents the maximum value of the trait in the global data. To further scale the values, making the data more concentrated within a specific range, and to avoid extreme 0 and 1 values, Scaling factor The offset is used as the offset. Through the above steps, the relationship between different ecological regions is revealed, the cross-regional shared trait combination pattern is extracted, and the correspondence between regions and patterns is established, ultimately forming a multidimensional dimensionless ecological region feature profile.

[0086] The present invention is further configured such that the customized maize breeding strategy formulates personalized breeding objectives for each ecological region based on the evolution patterns of maize traits, the correlation between key traits and yield, and the synergistic effects of traits in each ecological region. Specifically, in this embodiment, based on the spatiotemporal evolution patterns of maize traits in each ecological region analyzed by a multidimensional data spatiotemporal analysis framework, the correlation between key traits and yield, and the synergistic and trade-off relationships between traits, and combined with the regional trait profiles and dominant pattern profiles generated by principal component analysis and combinatorial clustering, customized maize breeding strategies are formulated for each ecological region. To improve maize yield and adaptability in different ecological regions; for example, Northeast China has the longest growth period and tallest plant type, resulting in high-yield varieties with outstanding ear length and number of ear rows. However, excessive plant height and ear position may lead to lodging risk. Therefore, breeding should focus on fully utilizing the high-yield potential brought about by the longer growth period, consolidating the advantages in yield and ear traits, while also emphasizing the improvement of plant structure to reduce lodging risk, such as appropriately lowering ear height and strengthening stem strength. The breeding objective should be to improve lodging resistance and maturity stability while maintaining high yield and large ear advantages. To achieve stable and high yields, the breeding strategy should focus on adjusting plant height and ear height to a more suitable range to optimize ventilation, light penetration, and light energy utilization efficiency within the population. Simultaneously, considering the region's climate characteristics, the growth period and the number of functional leaves should be appropriately adjusted to cultivate varieties that fully utilize the growing season without excessive growth, thus promoting stable and increased yields. Furthermore, attention should be paid to enhancing drought resistance and tolerance to dense planting to complement the relatively long growing season. High planting density further increases yield. While East China has a shorter growth period and shorter plant height, with relatively lower ear length and number of ear rows, it is well-suited for high-density planting. This indicates that varieties from East China are adapted to compact planting systems, especially double-cropping systems. Breeding strategies should prioritize early-maturing, high-yielding varieties to ensure rapid growth and grain filling within a limited growth period, further increasing thousand-grain weight and seed setting rate to compensate for the smaller ear size. Simultaneously, the tolerance to dense planting and resistance to disease and lodging should be strengthened to ensure stable yields under high-density planting and humid conditions.To increase yield per unit area, the genetic potential for ear length and ear row number should be appropriately increased to achieve yield increase without prolonging the growth period. The overall traits of the Central China region fall between those of the North and South, with a moderate growth period, relatively high plant height and ear position, but moderate ear traits and thousand-grain weight, and a yield level close to the national average. Breeding should focus on optimizing plant height and ear position to ensure ventilation and light penetration and reduce disease occurrence under hot and humid conditions. Simultaneously, improving ear grain structure and grain weight can compensate for slightly lower ear length and ear row number, enhancing single-ear productivity. Furthermore, strengthening the variety's resistance to diseases and pests and its tolerance to high temperature and humidity will ensure yield increase. To ensure stable yields in complex environments, new high-yielding and widely adaptable varieties can be developed in Central China by coordinating ear size and plant density. Northwest China, with its longer growth period, higher plant height, and greater yield potential, boasts outstanding thousand-grain weight and ear row number among all regions. Breeding strategies should fully leverage the advantages of high thousand-grain weight and ear row number in Northwest China, cultivating large-ear varieties tolerant of dense planting to achieve breakthroughs in yield per unit area. Furthermore, attention should be paid to maintaining suitable growth periods and plant height, ensuring compatibility with irrigation and mechanization conditions, improving drought resistance and tolerance to poor soil, and combining this with high-light-efficiency plant architecture design to fully realize yield-increasing potential. The Southwest region has a longer growing season, with higher panicle rows and panicle length, but relatively insufficient light and heat resources and complex terrain, resulting in a medium-to-high yield. Breeding should focus on increasing the yield potential per plant, especially by selecting materials with higher grain weight to increase yield per panicle, while appropriately shortening the growing season to ensure maturity in complex environments. Since planting density in the mountainous areas of Southwest China is greatly affected by terrain, breeding should select types with lower panicle position, compact plant type, and large panicles with many grains. At the same time, disease resistance, especially resistance to panicle rot and lodging, should be strengthened to adapt to high humidity and dispersed production conditions. The South China region has the shortest growing season. The plant is short in stature, with small ears and relatively low yield, and is adapted to hot and rainy tropical and subtropical environments. The primary goal of breeding is to enhance adaptability to extreme environments. Emphasis should be placed on increasing ear height and optimizing plant type to improve ventilation and light penetration within the plant population, thereby reducing disease occurrence under hot and humid conditions. At the same time, the variety's resistance to diseases and pests should be improved, especially resistance to southern rust and sheath blight, to ensure stable growth under complex biological stress. Genes that improve ear length and grain weight can be appropriately introduced to increase yield per plant without extending the growth period. Breeding should also select varieties that are tolerant to dense planting to further improve land utilization and yield.

[0087] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating regional breeding strategies based on big data on maize traits, characterized in that, include: Collect and integrate multi-source heterogeneous maize regional experimental data, extract agronomic trait data, and construct a structured dataset; Based on the maize regional trial data, the extracted agronomic traits include regional yield, growth period, plant height, ear length, thousand-grain weight, ear height, number of leaves at maturity, planting density, and number of ear rows. The agronomic trait data are cleaned, denoised, missing value processed, and standardized, and then fused according to preset key fields and stored in a relational database to form a structured dataset. Based on the characteristics of the agricultural ecological environment and climate zoning methods, the main corn-producing areas are divided into a predetermined number of ecological zones; A multidimensional spatiotemporal analysis framework is constructed, sequentially performing global descriptive statistical analysis, trait spatiotemporal evolution analysis, trait synergy analysis, trait feature dimensionality reduction analysis, and ecological regionalization and feature profiling analysis on structured datasets. The ecological regionalization and feature profiling analysis includes: calculating the arithmetic mean of all agronomic traits within each geographic region, generating regional feature vectors, and performing Z-score standardization on the resulting regional feature vector matrix; then performing agglomerative hierarchical clustering based on Euclidean distance and the Ward variance minimization join method to generate a hierarchical clustering dendrogram of agronomic traits; and finally, performing Z-score standardization on the agronomic traits of all data samples. The standardization process involves iteratively performing K-Means clustering within a preset cluster size range. The optimal number of clusters is determined based on the principle of maximizing the silhouette coefficient. K-Means clustering is then performed based on this optimal number to extract cluster centers as universal trait combination patterns. The silhouette coefficient is obtained by calculating the ratio of intra-cluster density to inter-cluster separation in the data sample. The Euclidean distance between the feature vectors of each region and the trait combination patterns is calculated, and the pattern corresponding to the minimum Euclidean distance is determined as the dominant trait combination pattern for that region. The region feature vector values ​​and cluster center values ​​are normalized to a preset numerical range through linear mapping, achieving dimensionless visualization of multidimensional features. Based on the spatiotemporal analysis results, customized maize breeding strategies for different ecological regions are generated; The customized maize breeding strategy formulates individualized breeding objectives for each region based on the evolution patterns of maize traits, the correlation between key traits and yield, and the synergistic effects of traits in each ecological region.

2. The method for generating regional breeding strategies based on big data of maize traits according to claim 1, characterized in that, The ecological regions include seven major geographical regions: Northeast, North China, East China, Central China, South China, Southwest, and Northwest.

3. The method for generating regional breeding strategies based on big data of maize traits according to claim 1, characterized in that, The global descriptive statistical analysis includes: By reading the raw data file from the specified path and converting it into a data table for storage, the numerical agronomic trait column in the data is preprocessed, and entries that cannot be identified as numerical values ​​are converted into null values. All null values ​​are then filled with the arithmetic mean of the column. Using preset statistical functions, a comprehensive statistical analysis is performed on all numerical agronomic trait columns to calculate global descriptive statistics for the structured dataset. These descriptive statistics include the total number of samples, arithmetic mean, standard deviation, minimum value, quantile value, and maximum value. The data are grouped according to the preset geographic region classification variables, and box plots are generated for different regions for each agronomic trait to compare the differences in median, interquartile range and outlier distribution of each region for that agronomic trait.

4. The method for generating regional breeding strategies based on big data of maize traits according to claim 1, characterized in that, The spatiotemporal evolution analysis of the traits includes: For all agronomic trait columns in a structured dataset, fill missing values ​​with the arithmetic mean of all non-empty values ​​in that column; Using provinces as the basic geographical unit, the dataset is divided into multiple data subsets. In each data subset, a univariate linear regression model is constructed with year as the independent variable and a single agronomic trait as the dependent variable. It is verified whether the year records in each data subset cover more than one year and contain data points from at least two different years. If the conditions are not met, the calculation for that data subset is terminated. If the conditions are met, a univariate linear regression model is established based on the year data, and the regression slope is calculated. The relative annual rate of change of each agronomic trait in each province is calculated based on the regression slope and the arithmetic mean of the individual agronomic trait in the corresponding provincial data subset. If the arithmetic mean of the agronomic trait in the provincial data subset is zero, the relative annual rate of change is considered invalid.

5. The method for generating regional breeding strategies based on big data of maize traits according to claim 1, characterized in that, The trait synergy analysis includes For all agronomic trait columns to be analyzed in the structured dataset, use the median of all non-empty values ​​in that column to fill in missing values; The data is divided into corresponding time intervals by year using a preset time window. Within each time interval, the data is grouped according to province and variety name, and the arithmetic mean of all agronomic traits in each group is calculated to generate an aggregated dataset. Provinces are mapped to geographic regions. Within each geographic region, with regional yield per acre as the target trait, Pearson correlation coefficients and significance p-values ​​between other agronomic traits and the target trait are calculated based on the aggregated dataset. The Benjamini-Hochberg method was used to perform multiple hypothesis testing to correct the significance p-values ​​in order to control the false positive rate. The significance level was determined based on the corrected p-values, and significance was marked accordingly.

6. The method for generating regional breeding strategies based on big data of maize traits according to claim 1, characterized in that, The dimensionality reduction analysis of the trait features includes: The data is grouped according to geographic regions, and partitioned principal component analysis is performed independently on the subset of data for each geographic region. For data within each geographic region, Z-Score standardization is performed to ensure that all agronomic traits are listed with a uniform standard within that geographic region. The covariance matrix or correlation coefficient matrix is ​​calculated based on the standardized agronomic trait data. The loading matrix is ​​generated by eigenvalue decomposition. The principal component score matrix is ​​obtained by multiplying the standardized agronomic trait data with the loading matrix. Extract the score of each data sample on a preset number of principal components, and extract the loading values ​​of each original agronomic trait on a preset number of principal components. The absolute value of the loading reflects the degree of contribution of the agronomic trait to the principal components, and the sign of the loading indicates the direction of contribution.

Images