Drip irrigation cotton nitrogen fertilizer demand estimation method based on remote sensing diagnosis
By combining Bayesian critical nitrogen concentration dilution curves with remote sensing technology, a dynamic nitrogen management model for cotton was constructed, which solved the problem of low nitrogen fertilizer utilization in cotton and achieved precise nitrogen management and efficient nitrogen fertilizer use.
Patent Information
- Application Number
- CN202511164352.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-20
- Publication Date
- 2025-12-23
AI Technical Summary
In existing technologies, the utilization rate of nitrogen fertilizer in cotton is low, and there is a lack of real-time monitoring of plant nitrogen concentration and quantitative understanding of nitrogen requirements under different planting density conditions, resulting in inaccurate nitrogen fertilizer management and affecting yield and environmental sustainability.
By combining Bayesian critical nitrogen concentration dilution curves with remote sensing technology, and using UAV remote sensing data and meteorological data, a model of nitrogen content in cotton plants was constructed. Combined with the Stacking ensemble learning model, the cumulative nitrogen deficit and nutrient index were accurately diagnosed, a dynamic prediction model of nitrogen demand under density effect was established, and a precise nitrogen application plan was formulated.
It enables accurate prediction of nitrogen demand under different density conditions, improves nitrogen fertilizer utilization efficiency, reduces nitrogen fertilizer input by 30-35%, maintains yield, and enhances economic and environmental benefits.
Smart Images
Figure CN121189676A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of agricultural technology, and in particular to a remote sensing diagnostic method for estimating nitrogen fertilizer requirements of drip-irrigated cotton. Background Technology
[0002] As an important economic crop, one of the key challenges to the efficient production of cotton is achieving precise nitrogen management to balance yield targets with environmental sustainability requirements. Currently, the nitrogen fertilizer utilization rate of cotton is generally only 30-40%, significantly lower than the theoretical potential. This inefficiency stems from two key problems in traditional nitrogen fertilizer management: firstly, the lack of real-time and precise monitoring of plant nitrogen concentration (PNC); and secondly, the lack of a quantitative understanding of the dynamics of nitrogen demand (Nfr) under different planting densities.
[0003] In recent years, the Critical Nitrogen Dilution Curve (CNDC), based on the relationship between nitrogen concentration and biomass, has become a key tool for diagnosing plant nitrogen nutrition status. The CNDC describes the minimum nitrogen concentration (%Nc) required for crops to maintain optimal growth at different biomass levels, providing a theoretical basis for precision nitrogen fertilizer management.
[0004] Existing cotton CNDC research is typically limited to specific growth conditions or single genotypes (G), environments (E), and management practices (M), lacking systematic evaluation under multiple combinations of conditions. This limitation not only restricts the applicability of cotton CNDC under diverse conditions but also increases the uncertainty and difficulty of its application in actual production. Traditional cotton CNDC construction methods often employ a segmented analysis approach: first, statistical methods are used to determine the critical point of nitrogen limitation, and then nonlinear regression is used to fit a negative exponential curve. However, this method requires high-quality and high-quantity data, and it is difficult to accurately capture the %Nc variation under different G×E×M conditions.
[0005] The nitrogen nutrient index (NNI), based on the critical nitrogen dilution curve (CNDC) theory, provides a theoretical framework for quantitatively characterizing crop nitrogen status. This index achieves precise quantification of nitrogen nutrient status by comparing the estimated plant nitrogen concentration with the critical nitrogen concentration required to reach maximum biomass. Cumulative nitrogen deficit (NAD), as the cumulative deficit of NNI, further reflects the cumulative effect of crop nitrogen nutrient status on yield. However, conventional NNI measurement requires destructive sampling and complex laboratory analysis, severely limiting its application in production practice.
[0006] Planting density significantly influences crop nitrogen allocation patterns and utilization efficiency by regulating light energy use efficiency, nutrient uptake dynamics, and population competition intensity. Studies have confirmed that high-density conditions exhibit characteristic source-sink relationship remodeling and nutrient competition patterns, leading to systematic changes in the critical nitrogen concentration curve and uptake dynamics. Research reveals that the response of maize's critical nitrogen concentration to density exhibits typical nonlinear characteristics, indicating that the density effect indirectly regulates crop nitrogen requirement by affecting canopy photosynthetic efficiency. However, for cotton, a crop with complex compensation mechanisms, the precise quantification of nitrogen dynamics and cumulative deficit (NAD) under wide-narrow row high-density cultivation conditions remains a theoretical gap.
[0007] In contemporary precision agriculture, the deep integration of remote sensing technology and advanced algorithms provides a new path for non-destructive diagnosis of crop nutrient status. Machine learning algorithms have made significant progress in agriculture, but single models still suffer from insufficient generalization ability when facing high-dimensional features and nonlinear relationships. Precisely quantifying cotton nitrogen requirements and establishing adaptive fertilization programs are core aspects of improving nitrogen use efficiency. Traditional fertilization programs are mainly based on statistical response curves, lacking accurate calculation of dynamic crop requirements. A dynamic nitrogen requirement estimation framework based on critical nitrogen concentration offers a new approach, but existing models do not fully consider the profound impact of density effects on nitrogen uptake and distribution.
[0008] Therefore, there is a need to provide an improved technical solution that addresses the shortcomings of the existing technology. Summary of the Invention
[0009] The purpose of this application is to provide a remote sensing diagnostic method for estimating nitrogen fertilizer requirements of drip-irrigated cotton, which is particularly suitable for the accurate diagnosis and control of nitrogen requirements during the growth period of cotton under different planting densities.
[0010] To achieve the above objectives, this application provides the following technical solution:
[0011] A remote sensing diagnostic method for estimating nitrogen fertilizer demand in drip-irrigated cotton includes the following steps:
[0012] S1. Bayesian critical nitrogen concentration dilution curve: Obtain cotton biomass and nitrogen concentration data under different genotypes, environmental and management conditions, and construct a universal critical nitrogen concentration dilution curve (CNDC) based on a Bayesian stratified model.
[0013] S2. Remote sensing estimation of plant nitrogen concentration: Acquire cotton growth data, UAV remote sensing RGB image data and meteorological data, and estimate the nitrogen content (PNC) of cotton plants based on the Stacking ensemble learning model to construct a cotton plant nitrogen concentration estimation model.
[0014] S3. Nitrogen nutrition status diagnosis: Based on the general critical nitrogen concentration dilution curve and the cotton plant nitrogen concentration estimation model, calculate the cumulative nitrogen deficit (NAD) and nitrogen nutrient index (NNI) under different density conditions;
[0015] S4. Dynamic prediction of nitrogen demand, including the following steps:
[0016] S41. Establish a quadratic function relationship model between the slope S and the effective accumulated temperature GDD;
[0017] S = a × GDD + b × GDD 2 +c (1);
[0018] In the formula, GDD is the effective accumulated temperature, and a, b, and c are coefficients;
[0019] S42. Construct a linear regression model of NAD and NNI under different densities;
[0020] NAD = S × (1 - NNI) (2);
[0021] In the formula, NAD is the cumulative nitrogen deficit, S is the effective accumulated temperature, and NNI is the nitrogen nutrition index.
[0022] S43. Based on the cumulative nitrogen deficit (NAD) and nitrogen recovery efficiency (REN), calculate the nitrogen demand (Nfr) and predict the nitrogen demand of cotton under different density conditions.
[0023] Nfr = NAD / REN (3);
[0024] S5. Determine the precise nitrogen application plan for cotton throughout its entire growth period under different densities based on the nitrogen requirement of cotton under different density conditions.
[0025] Existing research largely focuses on constructing the relationship between nitrogen nutrient index (NNI) and relative yield (RY) under single-density conditions, while the regulatory mechanism of density-nitrogen interaction on the critical NNI threshold remains to be elucidated. Ensemble learning, by combining the advantages of multiple basic learners, has demonstrated performance superior to single models in several fields. In particular, the Stacking framework, by introducing a meta-learning mechanism, can adaptively adjust the contribution weights of each basic learner, significantly improving prediction stability and accuracy. The Bayesian hierarchical model method allows for the simultaneous identification of nitrogen limitation critical points and fitting of CNDCs through a unified statistical framework, overcoming the dependence of traditional methods on data partitioning and critical point selection.
[0026] Furthermore, step S3 includes the following steps:
[0027] S31. Based on the general critical nitrogen concentration dilution curve constructed in step S1, calculate the critical nitrogen concentration;
[0028] S32. Calculate the nitrogen nutrient index (NNI) based on the plant nitrogen concentration estimated in step S2 and the critical nitrogen concentration calculated in step S31.
[0029] S33. Calculate the cumulative nitrogen deficit (NAD).
[0030] Furthermore, step S5 includes the following steps:
[0031] S51. Determine the total nitrogen application rate for different density treatments;
[0032] S52. Determine the nitrogen application ratio between the budding stage and the flowering and boll-forming stage;
[0033] S53. Determine the critical nitrogen application period based on accumulated temperature data;
[0034] S54. Implement coupled management of drip irrigation and fertilization;
[0035] The key nitrogen application periods and corresponding nitrogen application plans include:
[0036] During the transition from vegetative to reproductive growth at an accumulated temperature of 600-900℃, the nitrogen nutrient index (NNI) should be maintained within the range of 0.85-1.0. During the peak flowering and boll-forming period at an accumulated temperature of 900-1200℃, nitrogen application should be appropriately increased to address the rapid increase in nitrogen accumulation deficit (NAD). In the later stages of the boll-forming period at an accumulated temperature >1200℃, nitrogen application should be adjusted appropriately based on changes in nitrogen accumulation deficit (NAD) to prevent nitrogen deficiency in the later stages.
[0037] Furthermore, based on different genotypes, environmental and management conditions, a Bayesian hierarchical model was used to construct 30 specific critical nitrogen concentration dilution curves (CNDC). Several candidate curves were reconstructed using three methods: data mixing, differential quantification, and importance analysis. The best one was then selected as the universal critical nitrogen concentration dilution curve.
[0038] Furthermore, in step S1, the Bayesian hierarchical model comprises three levels: the first level describes the relationship between biomass and nitrogen content observations, where changes in biomass and nitrogen content are modeled using a linear-plateau function; the second level describes the parameters of the critical nitrogen concentration dilution curve based on a probability distribution, while also considering the variability of biomass and nitrogen concentration during the measurement period; and the third level incorporates prior information, including knowledge about the reasonable range of parameter values.
[0039] Furthermore, the model's first-level assumption is a biomass W. ij Follows the mean μ ij The sum and variance are Gaussian distribution:
[0040]
[0041] In equation (1), BMAX,i S represents the biomass that reaches its maximum (i.e., the biomass when nitrogen is no longer a limiting factor) on the i-th observation date. i The slope of the linear portion represents the rate of biomass response to an increase in nitrogen content. C,i It is the critical nitrogen concentration for the i-th observation date;
[0042] In the second stage of the model, it is assumed that the relationship between the critical nitrogen concentration and the maximum biomass follows a power function:
[0043]
[0044] In equation (2), a and b are the key model parameters;
[0045] To capture variability between measurement dates, assume B MAX,i The values of Si change with the date and follow a truncated normal distribution;
[0046] The third level of the model sets the prior distribution of parameters. Parameters a and b are uniformly distributed and set to range from 2 to 6 and from 0 to 0.5, respectively, to ensure the flexibility of the parameters within a reasonable range.
[0047] Furthermore, in step S1, for the Bayesian hierarchical model, parameter estimation is performed using the Markov Chain Monte Carlo (MCMC) algorithm. During the MCMC sampling process of the model, the first 20,000 iterations are discarded, and the posterior median and 95% confidence interval of the parameters are calculated based on the results of more than 30,000 subsequent iterations. The critical nitrogen concentration dilution curve (CNDC) formula is obtained as Nc = 4.74 × DW. -0.25 Where Nc is the critical nitrogen concentration and DW is the dry matter mass of the plant, in t / ha.
[0048] Furthermore, in step S2, the cotton growth data includes nitrogen content and biomass; the meteorological data includes daily maximum temperature, daily minimum temperature, and daily average temperature.
[0049] Furthermore, the UAV remote sensing RGB image data includes the extraction of vegetation index features and texture features from the RGB images. The index features include seven vegetation indices: Excess Green minus Excess Red (ExGR), Color Index Vegetation Extraction (CIVE), Green Leaf Area Index (GLA), Red-Green-Blue Vegetation Index (RGBVI), Normalized Green Channel Value (g), Excess Green Index (ExG), and Green Leaf Index (GLI). These seven vegetation indices were optimized and selected from 22 vegetation indices extracted through comprehensive feature extraction. In the comprehensive feature extraction stage, up to 22 vegetation indices were extracted from the UAV RGB images, covering different spectral features and calculation methods. The basic channel indices of these 22 vegetation indices include: r (normalized red channel value), g (normalized green channel value), and b (normalized blue channel value). Ratio-based indices include GRRI (Green-Red Ratio Index), RBRI (Red-Blue Ratio Index), and GBRI (Green-Blue Ratio Index). Difference-based indices include NDI (Normalized Difference Index), ExR (Excess Red Index), ExB (Excess Blue Index), ExG (Excess Green Index), and ExGR (Excess Green Minus Excess Red Index). Comprehensive indices include INT (Intensity Index), WI (Wardman Index), VEG (Vegetation Index), IKAW (Kawashima Vegetation Index), RGBVI (Red-Green-Blue Vegetation Index), MGRVI (Modified Green-Red Vegetation Index), GLI (Green Leaf Index), GRVI (Green-Red Vegetation Index), VARI (Visible Light Atmospheric Impedance Vegetation Index), CIVE (Color Index Vegetation Extraction), and GLA (Green Leaf Area Index). Considering the computational efficiency and prediction accuracy of the model, this invention uses a random forest algorithm to evaluate the importance of all vegetation indices. Based on a Gini coefficient greater than 8.0, the above seven key vegetation indices were finally selected for constructing the nitrogen concentration estimation model.
[0050] Furthermore, the texture features undergo two levels of processing. The first level is the extraction of original texture features: based on the Gray-Level Co-occurrence Matrix (GLCM) method, eight texture features are extracted from the three color channels (red, green, and blue) of the RGB image, including variance (VAR), entropy (EN), correlation (COR), homogeneity (HOM), second moment (SE), dissimilarity (DIS), contrast (CON), and mean (MEA); a total of 24 texture features are obtained (3 channels × 8 features).
[0051] Second level: Feature selection and optimization: Considering model efficiency and avoiding feature redundancy, this invention uses the random forest algorithm to evaluate feature importance. Based on the threshold standard of Gini coefficient greater than 8.0, the seven most discriminative (most critical) texture features are finally selected: BCON (blue channel contrast), BDIS (blue channel dissimilarity), BVAR (blue channel variance), RCON (red channel contrast), and BCOR (blue channel correlation).
[0052] Combining spectral features with texture parameters is a key approach to improving remote sensing estimation performance. Vegetation indices primarily reflect physiological characteristics such as chlorophyll content and biomass, while texture parameters capture morphological features such as canopy structure and leaf arrangement. These two types of parameters are theoretically complementary; as crop density increases, the contribution of texture features to physiological parameter prediction significantly increases, possibly due to the more complex canopy structure and light environment heterogeneity under high-density conditions. This invention systematically evaluates the synergistic effect of these two types of features within an ensemble learning framework, which is of great significance for constructing robust cotton nitrogen diagnostic models.
[0053] Furthermore, in step S2, the Stacking ensemble learning model has a two-layer structure: the first layer consists of six basic learners: XGBoost, Support Vector Regression (SVR), Random Forest (RF), Extreme Learning Machine (ELM), Stepwise Multiple Linear Regression (SMLR), and Long Short-Term Memory Network (LSTM); the second layer uses XGBoost as the meta-learner.
[0054] Furthermore, in step S41, at a density of 69,000 plants / ha, the slope S = -0.000149 × GDD 2 +0.54×GDD-211.37; at a density of 138,000 plants / ha, the slope S=-0.000198×GDD 2 +0.71×GDD-263.80; At a density of 207,000 plants / ha, the slope of S = -0.0000677×GDD 2 +0.48×GDD-149.24; at a density of 240,000 plants / ha, the slope S=-0.0000231×GDD 2 +0.43×GDD-167.27.
[0055] The beneficial effects of this invention are:
[0056] This invention organically combines Bayesian critical nitrogen concentration dilution curves, remote sensing diagnostic technology, and density effects to propose a comprehensive solution for precise nitrogen management of drip-irrigated cotton under different density conditions. It innovatively applies a Bayesian stratified model to construct critical nitrogen concentration dilution curves for cotton, overcoming the limitations of traditional CNDC construction methods. This achieves a precise description of cotton nitrogen dynamics under different genotypes, environmental conditions, and management practices. The constructed universal critical nitrogen concentration dilution curve is Nc = 4.74 × DW. -0.25 It has wide applicability;
[0057] By integrating the advantages of multiple algorithms through the Stacking ensemble learning framework, high-precision remote sensing estimation of nitrogen concentration in cotton plants was achieved.
[0058] This study revealed the spatiotemporal dynamic characteristics of cotton nitrogen nutrition index and cumulative deficit under different density conditions, quantified the regulatory effect of density on critical nitrogen nutrition index, and determined the critical NNI value for maximizing yield potential to be 0.85. A dynamic prediction model of nitrogen demand considering density effect was constructed. By quantifying the quadratic function relationship between the NAD-NNI regression slope and accumulated temperature, the model achieved accurate prediction of nitrogen demand under different density conditions, with a verification accuracy of over 97%.
[0059] Precision nitrogen application schemes for different densities were developed, which reduced nitrogen fertilizer input by 30-35% while maintaining yield, significantly improving nitrogen fertilizer utilization efficiency and demonstrating significant economic and environmental benefits. Attached Figure Description
[0060] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. Wherein:
[0061] Figure 1 This is a technical roadmap of the present invention;
[0062] Figure 2 This is an overview diagram of the Bayesian three-level hierarchical model;
[0063] Figure 3 Dilution curves for candidate universal critical nitrogen concentrations;
[0064] Figure 4 A schematic diagram of the Stacking integrated learning framework;
[0065] Figure 5 The graph shows the change of nitrogen nutrition index of cotton with effective accumulated temperature at different densities.
[0066] Figure 6 The graph shows the change in nitrogen nutrient index of cotton with effective accumulated temperature at a density of 207,000 plants / ha.
[0067] Figure 7 Curves showing the cumulative nitrogen deficit of cotton at different densities as a function of effective accumulated temperature.
[0068] Figure 8 The graph shows the cumulative nitrogen deficit of cotton as a function of effective accumulated temperature at a density of 207,000 plants / ha.
[0069] Figure 9 A framework diagram of a cotton nitrogen demand estimation model based on the critical nitrogen concentration dilution curve;
[0070] Figure 10 Linear-plateau regression diagram of nitrogen nutrient index and relative yield under different density conditions;
[0071] Figure 11 The graph shows the relationship between the NAD-NNI regression slope and the effective accumulated temperature under different density conditions. Detailed Implementation
[0072] The present application will now be described in detail with reference to the accompanying drawings and embodiments.
[0073] like Figure 1 As shown, a remote sensing diagnostic method for estimating nitrogen fertilizer demand in drip-irrigated cotton includes the following steps:
[0074] S1. Bayesian critical nitrogen concentration dilution curve: Obtain cotton biomass and nitrogen concentration data under different genotypes, environmental and management conditions, and construct a universal critical nitrogen concentration dilution curve (CNDC) based on a Bayesian stratified model.
[0075] S2. Remote sensing estimation of plant nitrogen concentration: Acquire cotton growth data, UAV remote sensing RGB image data and meteorological data, and estimate the nitrogen content (PNC) of cotton plants based on the Stacking ensemble learning model to construct a cotton plant nitrogen concentration estimation model.
[0076] S3. Nitrogen nutrition status diagnosis: Based on the general critical nitrogen concentration dilution curve and the cotton plant nitrogen concentration estimation model, calculate the cumulative nitrogen deficit (NAD) and nitrogen nutrient index (NNI) under different density conditions;
[0077] S4. Dynamic prediction of nitrogen demand;
[0078] S5. Determine the precise nitrogen application plan for cotton throughout its entire growth period under different densities based on the nitrogen requirement of cotton under different density conditions.
[0079] Furthermore, in step S1, the construction of the Bayesian critical nitrogen concentration dilution curve involves experimental design and data collection based on different genotypes, environments, and management conditions. In this embodiment, experiments were conducted on four genotypes (varieties): Lumian 24, Xinluzao 76, Xinluzao 79, and Xinluzao 83; four environments (locations and years): Shihezi, Shawan, Tacheng, and Manas, spanning 2018-2021; and different management practices (planting density and phosphorus / potassium application), including four planting densities of 69,000 plants / ha, 138,000 plants / ha, 207,000 plants / ha, and 240,000 plants / ha; and four phosphorus / potassium application strategies: P0K0, P1K1, P2K2, and P3K3. Aboveground samples were collected during key growth stages: budding, flowering, early boll formation, late boll formation, and boll opening, to measure dry matter and nitrogen concentration.
[0080] Furthermore, in S1, such as Figure 2 As shown, the Bayesian hierarchical model consists of three levels: the first level describes the relationship between biomass and nitrogen content observations, where changes in biomass and nitrogen content are modeled using a linear-plateau function; the second level describes the parameters of the critical nitrogen concentration dilution curve based on probability distributions, while also considering the variability of biomass and nitrogen concentration during the measurement period; the third level incorporates prior information, including knowledge about the reasonable range of parameter values.
[0081] The first-level hypothesis of the model is biomass W. ij Follows the mean μ ij The sum and variance are Gaussian distribution:
[0082]
[0083] In equation (1), B MAX,i S represents the biomass that reaches its maximum (i.e., the biomass when nitrogen is no longer a limiting factor) on the i-th observation date. i The slope of the linear portion represents the rate of biomass response to an increase in nitrogen content. C,i It is the critical nitrogen concentration for the i-th observation date;
[0084] In the second stage of the model, it is assumed that the relationship between the critical nitrogen concentration and the maximum biomass follows a power function:
[0085]
[0086] In equation (2), a and b are the key model parameters;
[0087] To capture variability between measurement dates, assume B MAX,i The values of Si change with the date and follow a truncated normal distribution;
[0088] The model's third-level parameter prior distributions are a ~ U(2,6) and b ~ U(0,0.5), where U represents a uniform distribution.
[0089] Furthermore, in S1, during parameter estimation and curve fitting of the Bayesian hierarchical model, the Markov Chain Monte Carlo (MCMC) algorithm implemented using the R package rjags was used for parameter estimation. During the MCMC sampling process, convergence was achieved after approximately 20,000 iterations based on the Gelman-Rubin diagnostics. The first 20,000 iterations were discarded, and an additional 30,000 iterations were run. Based on these iterations, the parameters (NC = aW) were calculated. -b The posterior median and 95% confidence interval of parameters a and b).
[0090] Furthermore, the universal critical nitrogen concentration dilution curve was constructed based on 30 specific CNDCs built under different genotypes, environmental and management conditions. Several candidate curves were reconstructed using three methods: data pooling, differential quantification, and importance analysis, and the relevant index R was selected from these. 2 The largest single line was used as the universal critical nitrogen concentration dilution curve. The final determined universal CNDC was Nc = 4.74 × DW. -0.25 Where DW is the plant dry matter mass (t / ha), parameter a is 4.74, and parameter b is 0.25. The goodness of fit R0 of this general curve on the modeling dataset is... 2 The goodness-of-fit R-value is 0.83 on the validation dataset. 2 The value of 0.88 indicates that it has good predictive performance and wide applicability.
[0091] Note that the general CNDC is a new curve reconstructed based on the data of 30 curves, rather than a single curve selected from the 30. (1) The data mixing method randomly mixes all the original data (including biomass and nitrogen content data under different genotypes, environmental and management conditions) corresponding to the 30 specific CNDCs to form a comprehensive dataset. Then, the mixed data is fitted using a Bayesian hierarchical model to obtain the posterior distribution of parameters a and b. Based on the statistical characteristics of the posterior distribution (mean, median, mode), three candidate general CNDCs are constructed respectively. (2) The importance analysis method first uses the random forest algorithm to evaluate the importance of parameters a and b of the 30 specific curves, and identifies the specific condition combination that contributes the most to the initial value of critical nitrogen concentration (parameter a) and dilution rate (parameter b); then, the original data corresponding to the curve with higher importance score are selected, the data is mixed, and then the Bayesian hierarchical model is applied to fit the data. Similarly, the general CNDC is constructed based on the statistical characteristics of the posterior distribution. (3) The difference quantification method uses the curve with the best fit among the 30 curves as a benchmark to analyze the differences between the other curves and the reference curve in the estimated nitrogen concentration. Statistical tests are used to identify curves that do not differ significantly from the reference curve. The original data corresponding to these curves are then selected for Bayesian modeling, and a general CNDC is constructed based on the posterior distribution parameters. The criterion for judging the best general CNDC is: goodness of fit (coefficient of determination R0). 2 The maximum value must also be considered; at the same time, the stability and prediction accuracy of the model (R² on the independent validation set) must be taken into account. 2 ≥8.0). Figure 3 Dilution curves of candidate universal critical nitrogen concentrations selected by the three methods; Figure 3 (b) The relevant index R 2 The largest and most independent R on the validation set 2 Since the value is 0.88, the final determined universal critical nitrogen concentration dilution curve (CNDC) is Nc = 4.74 × DW. -0.25 .
[0092] Furthermore, in this embodiment, the data source for the remote sensing estimation of nitrogen concentration in S2 plants was a three-year (2019-2021) fixed-point field trial conducted by the Shihezi University Agronomic Experiment Station. This field trial set four planting densities (69,000 plants / ha, 138,000 plants / ha, 207,000 plants / ha, and 240,000 plants / ha) and five nitrogen fertilizer levels (0, 199 kg / ha, 299 kg / ha, 402 kg / ha, and 506 kg / ha). RGB image data was acquired using a drone, and cotton plant samples were collected and their nitrogen concentration measured as reference data. A total of 600 cotton plant nitrogen content samples were obtained, randomly divided into 500 training samples and 100 validation samples.
[0093] Furthermore, the UAV remote sensing RGB image data includes vegetation index features and texture features extracted from the RGB images. Considering computational efficiency, the most critical features were selected from the fully extracted features. The key index features include seven vegetation indices: Excess Green Minus Excess Red Index (ExGR), Color Index Vegetation Extraction (CIVE), Green Leaf Area Index (GLA), Red-Green-Blue Vegetation Index (RGBVI), Normalized Green Channel Value (g), Excess Green Index (ExG), and Green Leaf Index (GLI). The key texture features include seven types: Blue Channel Contrast (BCON), Blue Channel Dissimilarity (BDIS), Blue Channel Variance (BVAR), Red Channel Contrast (RCON), and Blue Channel Correlation (BCOR). Meteorological data includes daily maximum temperature, daily minimum temperature, and daily average temperature. By comparing the prediction effects of different combinations of input variables, it was found that the combination of vegetation indices and texture features can effectively improve the model's prediction accuracy for cotton plant nitrogen content (PNC). When using only texture features, all models generally outperform those based solely on vegetation indices. The goodness-of-fit R-squared of the texture feature model on the validation set is [value missing]. 2 The values range from 0.85 to 0.90, while the goodness-of-fit R-value of the model based on vegetation indices is [value missing]. 2 The values are relatively low, ranging from 0.77 to 0.86.
[0094] Furthermore, such as Figure 4 As shown, in step S2, the Stacking ensemble learning model has a two-layer structure, and the specific construction steps are as follows:
[0095] a. Dataset partitioning: Divide the original training data into 6 subsets in chronological order; for each base learner, train using 5 subsets and use the remaining 1 for validation;
[0096] b. Selection and Training of Base Learners: Six algorithms were selected as base learners: XGBoost, Support Vector Regression (SVR), Random Forest (RF), Extreme Learning Machine (ELM), Stepwise Multiple Linear Regression (SMLR), and Long Short-Term Memory Network (LSTM), and trained on their respective training sets. XGBoost has strong feature selection and combination capabilities, can adaptively adjust the weights of each base learner, and has good control over overfitting. RF is good at handling nonlinear relationships and feature interactions. SVR performs well in handling high-dimensional features and noisy data. ELM has fast learning and generalization capabilities. SMLR can select important features and build models with strong interpretability. LSTM is particularly suitable for capturing long-term dependencies in time series data.
[0097] c. New training set generation: Each base learner makes predictions on the validation set, and combines the prediction results with the actual values to generate a new dataset for training the meta-learner.
[0098] d. Meta-learner training: The XGBoost meta-learner was trained using the newly generated training dataset to obtain the final cotton plant nitrogen content (PNC) prediction model; the model performance was evaluated on the validation dataset, and the results showed that the model achieved a goodness of fit R on the validation dataset. 2 =0.94, RMSE=0.23%, and rRMSE is only 5.90%, indicating that the model performance is excellent and has strong fitting and prediction capabilities for the data.
[0099] Furthermore, in step S3, the specific steps for diagnosing the nitrogen nutrition status of cotton are as follows:
[0100] S31. The general critical nitrogen concentration dilution curve Nc = 4.74 × DW constructed based on step S1. -0.25 Calculate the critical nitrogen concentration, where DW is the dry matter mass of the plant (t / ha);
[0101] S32. Calculate the nitrogen nutrient index (NNI), NNI = %N / Nc, where %N is the remote sensing estimated plant nitrogen concentration using the cotton plant nitrogen content (PNC) prediction model in step S2, and Nc is the critical nitrogen concentration; when NNI equals 1, it indicates that the crop is in the optimal nitrogen nutrient state; NNI>1 indicates nitrogen excess, while NNI<1 indicates nitrogen deficiency.
[0102] S33. Calculate the cumulative nitrogen deficit (NAD); NAD = Nca - Na, where Nca is the critical nitrogen accumulation (the product of critical nitrogen concentration and biomass), and Na is the remotely sensed estimated nitrogen accumulation (the product of estimated nitrogen concentration and biomass); when NAD is zero, it indicates that the nitrogen supply is in an optimal state; a positive NAD value reflects a nitrogen deficit, while a negative value indicates nitrogen luxury absorption.
[0103] The inventors also analyzed the spatiotemporal dynamics of nitrogen nutrient index (NNI) and nitrogen cumulative deficit (NAD) under different density conditions; such as Figures 5-8 As shown, taking the data source in step S1 of this embodiment as an example, the dynamic changes of NNI and NAD with effective accumulated temperature (GDD) under different densities (69,000 plants / ha, 138,000 plants / ha, 207,000 plants / ha, and 240,000 plants / ha) are analyzed. In the figure, D6.9 represents the treatment with a density of 69,000 plants / ha, D13.8 represents the treatment with a density of 138,000 plants / ha, D20.7 represents the treatment with a density of 207,000 plants / ha, and D24.0 represents the treatment with a density of 240,000 plants / ha. The results show:
[0104] (1) The nitrogen nutrient index (NNI) of cotton showed a significant density-responsive characteristic to the dynamic changes of growth days (GDD). The peak NNI values of all treatments generally appeared in the range of 900-1100 GDD, with a peak value range of 0.9-1.2. With the increase of nitrogen fertilizer application, the NNI curve generally showed an upward trend, indicating that the sufficiency of nitrogen supply gradually improved.
[0105] (2) Under low-density treatment (D6.9), NNI was more sensitive to nitrogen fertilizer. The N506 (506 kg / ha) treatment increased by an average of 25-30% compared with the N199 (199 kg / ha) treatment, and the differences between the nitrogen application levels were significant. However, as the planting density increased, this difference gradually weakened. Under the D24.0 (240,000 plants / ha) treatment, the difference in NNI between the N506 (506 kg / ha) and N199 (199 kg / ha) treatments decreased to less than 15%. High-density conditions partially alleviated the nitrogen supply shortage through population regulation.
[0106] (3) During the GDD 600-900℃ stage, the NNI of medium and high density treatments (138,000 plants / ha and 207,000 plants / ha) remained within the suitable range of 0.95-1.15, and the response to nitrogen fertilizer level was relatively stable. This period is the key stage for the transition of cotton from vegetative growth to reproductive growth, and suitable density is conducive to maintaining a relatively stable nitrogen nutrition status. In the later stage of growth (>1200 GDD), the NNI of all density treatments showed a downward trend. Even under the high nitrogen treatment (506 kg / ha), the NNI generally dropped below 0.8, with the ultra-high density treatment (240,000 plants / ha) showing the most significant decrease, dropping to 0.6-0.7.
[0107] (4) Nitrogen cumulative deficit (NAD) analysis showed that during the early growth stage (600-800℃), the NAD values of each treatment fluctuated relatively little and were close to zero, indicating that the nitrogen supply and demand of cotton was basically in dynamic equilibrium during this stage. However, with the increase of effective accumulated temperature (GDD), the NAD values gradually diverged, especially showing a significant upward trend during the 1000-1400℃ period, reflecting the sharp increase in the nitrogen demand of cotton during this stage.
[0108] (5) Establishment of the relationship between nitrogen nutrient index and relative yield: such as Figure 10As shown, the relationship between nitrogen nutrient index and relative yield under different densities was quantified using a linear-plateau model, and the critical NNI value for maximizing yield potential was determined to be 0.85. Analysis based on the linear-plateau model showed a significant piecewise functional relationship between the average nitrogen nutrient index (NNI-int) and relative yield (RY) throughout the cotton growth period (P<0.001). When NNI-int < 0.85, RY increased linearly with increasing NNI-int under different density treatments; when NNI-int ≥ 0.85, RY tended to stabilize and reach a plateau value. Specifically, below the critical value (NNI-int = 0.85), the slopes of the D6.9 (69,000 plants / ha), D13.8 (138,000 plants / ha), D24.0 (240,000 plants / ha), and D20.7 (207,000 plants / ha) treatments were 0.97, 0.51, 0.50, and 0.95, respectively, indicating significant differences in nitrogen response efficiency among different densities. Correspondingly, the RY plateau values reached for each density treatment were 0.79, 0.91, 0.91, and 0.91, respectively, with goodness-to-fit (R0) values of [missing value]. 2 The values were 0.86, 0.79, 0.73, and 0.88, respectively. All density treatments showed a clear critical value effect, that is, an inflection point occurred at the critical point of NNI-int = 0.85, indicating that there is a universal nutrient threshold for cotton's nitrogen requirement regardless of the planting density.
[0109] Furthermore, such as Figure 9 As shown, step S4, dynamic prediction of nitrogen demand, includes the following steps:
[0110] S41. Under different densities, establish a linear regression model between NAD and NNI to obtain the regression slope S;
[0111] S = a × GDD + b × GDD 2 +c (1);
[0112] In the formula, GDD is the effective accumulated temperature, and a, b, and c are coefficients;
[0113] The quadratic function relationship between the slope S and the effective accumulated temperature (GDD) differs for different density treatments:
[0114] At a density of 69,000 plants / ha, the slope S = -0.000149 × GDD 2 +0.54×GDD-211.37;
[0115] At a density of 138,000 plants / ha, the slope S = -0.000198 × GDD 2 +0.71×GDD-263.80;
[0116] At a density of 207,000 plants / ha, the slope of S is -0.0000677 × GDD. 2 +0.48×GDD-149.24;
[0117] At a density of 240,000 plants / ha, the slope S = -0.0000231 × GDD 2 +0.43×GDD-167.27;
[0118] Under different density treatments, a significant nonlinear correlation was observed between the regression slope S and the effective accumulated temperature (GDD). As density increased, the opening of the quadratic function curve gradually widened, indicating that the regression slope S was more sensitive to GDD under high-density conditions. All four density treatments exhibited a typical saturation growth trend: the regression slope S rose rapidly in the GDD range of 600-800℃, then the growth rate gradually slowed, and stabilized in the 1200-1400℃ range. Notably, the initial values (50-100) and saturation values (300-400) of each treatment group showed consistency, while the moderate fluctuations in the regression slope S (2.31-4.54) reflected the differentiated response characteristics of different treatments to temperature accumulation.
[0119] S42. Construct a linear regression model of NAD and NNI under different densities;
[0120] NAD = S × (1 - NNI) (2);
[0121] In the formula, NAD represents the cumulative nitrogen deficit, S represents the effective accumulated temperature, and NNI represents the nitrogen nutrient index; for example, Figure 11 As shown, nitrogen nutrition index (NNI) and nitrogen accumulation deficit (NAD) at different growth stages exhibit a significant negative correlation, with a coefficient of determination (R²) of 1 / 2 ≤ ... 2 All are above 0.97;
[0122] S43. Based on the cumulative nitrogen deficit (NAD) and nitrogen recovery efficiency (REN), calculate the nitrogen demand (Nfr) and predict the nitrogen demand of cotton under different density conditions.
[0123] Nfr = NAD / REN (3);
[0124] The nitrogen recovery efficiency (REN) is calculated as follows: REN = (fertNa - unfertNa) / Nrate, where fertNa represents the actual nitrogen accumulation in the aboveground part of the nitrogen-treated plant (kg / ha), unfertNa is the corresponding value for the non-nitrogen-treated plant, and Nrate is the amount of nitrogen applied (kg / ha).
[0125] Density makes a substantial contribution to this invention. The density-specific S-GDD quadratic function model is the core of dynamic nitrogen demand prediction. Density directly regulates the regression slope S of the NAD-NNI linear regression model, and the S value varies significantly under different densities. Although the critical NNI value for maximizing yield potential is uniformly 0.85, the amount of nitrogen required to reach this value varies significantly under different densities, which is the key to precise nitrogen application.
[0126] Model Validation To verify the predictive stability and interannual applicability of the model, the fixed-point field test data from the aforementioned Shihezi University Agronomic Experiment Station over three years (2019-2021) were used to conduct multi-year validation analysis on the nitrogen accumulation deficit (NAD) of cotton under different density treatments (69,000 plants / ha, 138,000 plants / ha, 207,000 plants / ha, and 240,000 plants / ha).
[0128] The results show that the goodness of fit (R0) between the model predictions and the measured values is high under all density treatments. 2 All reached above 0.97 (R 2 The predicted values were 0.98, 0.97, 0.98, and 0.99 respectively, and the data points were evenly distributed on both sides of the 1:1 line. Within the cotton nitrogen accumulation deficit (NAD) range of -150 to 100 kg / ha, the predicted values and measured values showed a high degree of agreement, indicating that the model can accurately predict the nitrogen accumulation deficit of cotton under different planting densities and years. The distribution pattern of the validation data points in different years (2019-2021) was consistent, and the dispersion between the predicted and measured values was small, further validating the model's prediction accuracy.
[0129] The data sources in this application cover field trials from multiple locations, years, genotypes, and management practices, demonstrating strong representativeness and breadth. Specifically, they include:
[0130] 1. Multi-environment and multi-year validation:
[0131] Four typical cotton-growing areas (Shihezi, Shawan, Tacheng, and Manas) were set up, spanning four growing seasons from 2018 to 2021, covering different climate and soil conditions, ensuring that the data reflected regional environmental differences.
[0132] 2. Multigenotype experimental materials:
[0133] The test materials included varieties such as Lumian 24, Xinluzao 76, Xinluzao 79, and Xinluzao 83, covering representative genotypes widely used in local production. The introduction of different varieties enabled the constructed critical nitrogen concentration dilution curves to reflect the influence of genotype differences on the nitrogen dilution effect.
[0134] 3. Design of multiple management measures:
[0135] Planting density: Four levels were set, ranging from 69,000 plants / ha to 240,000 plants / ha, covering low-density to high-density cultivation patterns.
[0136] Nitrogen fertilizer level: designed with a gradient from 195-506 kg / ha based on farmers' conventional fertilizer application rate (approximately 299 kg / ha).
[0137] Phosphorus and potassium management strategies included four combinations: 100% drip application of phosphorus and potassium fertilizer during the budding stage (P0K0), 25% drip application during the budding stage and 75% drip application during the flowering and boll-forming stage (P1K1), 50% drip application during the budding stage and 50% drip application during the flowering and boll-forming stage (P2K2), and 75% drip application during the budding stage and 25% drip application during the flowering and boll-forming stage (P3K3). The regulatory role of nitrogen and nutrient application strategies on nitrogen dilution effect was systematically evaluated.
[0138] The universal critical nitrogen concentration dilution curve (CNDC) constructed using the above data, and the prediction of nitrogen requirements for cotton under different density conditions, demonstrate universality and stability, as shown in:
[0139] 1. The general-purpose CNDC achieves R on both the modeling and validation sets. 2 =0.83 and 0.88, which show strong explanatory power.
[0140] 2. Under different density and year conditions, the validation accuracy of both the nitrogen nutrient index and the cumulative deficit model remained at R. 2 A value of ≥0.97 indicates that the method has good applicability across regions and years, and can be extended to cotton-growing areas and even wider regions.
[0141] This application innovatively applies a Bayesian stratified model to construct a critical nitrogen concentration dilution curve for cotton, overcoming the limitations of traditional CNDC construction methods. It achieves an accurate description of cotton nitrogen dynamics under different genotypes, environmental, and management conditions, and the constructed universal critical nitrogen concentration dilution curve Nc = 4.74 × DW. -0.25 It has wide applicability;
[0142] This application integrates the advantages of multiple algorithms through the Stacking ensemble learning framework, achieving high-precision remote sensing estimation of nitrogen concentration in cotton plants, and validating the goodness of fit R on the dataset. 2 The accuracy reached 0.94, and the RMSE was only 0.23%, which improved the prediction accuracy by 20-25% compared with a single feature.
[0143] This application reveals the spatiotemporal dynamic characteristics of the nitrogen nutrition index and cumulative deficit of cotton under different density conditions, quantifies the regulatory effect of density on the critical nitrogen nutrition index, and determines the critical NNI value for maximizing yield potential to be 0.85.
[0144] This application constructs a dynamic prediction model for nitrogen demand that considers the density effect. By quantifying the quadratic function relationship between the NAD-NNI regression slope and accumulated temperature, it achieves accurate prediction of nitrogen demand under different density conditions, with a verification accuracy of over 97%.
[0145] This application develops a precise nitrogen application scheme for different densities, which reduces nitrogen fertilizer input by 30-35% while maintaining yield, significantly improving nitrogen fertilizer utilization efficiency and having obvious economic and environmental benefits.
[0146] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A remote sensing diagnostic method for estimating nitrogen fertilizer requirements in drip-irrigated cotton, characterized in that, Includes the following steps: S1. Bayesian critical nitrogen concentration dilution curve: Obtain cotton biomass and nitrogen concentration data under different genotypes, environmental and management conditions, and construct a universal critical nitrogen concentration dilution curve (CNDC) based on a Bayesian stratified model. S2. Remote sensing estimation of plant nitrogen concentration: Acquire cotton growth data, UAV remote sensing RGB image data and meteorological data, and estimate the nitrogen content (PNC) of cotton plants under various conditions based on the Stacking ensemble learning model, and construct a cotton plant nitrogen concentration estimation model. S3. Nitrogen nutrition status diagnosis: Based on the general critical nitrogen concentration dilution curve and the cotton plant nitrogen concentration estimation model, calculate the cumulative nitrogen deficit (NAD) and nitrogen nutrient index (NNI) under different density conditions; S4. Dynamic prediction of nitrogen demand, including the following steps: S41. Establish a quadratic function relationship model between the slope S and the effective accumulated temperature GDD; S=a×GDD+b×GDD 2 +c (1); In the formula, GDD is the effective accumulated temperature, and a, b, and c are coefficients; S42. Construct linear regression models of NAD and NNI under different densities; NAD = S × (1 - NNI) (2); In the formula, NAD is the cumulative nitrogen deficit, S is the effective accumulated temperature, and NNI is the nitrogen nutrition index. S43. Based on the cumulative nitrogen deficit (NAD) and nitrogen recovery efficiency (REN), calculate the nitrogen demand (Nfr) and predict the nitrogen demand of cotton under different density conditions. Nfr = NAD / REN (3); S5. Determine the precise nitrogen application plan for cotton throughout its entire growth period under different densities based on the nitrogen requirement of cotton under different density conditions.
2. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that, Step S3 includes the following steps: S31. Based on the general critical nitrogen concentration dilution curve constructed in step S1, calculate the critical nitrogen concentration; S32. Calculate the nitrogen nutrient index (NNI) based on the plant nitrogen concentration estimated in step S2 and the critical nitrogen concentration calculated in step S31. S33. Calculate the cumulative nitrogen deficit (NAD).
3. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that, Step S5 includes the following steps: S51. Determine the total nitrogen application rate for different density treatments; S52. Determine the nitrogen application ratio between the budding stage and the flowering and boll-forming stage; S53. Determine the critical nitrogen application period based on accumulated temperature data; S54. Implement coupled management of drip irrigation and fertilization; The key nitrogen application periods and corresponding nitrogen application plans include: During the transition from vegetative to reproductive growth when the accumulated temperature is 600-900℃, the nitrogen nutrient index (NNI) should be maintained within the range of 0.85-1.
0. During the full bloom and boll-forming period when the accumulated temperature is 900-1200℃, the amount of nitrogen applied should be appropriately increased to address the rapid increase in nitrogen accumulation deficit (NAD). During the later stages of the boll-forming period when the accumulated temperature is >1200℃, the amount of nitrogen applied should be adjusted appropriately according to changes in nitrogen accumulation deficit (NAD) to prevent nitrogen deficiency in the later stages.
4. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that: Based on different genotypes, environmental and management conditions, 30 specific critical nitrogen concentration dilution curves (CNDC) were constructed using a Bayesian hierarchical model. Several candidate curves were reconstructed using three methods: data mixing, differential quantification, and importance analysis. The best one was then selected as the universal critical nitrogen concentration dilution curve.
5. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 4, characterized in that: In step S1, the Bayesian hierarchical model comprises three levels: the first level describes the relationship between biomass and nitrogen content observations, where changes in biomass and nitrogen content are modeled using a linear-plateau function; the second level describes the parameters of the critical nitrogen concentration dilution curve based on probability distributions, while also considering the variability of biomass and nitrogen concentration during the measurement period; and the third level incorporates prior information, including knowledge about the reasonable range of parameter values.
6. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 5, characterized in that: The first-level hypothesis of the model is biomass W. ij Follows the mean μ ij The sum and variance are Gaussian distribution: In equation (1), B MAX,i S represents the biomass that reaches its maximum (i.e., the biomass when nitrogen is no longer a limiting factor) on the i-th observation date. i The slope of the linear portion represents the rate of biomass response to an increase in nitrogen content. C,i It is the critical nitrogen concentration for the i-th observation date; In the second stage of the model, it is assumed that the relationship between the critical nitrogen concentration and the maximum biomass follows a power function: In equation (2), a and b are the key model parameters; To capture variability between measurement dates, assume B MAX,i The values of Si change with the date and follow a truncated normal distribution; The third level of the model sets the prior distribution of parameters. Parameters a and b are uniformly distributed and set to range from 2 to 6 and from 0 to 0.5, respectively, to ensure the flexibility of the parameters within a reasonable range.
7. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that: In step S1, for the Bayesian hierarchical model, parameter estimation is performed using the Markov Chain Monte Carlo (MCMC) algorithm. During the MCMC sampling process of the model, the first 20,000 iterations are discarded, and the posterior median and 95% confidence interval of the parameters are calculated based on the results of more than 30,000 subsequent iterations. The critical nitrogen concentration dilution curve (CNDC) formula is obtained as Nc = 4.74 × DW. -0.25 Where Nc is the critical nitrogen concentration and DW is the dry matter mass of the plant, in t / ha.
8. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that: In step S2, cotton growth data includes nitrogen content and biomass; The UAV remote sensing RGB image data includes vegetation index features and texture features extracted from RGB images; the index features include 7 vegetation indices, namely Excess Green Minus Excess Red Index (ExGR), Color Index Vegetation Extraction (CIVE), Green Leaf Area Index (GLA), Red Green Blue Vegetation Index (RGBVI), Normalized Green Channel Value (g), Excess Green Index (ExG), and Green Leaf Index (GLI). Texture features were extracted from the red, green, and blue color channels of the RGB image using the Gray-Level Co-occurrence Matrix (GLCM) method. These features were: Blue Channel Contrast (BCON), Blue Channel Dissimilarity (BDIS), Blue Channel Variance (BVAR), Red Channel Contrast (RCON), and Blue Channel Correlation (BCOR). Meteorological data included daily maximum temperature, daily minimum temperature, and daily average temperature.
9. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that: In step S2, the Stacking ensemble learning model has a two-layer structure: the first layer consists of six basic learners: XGBoost, Support Vector Regression (SVR), Random Forest (RF), Extreme Learning Machine (ELM), Stepwise Multiple Linear Regression (SMLR), and Long Short-Term Memory Network (LSTM); the second layer uses XGBoost as the meta-learner.
10. The method for estimating nitrogen fertilizer demand of drip-irrigated cotton using remote sensing diagnostics according to claim 1, characterized in that: In step S41, at a density of 69,000 plants / ha, the slope S = - 0.000149×GDD 2 +0.54×GDD-211.37; at a density of 138,000 plants / ha, the slope S=-0.000198×GDD 2 +0.71×GDD-263.80; At a density of 207,000 plants / ha, the slope S = -0.0000677×GDD 2 +0.48×GDD-149.24; at a density of 240,000 plants / ha, the slope S=-0.0000231×GDD 2 +0.43×GDD-167.27.