Soybean variety environmental adaptability identification and growth regulation method

Abstract

The invention provides a soybean variety environmental adaptability identification and growth regulation and control method, which comprises the following steps: carrying out a multi-environmental factor interaction test, and carrying out observation data acquisition and pretreatment containing censored data; constructing a Bayesian Weibull AFT model considering the random difference between the multi-environment factor and the soybean variety; fitting the Bayesian Weibull AFT model by using a Markov chain Monte Carlo method, and carrying out model convergence diagnosis and goodness of fit verification; respectively extracting posterior distribution of multiple environmental factors and posterior distribution of soybean variety random differences from the fitted Bayesian Weibull AFT model, and respectively carrying out fixed effect analysis and random effect analysis; and performing variety screening and environment optimization based on analysis results of the fixed effect and the random effect. According to the method, effective utilization of the deleted data, multi-environmental factor interaction quantification and accurate evaluation of varieties and environmental effects are realized.

Description

Technical Field

[0001] This invention relates to the technical field of agricultural biotechnology, and in particular to a method for identifying the environmental adaptability and regulating the growth of soybean varieties. Background Technology

[0002] With the development of controlled environment agriculture (CEA, such as plant factories and smart greenhouses), it has become possible to maximize crop yield and quality through precise control of environmental factors. Soybeans, as an important food and economic crop, are heavily influenced by CO2 concentration, light quality, and photoperiod. However, these environmental factors exhibit complex interactions, and different soybean varieties show significant genetic differences in their responses.

[0003] Traditional research methods often employ analysis of variance (ANOVA) and similar methods for comparison. These methods have significant limitations: 1) They struggle to effectively handle data censoring, as many soybean varieties fail to complete their growth cycle (R6 stage) under specific treatments, resulting in the loss of valuable information using traditional methods; 2) Existing studies often analyze single environmental factors individually, lacking the ability to analyze complex interactions between environmental factors; 3) They cannot simultaneously assess both fixed environmental effects (such as the average impact of a certain CO2 concentration) and random effects (such as genetic differences within the variety itself), making it difficult to quantify variety-environment interactions and leading to low variety selection efficiency.

[0004] In addition, for example, the invention patent with publication number CN116561952A discloses a soybean growth period prediction method based on a soybean growth period model. By dynamically simulating physiological development time and photothermal effects, and combining the Monte Carlo-Markov chain algorithm to optimize parameters, a soybean growth period model based on physiological development time is constructed. This solves the problem of inaccurate soybean growth period simulation in the prior art, and realizes accurate prediction of soybean growth period in multiple stages and clear physiological significance of variety parameters. However, this method cannot analyze the complex interaction effects between environmental factors, nor can it simultaneously evaluate environmental fixed effects and random effects. Summary of the Invention

[0005] To address the technical problems of existing methods in effectively handling data truncation, insufficient ability to analyze complex interactions between environmental factors, and inability to simultaneously assess fixed and random effects, this invention proposes a method for identifying the environmental adaptability and growth regulation of soybean varieties. By utilizing the Bayesian Weibull AFT survival analysis model, it achieves effective utilization of censored data, quantification of interactions between multiple environmental factors, and accurate assessment of variety-environment effects, ultimately providing a scientific basis for variety selection and environmental optimization.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A method for identifying the environmental adaptability and regulating the growth of soybean varieties, comprising the following steps:

[0008] S1: Conduct multi-environmental factor interaction experiments and collect and preprocess observational data including censored data;

[0009] S2: Construct a Bayesian Weibull AFT model that considers multiple environmental factors and random differences in soybean varieties;

[0010] S3: The Bayesian Weibull AFT model is fitted using the Markov chain Monte Carlo method, and model convergence diagnosis and goodness-of-fit verification are performed.

[0011] S4: Extract the posterior distributions of multiple environmental factors and the posterior distributions of random differences in soybean varieties from the fitted Bayesian Weibull AFT model, and perform fixed-effects and random-effects analysis respectively;

[0012] S5: Variety selection and environmental optimization based on the analysis results of fixed and random effects;

[0013] S6: Use the fitted Bayesian Weibull AFT model to predict the expected reproductive time under any variety-environment combination, and perform variety-environment interaction analysis to generate a core heat map, recommending the optimal environment for the variety or configuring the most suitable variety for the environment.

[0014] Furthermore, the aforementioned multiple environmental factors include, but are not limited to, CO2 concentration, light quality, and photoperiod.

[0015] Furthermore, the method for constructing a Bayesian Weibull AFT model that considers multiple environmental factors and random differences in soybean varieties is as follows:

[0016] S21: Using the logarithm of survival time as the response variable, an AFT model is constructed by incorporating environmental fixed effects and variety random effects;

[0017] S22: Correlation between AFT model and Weibull distribution;

[0018] S23: Calculate the total likelihood function based on the probability density function and survival function of the Weibull distribution;

[0019] S24: Define the prior distributions for environmental fixed effects and variety random effects in the AFT model;

[0020] S24: Based on the prior distribution and the total likelihood function, calculate the posterior distribution of fixed effects and variety random effects in the AFT model using the Bayesian method.

[0021] Furthermore, the AFT model is as follows:

[0022]

[0023] in, Here, i represents the survival time, i is the variety number, and j is the j-th environment for variety i. Design a matrix for the fixed effects of environmental factors, including main effect terms and interaction terms. These are environmental fixed effects coefficients, including main effects coefficients and interaction effects coefficients. Let be the random intercept for the i-th variety. Let Variance be the random intercept of the variety. α is the error term, which follows a Gumbel distribution, and α is the Weibull shape parameter.

[0024] Furthermore, the method for linking the AFT model with the Weibull distribution is as follows:

[0025] For AFT model Inverse transformation, we get Substituting the scale parameter of the Weibull distribution, let the scale parameter... ,but .

[0026] Furthermore, the method for calculating the total likelihood function based on the probability density function and survival function of the Weibull distribution is as follows:

[0027]

[0028]

[0029] in, This represents the likelihood probability of a single variety. The total number of environmental factor levels. For the set of all observation data, This represents the probability density function of the Weibull distribution. Let Weibull be the survival function. The truncated indicator variable is V; V is the total number of varieties. This represents the set of random intercepts for all varieties.

[0030] Furthermore, the model convergence diagnosis is achieved using R-hat statistics and trace plots;

[0031] The goodness-of-fit verification is performed by checking the goodness-of-fit through posterior prediction:

[0032] Density superposition: The overlap between the density distribution curves of the observed data and the model prediction data is greater than or equal to the overlap threshold, indicating a good fit;

[0033] Mean comparison: If the deviation between the censoring ratio of the observed data and the censoring ratio of the model-predicted data is less than or equal to the deviation threshold, it indicates that the model can effectively handle censored data.

[0034] Furthermore, the method for analyzing fixed effects is as follows:

[0035] Extracting environmental fixation effects from the fitted Bayesian Weibull AFT model The posterior distribution was used to calculate the acceleration factor at different levels of environmental factors for the same variety. , The environmental fixed effects coefficient representing the environmental reference baseline is extracted. The acceleration factor of the interaction effect is represented in the middle. ;

[0036] If the accelerating factor at a certain level of environmental factor <1 and The preset confidence interval does not contain 1, indicating that the level of this environmental factor will accelerate soybean growth to the R6 stage; if >1 and The pre-defined confidence interval does not contain 1, indicating that the level of this environmental factor will slow soybean growth to the R6 stage. If... If the pre-set confidence interval contains 1, it means that the level of this environmental factor has no significant effect on soybean growth.

[0037] Furthermore, the method for analyzing the random effects of varieties is as follows:

[0038] Extracting the variety random intercept from the fitted Bayesian Weibull AFT model The posterior distribution was used to calculate the variety-specific acceleration factor for each variety at the same environmental factor level. , The random intercept of varieties serving as a reference benchmark;

[0039] The adaptability of all varieties at the same environmental factor level was ranked and classified: If <1 and If the preset information interval does not contain 1, then the variety is an acceleration type. >1 and If the preset information interval does not contain 1, then the variety is a slow-down type. If the preset confidence interval contains 1, then the variety is neutral.

[0040] Furthermore, the method for variety selection and environmental optimization based on the analysis results of fixed effects and random effects is as follows:

[0041] Based on the variety-specific acceleration factor obtained in step S4 To screen out superior varieties that exhibit stable performance and short growth cycles under different environments: determine the target environment and iterate through the variety-specific acceleration factors of all varieties under the target environment. Selecting variety-specific accelerating factors Accelerated varieties with a value <1 and whose preset information interval does not contain 1 are considered superior varieties;

[0042] Acceleration factor based on the environmental factor level obtained in step S4 To determine the optimal combination of environmental factors that can maximize the acceleration of soybean growth and shorten the growth cycle: identify the target variety and iterate through all environmental factor levels for the target variety to identify the acceleration factors. Select For acceleration environment combinations where the value is less than 1 and the preset confidence interval does not contain 1, calculate the comprehensive benefit value. (1 - Energy consumption increment).

[0043] The beneficial effects of this invention are:

[0044] By using the Bayesian Weibull AFT model to process censored data, the problems of information loss or biased estimation in traditional methods are avoided, resulting in more accurate and reliable results. The AF value provides an intuitive and quantitative interpretation of the effect size.

[0045] By setting up a three-factor cross-treatment experiment involving CO2 concentration, light quality, and photoperiod, a Bayesian Weibull AFT model was constructed to incorporate the main effects and potential interactions of the three factors into the environmental fixed effects. The interaction effects were quantified with the acceleration factor AF, thus achieving the analysis of multiple key environmental factors and their interactions.

[0046] By employing a Bayesian Weibull AFT model with a dual-effect structure (environmental fixed effects to analyze the average environmental impact and variety random intercepts to remove genetic differences), combined with MCMC sampling and posterior inference, and supplemented by variety ranking, bidirectional clustering heatmaps, the variety-environment interaction is quantified and visualized, providing a basis for variety selection and environmental optimization.

[0047] A well-fitted model can be used as a prediction tool to predict the performance of new varieties in new environments without having to repeat all the experiments, thus saving a lot of time and costs.

[0048] For breeders, it can quickly screen out varieties that perform robustly under specific adverse conditions (such as T9 and T10) or that develop rapidly under high-yield environments (such as T4); for growers, it can use the model to query the optimal combination of light, temperature, and air environment parameters recommended by the model based on the varieties they own, so as to achieve precise control and energy saving and efficiency improvement. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0050] Figure 1 This is a flowchart of the method of the present invention.

[0051] Figure 2 The heatmap shows the R6 stage of 21 soybean varieties under 10 different environmental treatments (T1-T10) in this embodiment of the invention.

[0052] Figure 3 The Kaplan-Meier survival curves of soybean populations reaching R6 under different environmental treatments (T1-T10) according to embodiments of the present invention are shown.

[0053] Figure 4 This is a trace diagram of the MCMC in an embodiment of the present invention.

[0054] Figure 5 This is a convergence diagnostic plot (R-hat value) of the Bayesian model in an embodiment of the present invention.

[0055] Figure 6 This is a goodness-of-fit verification graph for the posterior prediction test in an embodiment of the present invention.

[0056] Figure 7 This is a diagram illustrating the fixed effects of environmental factors in an embodiment of the present invention.

[0057] Figure 8 This is a ranking diagram of random effects on varieties according to an embodiment of the present invention.

[0058] Figure 9 This is a comparison chart showing the predicted development time of the varieties in this invention under optimal (T4) and adverse (T9) conditions.

[0059] Figure 10 This is a heatmap of bidirectional cluster analysis based on the Bayesian Weibull AFT model, showing the time of reaching R6 for 21 soybean varieties under 10 environmental treatments, as predicted in this embodiment of the invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] A method for identifying the environmental adaptability and regulating the growth of soybean varieties, such as Figure 1 As shown, the steps include:

[0062] S1: Conduct multi-environmental factor interaction experiments and collect and preprocess observational data including censored data.

[0063] In this embodiment of the application, the method for conducting multi-environmental factor interaction experiments is as follows:

[0064] First, the core environmental factors were identified, including three dimensions of environmental factors: CO2 concentration (400 / 1000 / 1400ppm), light quality (blue-red light ratio including: 5:1 / 2:1 / full spectrum / natural light), and photoperiod (4 / 6 / 8 / 16h); each environmental factor corresponds to 10 environmental factor levels (T1-T10).

[0065] The specific settings are as follows:

[0066] 21 soybean varieties, 10 experimental treatments:

[0067] T1: CO2 concentration 1000ppm, blue-red ratio 5:1, light duration 6h;

[0068] T2: CO2 concentration 1000ppm, blue-red ratio 5:1, light duration 8h;

[0069] T3: CO2 concentration 1000ppm, blue-red ratio 2:1, light duration 6h;

[0070] T4: CO2 concentration 1000ppm, blue-red ratio 2:1, light duration 8h;

[0071] T5: CO2 concentration 1400ppm, blue-red ratio 5:1, light duration 6h;

[0072] T6: CO2 concentration 1400ppm, blue-red ratio 5:1, light duration 8h;

[0073] T7: CO2 concentration 1400ppm, blue-red ratio 2:1, light duration 6h;

[0074] T8: CO2 concentration 1400ppm, blue-red ratio 2:1, light duration 8h;

[0075] T9: CO2 concentration is normal, full spectrum, light duration 16h;

[0076] T10: CO2 concentration is normal, simulated natural light, light duration 4h.

[0077] Furthermore, 21 representative soybean varieties were selected, and each variety was bred under the aforementioned 10 groups of environmental factor levels.

[0078] In this embodiment, the data acquisition method is as follows: record the time required for soybeans under each variety-environment combination to reach the growth stage (R6 stage); for observations that have not reached the R6 stage by the experimental cutoff time (e.g., 126 days), their time is recorded as the cutoff time, and this data is marked as censored data (NA). This avoids the loss of key information about incomplete growth stages by traditional methods.

[0079] In this embodiment of the application, the preprocessing method is as follows:

[0080] First, environmental factors such as CO2 concentration, light quality, and photoperiod are uniformly coded. For example, the numbers 1 / 2 / 3 represent CO2 400 / 1000 / 1400ppm, 1 / 2 / 3 / 4 represent 5:1 / 2:1 / full spectrum / natural light, and 1 / 2 / 3 / 4 represent 4 / 6 / 8 / 16h, so as to avoid the model calculation being affected by the chaotic format.

[0081] Furthermore, the environmental fixed effects were evaluated using T4 as the reference group, and environmental factor levels were aligned, with the first level of each environmental factor set as the environmental factor level corresponding to T4.

[0082] Furthermore, the collected R6 arrival time data were converted into a survival analysis format, which includes data format of variety ID, CO2, light quality, photoperiod, time and status. Here, status=1 indicates that the event was observed during the experiment (R6 arrival), and status=0 indicates that the data was truncated (no event was observed).

[0083] S2: Construct a Bayesian Weibull AFT model that considers multiple environmental factors and random differences in soybean varieties to predict soybean growth time.

[0084] Soybean growth period data conforms to the Weibull distribution, characterized by slow growth in the early stage, accelerated growth in the middle stage, and stabilization in the later stage. It is more suitable for describing survival time data than the normal distribution. Compared with the Cox proportional hazards model, the AFT model can directly output the acceleration factor (AF), intuitively quantifying the multiplier of the impact of environment / variety on the growth period, and is easier to transform into practical application schemes. Bayesian methods can integrate existing research information through prior distribution (such as the known range of the impact of CO2 concentration on soybeans) and can output the posterior distribution of parameters. Compared with the frequentist approach (such as ANOVA), it is more suitable for the analysis of small sample and censored data.

[0085] The method for constructing a Bayesian Weibull AFT model that considers multiple environmental factors and random differences in soybean varieties is as follows:

[0086] First, using the logarithm of survival time as the response variable, an AFT model is constructed by incorporating environmental fixed effects and variety random effects to quantify the logarithmic failure time:

[0087]

[0088] in, Here, i represents the survival time, i is the variety number, and j is the j-th environment for variety i. Design a matrix for the fixed effects of environmental factors, including the main effects term ( ) and interaction items ( ),in Indicates CO2, Indicates light quality, Indicates the optical period. This is the environmental fixed effects coefficient, used to quantify the impact of the environment on the reproductive period. , This represents the three main effect coefficients. This represents the four interaction effect coefficients. Corresponding to the CO2 effect, Corresponding to the light quality effect, Corresponding to photoperiodic effect, This represents the interaction effect among three environmental factors, where T is the transpose. Let be the random intercept for the i-th variety, quantifying the random effect of the variety. , To quantify the genetic differences among varieties, we use the random intercept variance. The error term follows a Gumbel distribution. α is the Weibull shape parameter, which determines the steepness of the fertility period distribution (α>1 indicates concentrated fertility period, α<1 indicates dispersed fertility period).

[0089] Furthermore, the correlation between the AFT model and the Weibull distribution is examined: Inverse transformation, we get Substituting the scale parameter of the Weibull distribution, let the scale parameter... ,but Scale parameters It also includes contributions from both environmental fixed effects and variety random effects, scale parameters The larger the value, the longer the failure time, meaning the better the adaptability to environmental factors or the variety.

[0090] Furthermore, the total likelihood function is calculated based on the probability density function and survival function of the Weibull distribution:

[0091] Calculate the likelihood contribution within a single variety:

[0092]

[0093] in, This represents the likelihood probability of a single variety. This represents the total number of environmental factor levels (T1 to T10). For the set of all observation data, This represents the probability density function of the Weibull distribution. Let Weibull be the survival function. =1-status is the truncation indicator variable. =1 indicates truncation, meaning the observation time equals the truncation time. =0 indicates failure, meaning the observed time equals the actual failure time, which is the actual time required to reach period R6.

[0094] The probability density function of the Weibull distribution is:

[0095]

[0096] The survival function of the Weibull distribution is:

[0097] .

[0098] Calculate the total likelihood function:

[0099]

[0100] Where V represents the total number of varieties. This represents the set of random intercepts for all varieties.

[0101] Furthermore, for environmental fixation effects Random intercept of varieties The weak information prior distribution is defined by the Weibull shape parameter α:

[0102] In this embodiment, environmental fixation effect The prior distribution is set as ~Normal(0,5 2 Based on existing research, the absolute values ​​of the effect coefficients of CO2 and light quality on the growth period are usually <3.5 times the standard deviation, which can cover most possible ranges. The random intercept of varieties... The prior distribution is set as follows: ~Normal(0,5 2 To cover a reasonable range of differences in growth period among varieties, the prior distribution of the Weibull shape parameter is set as follows: ~Gamma(2,0.1), where Gamma is the gamma distribution.

[0103] Furthermore, based on the total likelihood function and the established weak-information prior distribution, Bayesian methods are used to infer model parameters, including environmental fixed effects. Random intercept of varieties Posterior distribution of the Weibull shape parameter:

[0104]

[0105] in, express{ }, As a prior distribution, For the posterior distribution, The marginal likelihood is merely a "normalization constant" that ensures the integral of the posterior distribution equals 1, and can be ignored in actual calculations.

[0106] S3: The Bayesian Weibull AFT model is fitted using the Markov Chain Monte Carlo (MCMC) method, and model convergence diagnosis and goodness-of-fit verification are performed.

[0107] First, the fitting process uses the brms package in the R language environment to obtain the environment fixed effects through the MCMC algorithm. Variety random intercept The Weibull shape parameter posterior distribution was used, and the sampling parameters were set as follows: number of independent chains (chains=4), number of iterations (iter=4000), and warmup period (warmup=2000).

[0108] MCMC sampling updates in blocks in the order of fixed effects → random effects → shape parameters, with each step calculated based on the current values ​​of other parameters.

[0109] Furthermore, convergence diagnosis is achieved using R-hat statistics and trace plots.

[0110] Specifically, all parameters, including environmental fixation effects, are calculated. Variety random intercept The R-hat statistic for the Weibull shape parameter is used. When the R-hat value of all parameters is <1.05, it indicates that the MCMC chain is well mixed and has no divergence.

[0111] Plot four chains for each parameter, i.e., trace plots. If the chain satisfies the conditions of no trend, no periodicity, and good overlap between chains, it indicates that the MCMC chains are well mixed and do not diverge.

[0112] If the convergence conditions of both the R-hat statistic and the trace plot are met, it indicates that the model has good convergence.

[0113] Furthermore, the goodness of fit was verified through posterior prediction check (PPC): density superposition: the overlap between the density distribution curves of the observed data and the model prediction data is ≥95% to indicate a good fit; mean comparison: the deviation between the censoring ratio of the observed data and the censoring ratio of the model prediction data is ≤2%, indicating that the model can effectively handle censored data.

[0114] S4: Extract the posterior distributions of multiple environmental factors and the posterior distributions of random differences in soybean varieties from the fitted Bayesian Weibull AFT model, and perform fixed-effects and random-effects analysis respectively.

[0115] After discarding the combustion period samples, the remaining M = iter − warmup posterior samples were used for inference, focusing on explaining the environmental fixed effects (environmental factors) and the variety random effects (variety variation).

[0116] Fixed effects analysis:

[0117] Extracting environmental fixation effects from the fitted Bayesian Weibull AFT model The posterior distribution was used to calculate the acceleration factor at different levels of environmental factors for the same variety. , The environmental fixed effects coefficient representing the environmental reference baseline is extracted. The acceleration factor of the interaction effect is represented in the middle. .

[0118] Using T4 as a reference, if the acceleration factor representing the interaction effect is at a certain level of environmental factor for the same variety... <1 and The 95% confidence interval (CI) does not include 1, indicating that the level of this environmental factor will accelerate soybean growth to the R6 stage; >1 and If the 95% confidence interval does not include 1, it means that the level of this environmental factor will slow soybean growth to the R6 stage. If the 95% confidence interval includes 1, it means that the level of this environmental factor has no significant effect on soybean growth.

[0119] Analysis of random effects on varieties:

[0120] Extracting the variety random intercept from the fitted Bayesian Weibull AFT model The posterior distribution was used to calculate the variety-specific acceleration factor for each variety at the same environmental factor level. .

[0121] The adaptability of all varieties at the same environmental factor level was ranked and classified: If <1 and If the 95% confidence interval does not include 1, then the variety is an accelerated type. >1 and If the 95% confidence interval does not include 1, then the variety is a slow-reducing type. If the 95% confidence interval includes 1, then the variety is neutral.

[0122] This invention incorporates the main effects of CO2, light quality, and photoperiod, as well as the potential interactions among them, into the environmental fixed effects using a Bayesian Weibull AFT model. The value is essentially a quantification of the overall effect of a single factor's main effect plus the interaction effect between factors.

[0123] S5: Based on the analysis results of fixed and random effects, conduct variety selection and environmental optimization.

[0124] Based on the variety-specific acceleration factor obtained in step S4 We selected superior varieties that exhibit stable performance and short growth cycles under different environments.

[0125] Specifically, the target environment is determined (such as an indoor farm environment with high CO2 and specific light quality, or an environment simulating natural stress), and all varieties are examined to identify the variety-specific acceleration factors for all varieties under the target environment. Selecting variety-specific accelerating factors Accelerated varieties with a value <1 and a 95% confidence interval that does not include 1 are considered superior varieties.

[0126] Acceleration factor based on the environmental factor level obtained in step S4 The goal is to determine the optimal combination of environmental factors that can maximize the acceleration of soybean growth and shorten the growth cycle.

[0127] Specifically, identify the target variety and iterate through all environmental factor levels for the target variety to identify the accelerating factors. Select For acceleration environment combinations where the value is less than 1 and the 95% confidence interval does not include 1, calculate the overall benefit value = (1 - Energy Consumption Increment). The energy consumption increment is the percentage increase in energy consumption for this environment relative to the baseline T4. For example, if the CO2 concentration is high in T7, the energy consumption increment is 20%.

[0128] S6: Use the fitted Bayesian Weibull AFT model to predict the expected reproductive time under any variety-environment combination, and perform variety-environment interaction analysis to generate a core heat map, recommending the optimal environment for the variety or configuring the most suitable variety for the environment.

[0129] The method for generating core heatmaps through variety-environment interaction analysis is as follows:

[0130] First, based on the fitted Bayesian Weibull AFT model, the predicted time for all varieties × environmental factor level combinations is generated;

[0131] Furthermore, the fitted() function was used to extract long-format data of predicted fertility time for each variety × environmental factor level combination;

[0132] Furthermore, the pivot_wider() function is used to convert the extracted long-format data of predicted fertility time into a predicted time matrix of variety (rows) × environmental factor level (columns);

[0133] Furthermore, the Viridis color coding is used to gradually change from purple (short time, fast development) to yellow (long time, slow development).

[0134] Furthermore, the distance matrix between varieties and the distance matrix between environmental factor levels are calculated;

[0135] Furthermore, using the complete linkage method, clustering trees at the variety level and environmental factor level were constructed respectively;

[0136] Further, line notes: Based on the average prediction time of varieties at all environmental factor levels, varieties are divided into three performance groups: "Fast" (green), "Medium" (yellow), and "Slow" (red);

[0137] Furthermore, the annotations are as follows: The specific combination of factors at each environmental factor level is displayed using color coding: CO2 concentration: 400 ppm (light yellow), 1000 ppm (orange), 1400 ppm (deep red); light quality: 5:1 (light purple), 2:1 (purple), Full (deep purple), Natural (very deep purple); photoperiod: 4h (light green), 6h (green), 8h (deep green), 16h (dark green).

[0138] Furthermore, the prediction time matrix, clustering tree, color coding, and row and column annotations are integrated using the ComplexHeatmap package, a bidirectional clustering heatmap tool, to generate the final heatmap.

[0139] Experimental results:

[0140] like Figure 2As shown, this heatmap displays the actual number of days observed for 21 soybean varieties to reach R6 stage under 10 different environmental treatments (T1-T10). It provides a clear and comprehensive view of the original structure and core characteristics of all experimental data, offering a visual basis for subsequent in-depth statistical modeling. This heatmap condenses a large amount of data (21 varieties × 10 treatments) into a single image, clearly showing the specific performance of each variety-treatment combination and avoiding information omissions. Color gradients preliminarily reveal the trends in the influence of environmental treatments, such as which treatments generally accelerate or delay development, and which varieties show stable or sensitive performance. The distribution of right-censored data (observations that did not reach R6 stage) is clearly displayed, intuitively illustrating the necessity and importance of using a survival analysis model capable of handling censored data.

[0141] Figure 2 The heatmap uses color intensity to represent the time (in days) required to reach R6 stage. The color gradients from light red (indicating shorter time and faster development) to dark brown (indicating longer time and slower development). Gray is typically used to indicate that the variety failed to reach R6 stage during the trial period under that treatment (i.e., censored data, recorded as NA). Rows represent 21 different soybean varieties, and columns represent 10 environmental treatments (T1 to T10). The T1-T8 treatments are predominantly light red, indicating that under these high CO2 and artificial light quality conditions, most varieties reached the R6 stage within a short period (approximately 70-80 days), with minimal inter-varietal variation. The T9 treatment column shows a distinct light red to yellowish-green hue, interspersed with a few gray and dark brown cells, indicating that under this treatment (normal CO2, full spectrum, long photoperiod), development in most varieties was significantly delayed, and some varieties failed to complete development, a stark contrast to T1-T8. The T10 treatment column is almost entirely gray / brown, strongly demonstrating that these extreme conditions (normal CO2, simulated natural light, short photoperiod) are completely unsuitable for soybean reproductive growth, severely inhibiting all varieties. The color pattern differences in the variety rows indicate inherent genetic variation among varieties. For example, some varieties reached the R6 stage under all treatments (e.g., jiyu86, wandou37), exhibiting broad adaptability; while some varieties were significantly redder under the T9 treatment than others (e.g., zhongdou63, zheng1307), showing particular sensitivity to long photoperiods.

[0142] This heatmap qualitatively and globally reveals the significant impact of environmental treatments on soybean growth stages and the substantial genotype × environment (G×E) interaction effect of different varieties. It first visually demonstrates the existence of the technical problem described in this invention (i.e., the combination of environmental factors and variety selection are crucial), and highlights the censoring features in the original data, thus naturally leading to the necessity of subsequent quantitative analysis using the Bayesian Weibull AFT model, an advanced method capable of properly handling such complex data.

[0143] like Figure 3 The figure shows Kaplan-Meier survival curves, used to visualize the cumulative survival probability of 21 soybean varieties reaching R6 stage over time under 10 environmental treatments. Each curve represents a treatment group. The steeper the curve's descent, the faster the population reaches R6 stage; the gentler the curve's descent, the slower the development; if the curve plateaus and does not reach zero, it indicates that some individuals never reach R6 stage (i.e., right censoring exists).

[0144] The curves for treatments T1 to T8 largely overlapped, all showing a sharp decline to near zero around day 70. This indicates that under these combinations of high CO2 (1000 ppm or 1400 ppm), suitable light quality (5:1 or 2:1), and short to medium photoperiods (6 h or 8 h), soybean populations developed rapidly and uniformly. The curve for treatment T9 (normal CO2, full spectrum, 16 h photoperiod) declined significantly more slowly, reaching a plateau after day 110. This indicates that this long photoperiod condition had a significant delaying effect on soybean development, and approximately one-third of the varieties failed to reach the R6 stage. The curve for treatment T10 (normal CO2, simulated natural light, 4 h photoperiod) was almost horizontal, with the survival probability consistently close to 1.0. This indicates that this short-day condition was extremely unsuitable for soybean reproductive growth, and almost all varieties failed to complete development within the experimental period. This figure qualitatively reveals the powerful regulatory role of the interaction between CO2 concentration, light quality, and photoperiod on soybean development, and preliminarily demonstrates the possibility that optimizing environmental combinations can significantly shorten the soybean growth period, laying a visual foundation for the subsequent construction of a precise quantitative model.

[0145] like Figure 4 The diagram shows the MCMC trace plot of key parameters of the Bayesian Weibull AFT model in this embodiment of the invention, used to visually evaluate the convergence and mixing performance of Markov Chain Monte Carlo (MCMC) sampling. It is a combination of multiple subplots, each corresponding to a key model parameter (such as the fixed effects coefficient, random effects standard deviation, and shape parameter). The horizontal axis of each subplot represents the number of MCMC iterations (typically from the first to the last, such as 4000), and the vertical axis represents the sampled value of the parameter in that iteration. The sampling trajectories of the four independent MCMC chains are represented by curves of different colors (such as black, blue, light blue, and gray). This trace plot provides intuitive visual evidence of a healthy and stable MCMC sampling process. The trajectories of all key parameters show excellent mixing and stability, with no signs of non-convergence. This result is consistent with... Figure 5The numerical results of the (R-hat diagnosis) are highly consistent, jointly confirming from both visual and quantitative dimensions that the Bayesian Weibull AFT model used in this invention has fully converged. Therefore, all parameter estimates, acceleration factors (AF), and predicted values ​​subsequently calculated based on this model are highly reliable and effective, providing a core computational guarantee for the scientific validity of the entire invention.

[0146] like Figure 5 The figure shown is a convergence diagnostic plot of the Bayesian Weibull AFT model in this embodiment of the invention, which evaluates the convergence of Markov chain Monte Carlo (MCMC) sampling using the R-hat statistic. This plot quantifies whether the MCMC sampling process adequately explores the posterior distribution of the parameters, ensuring that the posterior inference is based on a stable and convergent algorithm output; it demonstrates that the R-hat values ​​of all parameters are below the strict threshold of 1.05, indicating that multiple Markov chains are well-mixed, and the obtained parameter estimates (such as the acceleration factor AF) are stable and reliable statistical inference results, rather than computational errors; it provides a solid quality assurance for all subsequent analytical conclusions based on this model (environmental factor effects, variety adaptability, etc.), indicating that these findings reflect the true patterns in the data.

[0147] This diagnostic plot displays all model parameters (including fixed-effects coefficients, random-effects standard deviations, and Weibull shape parameters) on the vertical axis and their corresponding R-hat values ​​on the horizontal axis as a scatter plot. The clear red vertical dashed line in the plot is the reference line, located at R-hat=1.05, which is a widely accepted critical threshold for model convergence. Data points are typically distinguished by color and shape according to different parameter types (e.g., blue dots represent fixed-effects parameters, green triangles represent variety-specific random-effects parameters, and yellow squares represent model-specific parameters).

[0148] The results show that all data points are closely away from the red reference line of 1.05. This indicates that the R-hat values ​​of all parameters are very close to the ideal value of 1.0 and far below the critical criterion of 1.05. This figure provides strong evidence that the Bayesian Weibull AFT model constructed for this invention has undergone sufficient iterative sampling and has reached a stable convergence state. This result is consistent with... Figure 3 The numerical conclusions from the (MCMC trace plot) are highly consistent, jointly confirming from both visual and quantitative dimensions that the Bayesian Weibull AFT model used in this invention has fully converged. Therefore, all posterior statistics derived from this model (such as the acceleration factor of environmental factors, the random effect of species, etc.) are valid, reliable, and reasonably interpretable, laying a solid computational foundation for the analytical conclusions of the entire invention.

[0149] like Figure 6The figure shown is a goodness-of-fit verification plot for the posterior prediction test of this invention. The density overlay plot on the left shows that the density distribution of the posterior prediction has good consistency with the density distribution of the observed data, indicating that the model can reproduce the characteristics of the observed data well at the distribution level. The mean comparison plot on the right shows that the observed mean falls within the distribution range of the posterior prediction mean, indicating that the model's prediction of the "mean of soybean R6 stage time" is reasonable and has a good fitting effect at the mean level.

[0150] like Figure 7 As shown, this is the environmental factor fixed effect acceleration factor ( ) in the embodiment of the present invention. The posterior estimation forest plot is used to accurately and quantitatively analyze the independent effects of various environmental factor levels on soybean development rate. By using the reference treatment (T4) as a baseline, the acceleration or deceleration effects of other environmental levels on development time are calculated and visualized, providing specific effect values. ) and uncertainty range (95% CrI).

[0151] By checking whether the 95% confidence interval contains 1, the statistical significance of each environmental effect can be intuitively determined; this clearly indicates which changes in environmental factors can most effectively regulate the reproductive period, providing direct data support for the precise regulation of controlled environment agriculture (CEA).

[0152] The forest plot uses a fixed-effects acceleration factor ( The x-axis represents the area, and the y-axis represents the levels of various environmental factors. A prominent red vertical dashed line is located at... =1.0. This line represents the no-effect line; if the confidence interval of an effect crosses this line, it indicates that there is no significant difference between the control group and the control group. The effect of each environmental level is represented by a point (representing the posterior median estimate) and a horizontal line segment (representing the 95% confidence interval).

[0153] The results are as follows:

[0154] (1) CO2 effect: Compared with the reference level (1000 ppm), the point with CO2=400 ppm is far to the right of the reference line, and its AF value is much greater than 1 (1.46 [0.00, 1250.38]), and the entire confidence interval does not contain 1, indicating that reducing the CO2 concentration significantly slows down development. The point with CO2=1400 ppm is close to the reference line (0.99 [0.96, 1.02]), and the interval contains 1, indicating that further increasing the CO2 concentration has no significant effect on the development rate.

[0155] (2) Light quality effect: The effect of light quality = 5:1 is close to that of the reference (2:1) but the range is slightly higher than 1 (1.02 [0.99, 1.05]). The points for light quality = Full (full spectrum) and light quality = Natural (natural light) are both significantly located to the right of the reference line. The value is significantly greater than 1, indicating that both light quality conditions significantly slow down development, and the slowing effect of natural light is the strongest.

[0156] (3) Photoperiod effect: The effect of photoperiod = 6h is not significant, although =0.99 < 1, but the entire confidence interval includes 1 [0.96, 1.03]. Points with photoperiods of 16h and 4h are both significantly located to the right of the reference line. A value greater than 1 indicates that both excessively long and excessively short photoperiods will significantly slow down development.

[0157] This figure precisely quantifies the impact of various environmental factors deviating from reference levels on soybean development rate. The results indicate that insufficient CO2 concentration and unsuitable light quality and photoperiod are key limiting factors leading to delayed soybean development. This provides a clear scientific basis and quantitative targets for optimizing soybean growth period through precise regulation of these factors in environments such as plant factories.

[0158] like Figure 8 As shown, this is the random effect acceleration factor for 21 soybean varieties in this embodiment of the invention. This is a ranking and classification diagram of soybean varieties. The purpose is to reveal and quantify the inherent, environment-independent differences in genetic developmental rates among different soybean varieties, and to classify varieties based on their adaptability. After removing environmental effects, this diagram assesses whether each variety's inherent developmental rate is faster or slower relative to the "average variety," evaluating the inherent characteristics of the variety; based on the variety acceleration factor (… The size of the cultivar is used to sort all varieties from fastest to slowest, providing a priority order for variety selection and achieving objective sorting of varieties; it also identifies extreme types of varieties (such as "accelerated" varieties that develop extremely fast or "decelerated" varieties that are highly resistant to adversity), providing parental or varietal resources for targeted breeding and promotion in specific environments.

[0159] This diagram shows 21 varieties. The posterior estimates (points) and their 95% confidence intervals (line segments) are displayed in numerical order. A red vertical dashed line is located at... =1.0 represents the level of the "average variety". It is located to the left of the reference line ( Points with a value <1) are indicated in green, representing "accelerated" varieties; located to the right of the reference line ( >1) The dots are usually marked in red, representing "slow-down" varieties.

[0160] The results are as follows: The green dots on the left side of the chart (e.g., jiyu86) The confidence interval for this value lies entirely to the left of 1.0, indicating that the inherent developmental rate of this variety is significantly faster than average; the red dots on the right side of the chart (such as jinpin82, gandou7, huachun2) indicate that... The confidence intervals for these varieties are all to the right of 1.0, indicating that their intrinsic development rate is significantly slower than the average. Some varieties located in the center of the graph (such as wandou37, hefeng55, and keshan1) have confidence intervals that include 1.0, indicating that their intrinsic development rate is not significantly different from the average.

[0161] This figure clearly illustrates the significant genetic variation among soybean varieties. "Accelerating" varieties (such as jiyu86) are ideal for shortening the growth period and achieving rapid crop rotation. While "decelerating" varieties perform poorly under optimal conditions, they may possess genetic potential for stress adaptation (as confirmed by subsequent analysis). This figure provides breeders with direct and quantitative decision-making support for parent selection based on specific breeding objectives (speed or stability).

[0162] like Figure 9 The figure shown is a comparison of the predicted development time of varieties under optimal (T4) and adverse (T9) conditions in an embodiment of the present invention. The purpose is to visualize the response patterns of different soybean varieties to environmental stress and directly screen for superior varieties with adverse adaptability. It directly compares the performance differences of the same variety under optimal and adverse conditions to assess its stability (good stability = relatively good performance under adverse conditions); from numerous varieties, it selects those that perform best relatively under adverse (T9) conditions. These varieties are of extremely high value for coping with adverse environments or for promotion in less-than-ideal regions. This figure, in a very intuitive way, indicates to agricultural extension workers and breeders which varieties are more suitable for planting under specific adverse conditions.

[0163] The horizontal axis of this graph represents the predicted time (in days) to reach R6. The predicted values ​​for each variety under the T4 environment are represented by green dots / intervals, which are usually concentrated on the left side of the graph (short time); the predicted values ​​under the T9 environment are represented by red / orange dots / intervals, which are usually distributed on the right side of the graph (long time).

[0164] The results are as follows: the red dot (T9) of all varieties was clearly located to the right of its green dot (T4), indicating that the development of all varieties was delayed under the T9 stress, proving that T9 did indeed constitute a stress environment; the position of the red dot varied greatly among different varieties. For example, the red dots of varieties wandou37 and jiyu86 were relatively more to the left (~108-113 days), while the red dot of variety gandou7 was very much to the right (>120 days).

[0165] This means that although all varieties are affected by stress, wandou37 and jiyu86 are relatively less affected, exhibiting stronger stress adaptability. This figure not only reaffirms the main effect of environment, but more importantly, reveals the existence of variety-environment interactions. It visually identifies varieties like wandou37, which perform reasonably well under optimal conditions and are relatively optimal under stressful conditions, making them valuable materials for promotion in areas with long photoperiods, normal CO2, and other stressful environments, or as parents for stress-tolerant breeding.

[0166] like Figure 10 The image shown is a heatmap of bidirectional clustering analysis based on the Bayesian Weibull AFT model, depicting the time of 21 soybean varieties reaching R6 under 10 experimental treatments. This system reveals the complex interaction patterns between soybean varieties and environmental treatments and visualizes the adaptive differentiation patterns of varieties. Through bidirectional clustering, it simultaneously displays the similarity relationships between varieties and environmental treatments, revealing hidden G×E interaction patterns. Row annotations visually display the performance groupings (fast, medium, slow) of different varieties based on predicted performance, identifying variety resources with broad adaptability or specific adaptability. Column annotations display the specific factor combinations for each environmental treatment, identifying optimal and stressful environmental conditions. This provides an intuitive, data-driven scientific basis for variety breeding and environmental optimization.

[0167] This heatmap, built using the ComplexHeatmap package, is a multi-layered, comprehensive visualization chart.

[0168] Core heatmap matrix: Rows represent 21 soybean varieties, columns represent 10 experimental treatments (T1-T10), and the color of each cell indicates the predicted time (in days) for the corresponding variety to reach R6 under the corresponding treatment. The color scheme uses the Viridis color scheme to gradually change from purple (short time, fast development) to yellow (long time, slow development).

[0169] Row clustering tree (left side): Grouping varieties with similar environmental response patterns together to form variety groups through hierarchical clustering (Euclidean distance, fully connected method).

[0170] Column clustering tree (top): Groups environmental treatments with similar effect patterns together to form environmental clusters.

[0171] Row annotation (left): Based on the average prediction time of the variety under all treatments, the variety is divided into three performance groups, "Fast" (green), "Medium" (yellow) and "Slow" (red), using the ternary method to visually display the overall adaptability of the variety.

[0172] List notes (top): Use color coding to show the specific combination of factors for each environmental treatment: CO2 concentration: 400ppm (light yellow), 1000ppm (orange), 1400ppm (deep red); Light quality: 5:1 (light purple), 2:1 (purple), Full (deep purple), Natural (very deep purple); Photoperiod: 4h (light green), 6h (green), 8h (deep green), 16h (dark green).

[0173] Clustering trees categorize varieties into different groups and environmental treatments into different effect groups. Row clustering shows that some varieties (such as a branch in the clustering tree) perform well in all environments (overall color is purplish), belonging to widely adaptable varieties; while some varieties only perform poorly in certain specific environments, showing environment specificity. Column clustering shows that treatments T1-T8 (high CO2 concentration combined with suitable light quality and photoperiod) cluster together, with an overall color purplish, indicating that these conditions promote development; while treatments T9 and T10 (normal CO2 concentration combined with extreme light conditions) often form their own group, with an overall color yellowish, indicating that these conditions inhibit development. By identifying color anomalies in specific varieties under specific environments (such as a variety performing moderately in most environments but performing exceptionally well in a certain environment), specific variety-environment interactions can be identified, providing targets for precision breeding.

[0174] This bidirectional clustering heatmap is one of the core analytical results of this invention. It not only verifies the predictive power of the Bayesian Weibull AFT model, but more importantly, it reveals the complex patterns of variety-environment interactions through advanced visualization techniques. The trends shown in the graph corroborate the results of previous fixed-effects and random-effects analyses of varieties, jointly demonstrating that significant environmental adaptability differentiation does indeed exist among different soybean varieties. Breeders can use this graph to: 1) screen for broadly adaptable varieties that perform well in multiple environments as parents; 2) discover germplasm resources with specific adaptability that stand out under specific adverse conditions; and 3) select the most suitable varieties for the target environment. This invention, through this single graph, achieves the transformation from complex, multi-dimensional predictive data to actionable breeding insights, demonstrating the advanced nature and practical value of the method in analyzing complex agronomic problems.

[0175] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying the environmental adaptability and regulating the growth of soybean varieties, characterized in that, Including the following steps: S1: Conduct multi-environmental factor interaction experiments and collect and preprocess observational data including censored data; S2: Construct a Bayesian Weibull AFT model that considers multiple environmental factors and random differences in soybean varieties; S3: The Bayesian Weibull AFT model is fitted using the Markov chain Monte Carlo method, and model convergence diagnosis and goodness-of-fit verification are performed. S4: Extract the posterior distributions of multiple environmental factors and the posterior distributions of random differences in soybean varieties from the fitted Bayesian Weibull AFT model, and perform fixed-effects and random-effects analysis respectively; S5: Variety selection and environmental optimization based on the analysis results of fixed and random effects; S6: Use the fitted Bayesian Weibull AFT model to predict the expected reproductive time under any variety-environment combination, and perform variety-environment interaction analysis to generate a core heat map, recommending the optimal environment for the variety or configuring the most suitable variety for the environment.

2. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 1, characterized in that, The aforementioned environmental factors include, but are not limited to, CO2 concentration, light quality, and photoperiod.

3. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 2, characterized in that, The method for constructing a Bayesian Weibull AFT model that considers multiple environmental factors and random differences in soybean varieties is as follows: S21: Using the logarithm of survival time as the response variable, an AFT model is constructed by incorporating environmental fixed effects and variety random effects; S22: Correlation between AFT model and Weibull distribution; S23: Calculate the total likelihood function based on the probability density function and survival function of the Weibull distribution; S24: Define the prior distributions for environmental fixed effects and variety random effects in the AFT model; S24: Based on the prior distribution and the total likelihood function, calculate the posterior distribution of fixed effects and variety random effects in the AFT model using the Bayesian method.

4. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 3, characterized in that, The AFT model mentioned above is: In(T ij )=X ij β+γ i +ε ij Among them, T ij X represents the survival time, where i is the variety number, j is the j-th environment for variety i, and X represents the survival time. ij A design matrix for the fixed effects of environmental factors is provided, including main effect terms and interaction terms. β represents the environmental fixed effect coefficients, including both main effect coefficients and interaction effect coefficients. Let be the random intercept for the i-th variety. Let ε be the variance of the random intercept for each variety. ij ~Gumbel(0, 1 / α) is the error term, which follows a Gumbel distribution, and α is the Weibull shape parameter.

5. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 4, characterized in that, The method for linking the AFT model and the Weibull distribution is as follows: For the AFT model In(T) ij Inverse transformation yields T ij =exp(X ij β+γ i +ε ij Substituting the scale parameter of the Weibull distribution, let the scale parameter λ... ij =exp(X ij β+γ i ), then T ij ~Weibull(λ) ij ,α).

6. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 5, characterized in that, The method for calculating the total likelihood function based on the probability density function and survival function of the Weibull distribution is as follows: Among them, L ij (β,γ i ,α|data) represents the likelihood probability of a single variety, n i Let f(t) represent the total number of environmental factor levels, data represent the set of all observed data, and f(t) represent the total number of environmental factor levels. ij |λ ij Let S(t, α) represent the probability density function of the Weibull distribution. ij |λ ij ,α) is the survival function of the Weibull distribution, δ ij γ is the truncated indicator variable; V is the total number of varieties, {γ i } represents the set of random intercepts for all varieties.

7. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to any one of claims 3 to 6, characterized in that, The model convergence diagnosis is achieved using R-hat statistics and trace plots. The goodness-of-fit verification is performed by checking the goodness-of-fit through posterior prediction: Density superposition: The overlap between the density distribution curves of the observed data and the model prediction data is greater than or equal to the overlap threshold, indicating a good fit; Mean comparison: The deviation between the censoring ratio of the observed data and the censoring ratio of the model prediction data is less than or equal to the deviation threshold, indicating that the model can effectively handle censored data.

8. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to any one of claims 3 to 6, characterized in that, The method for analyzing fixed effects is as follows: The posterior distribution of the environmental fixed effect β was extracted from the fitted Bayesian Weibull AFT model, and the acceleration factor AF for each environmental factor level under the same species was calculated. Treatment =exp(β-β) ref ), β ref The environmental fixed effects coefficients representing the environmental reference baseline are extracted from the AF. Treatment The acceleration factor AF represents the interaction effect. β ; If the acceleration factor AF at a certain environmental factor level β <1 and AF β The pre-set confidence interval does not contain 1, indicating that the level of this environmental factor will accelerate soybean growth to the R6 stage; if AF β >1 and AF β The pre-set confidence interval does not contain 1, indicating that the level of this environmental factor will slow soybean growth to the R6 stage. If AF β If the pre-set confidence interval contains 1, it means that the level of this environmental factor has no significant effect on soybean growth.

9. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 8, characterized in that, The method for analyzing the random effects of varieties is as follows: Extract the variety random intercept γ from the fitted Bayesian Weibull AFT model. i The posterior distribution was used to calculate the variety-specific acceleration factor AF for each variety at the same environmental factor level. γ =exp(γ i -γ ref ), γ ref The random intercept of varieties serving as a reference benchmark; The fitness of all varieties at the same environmental factor level was ranked and classified: if AF γ <1 and AF γ If the preset confidence interval does not contain 1, then the variety is an acceleration type. If AF γ >1 and AF γ If the preset signal interval does not contain 1, then the variety is a deceleration type. γ If the preset confidence interval contains 1, then the variety is neutral.

10. The method for identifying the environmental adaptability and regulating the growth of soybean varieties according to claim 9, characterized in that, The method for variety selection and environmental optimization based on the analysis results of fixed effects and random effects is as follows: Based on the variety-specific acceleration factor AF obtained in step S4 γ To screen out superior varieties that exhibit stable performance and short growth cycles under different environments: The target environment was determined, and the variety-specific acceleration factor (AF) of all varieties under the target environment was analyzed. γ Selecting variety-specific accelerating factor AF γ Accelerated varieties with a value <1 and whose preset information interval does not contain 1 are considered superior varieties; Acceleration factor AF based on the environmental factor level obtained in step S4 β To determine the optimal combination of environmental factors that can maximize the acceleration of soybean growth and shorten the growth cycle: identify the target variety and iterate through all environmental factor levels for the target variety to determine the acceleration factor AF. β Select AF β For acceleration environment combinations where the value is less than 1 and the preset confidence interval does not contain 1, calculate the comprehensive benefit value = AF. β ×(1-energy consumption increment).

Images