Hainan island wild tea tree growth model construction method and system

Abstract

The invention belongs to the field of agricultural technology and ecological monitoring, and particularly discloses a Hainan island wild tea tree growth model construction method and system.The method comprises the steps that firstly, a microenvironment competition factor is obtained by quantifying a local competition index based on diameter-level distribution in a fixed quadrat around a target tea tree; selecting a preset cubic function growth mechanism model to calculate a theoretical growth index value according to a vegetation vertical zone corresponding to the altitude of the tea tree; then fusing the local competition index, the theoretical growth index value, the altitude, the canopy density and the growth environment type to construct a feature vector so as to train a gradient lifting decision tree model, and establishing a multi-factor growth state prediction model; and finally, predicting the growth state of an unknown tea tree by using the model. According to the method, the ecological mechanism and machine learning are effectively fused, the growth prediction precision is remarkably improved, and a quantitative basis is provided for precise conservation.

Description

Technical Field

[0001] This invention relates to the fields of agricultural technology and ecological monitoring, specifically to a method and system for constructing a growth model of wild tea trees on Hainan Island. Background Technology

[0002] Wild tea trees are important germplasm resources for tea and components of forest ecosystems, possessing strategic value for biodiversity conservation, tea breeding innovation, and the sustainable development of the tea industry. Hainan Island, as my country's only tropical island, boasts abundant wild tea tree resources due to its unique hydrothermal conditions, primarily distributed sporadically or in patches along the central mountains. Accurately understanding their growth status and dynamics is a prerequisite for implementing scientific protection and sustainable utilization.

[0003] Currently, research on wild tea trees mainly focuses on resource surveys and distribution records, while growth analysis generally employs traditional regression models. Existing techniques primarily describe growth patterns by constructing empirical formulas (such as linear or power function relationships) between single morphological indicators like tree height, diameter at breast height (DBH), and crown width. However, these methods have significant limitations: First, traditional models are mostly static descriptions, failing to fully consider the combined effects of dynamic biotic and abiotic factors such as microenvironmental competition and human disturbance, especially the quantification of intraspecific competitive pressure remains a gap; second, models typically assume uniform growth relationships across the entire region, ignoring the differentiation of adaptive growth strategies caused by varying environmental stresses at different altitudes, resulting in insufficient model universality and predictive accuracy; third, existing methods can only predict growth status and cannot reveal the dominant limiting factors affecting growth, making it difficult to support precise conservation management decisions. Summary of the Invention

[0004] To address this issue, the present invention provides a method and system for constructing a growth model of wild tea trees on Hainan Island, thereby solving the aforementioned technical problems.

[0005] This invention provides a method for constructing a growth model of wild tea trees on Hainan Island, comprising the following steps: For a target wild tea tree individual, its microenvironmental competition factors are obtained and quantified. The microenvironmental competition factors include at least the local competition index calculated based on the diameter distribution of all other wild tea trees within a surrounding fixed radius quadrat. Obtain the altitude information of the target wild tea tree, and based on the preset vegetation vertical zone to which the altitude information belongs, select the corresponding preset growth mechanism model to calculate one or more theoretical growth index values ​​of the target wild tea tree. The growth mechanism model is a curve regression model that describes the relationship between tree height, crown width, diameter at breast height and ground diameter. Construct a feature vector, which includes at least the local competition index, the theoretical growth index value, the altitude information, the canopy closure, and the growth environment type; use the feature vector as input features, and use one or more actual growth morphology indicators of wild tea trees as prediction targets, train a gradient boosting decision tree model to establish a multi-factor growth state prediction model. For each wild tea tree individual to be evaluated, its corresponding feature vector is obtained and input into the trained multi-factor growth state prediction model, which then outputs the predicted value of its growth state.

[0006] Preferably, a circular quadrat with a radius of R is drawn with the target wild tea tree as the center; The diameter at breast height (DBH) of all other wild tea trees within the sample plot was counted, and each surrounding tea tree was assigned a competitive weight value according to a predetermined diameter weight table. The principle for setting the competitive weight value was that the diameter of young trees in the rapid growth and resource competition stage was given the highest weight, while the weight of seedlings and large-diameter trees was relatively low.

[0007] Preferably, the diameter weight table is as follows: the weight W1 of Grade I seedlings ranges from 0.5 to 0.8; the weight W2 of Grade II saplings ranges from 0.9 to 1.2; the weight W3 of Grade III medium trees ranges from 0.6 to 0.9; and the weight W4 of Grade IV and above large trees ranges from 0.3 to 0.6, and satisfies W2>W1, W2>W3, W2>W4.

[0008] Preferably, the growth mechanism model is a cubic function curve model based on different altitude zones h, specifically including: In the lowland rainforest zone where h < 700 meters, the tree height-crown width model is H = a1 + b1C + c1C² + d1C³. In the mountain rainforest zone where 700m ≤ h < 1300m, the diameter at breast height (DBH) - diameter at ground level (D) model is D = a² + b²d + c²d² + d²d³. In the mossy dwarf forest zone at the summit with h ≥ 1300 meters, the diameter at breast height (DBH) - diameter at ground level (D) model is D = a³ + b³d + c³d² + d³d³. Where H represents tree height, C represents crown width, d represents ground diameter, D represents diameter at breast height, and a1, b1, c1, d1, a2, b2, c2, d2, a3, b3, c3, and d3 are obtained by fitting statistical data.

[0009] Preferably, the establishment of the multi-factor growth state prediction model combines the theoretical growth index values ​​with the corresponding actual measured values ​​as input features; the theoretical growth index values ​​include at least the theoretical diameter at breast height (DBH) value calculated from the ground diameter using the DBH-ground diameter model, and / or the theoretical tree height value calculated from the crown width using the tree height-crown width model.

[0010] Preferably, the growth environment type is a categorical variable, which is digitized through unique thermal coding, and its types include at least: secondary forest, tea garden, roadside, and artificial forest.

[0011] Preferably, the method also includes tree age estimation. Specifically, for shrub-type wild tea trees whose diameter at breast height (DBH) cannot be directly measured, their ground diameter (DBH) is measured, and their theoretical DBH is calculated using the DBH-DBH growth mechanism model in high-altitude zones. Then, their tree age is estimated based on the DBH-age relationship model.

[0012] Preferably, it further includes outputting the diagnostic results of the dominant factors affecting its growth status based on the feature importance ranking obtained from model training. Specifically, the feature importance ranking is performed by extracting and outputting the contribution ranking of all input features to the prediction target after the multi-factor growth status prediction model is trained, so as to quantitatively identify the dominant environmental factors and competing factors affecting the growth of wild tea trees.

[0013] Preferably, when the diagnostic results of the dominant factor show that the local competition index is the key limiting factor, it is recommended to implement artificial thinning; when the canopy closure is the key limiting factor, it is recommended to implement forest gap thinning or light management.

[0014] In another aspect, this application also provides a system for constructing a growth model of wild tea trees on Hainan Island, comprising: The local competition index acquisition module is used to acquire and quantify the microenvironmental competition factors for a target wild tea tree individual. The microenvironmental competition factors include at least the local competition index calculated based on the diameter distribution of all other wild tea trees within a surrounding fixed radius quadrat. The curve regression model generation module is used to obtain the altitude information of the target wild tea tree, and based on the preset vegetation vertical zone to which the altitude information belongs, select the corresponding preset growth mechanism model to calculate one or more theoretical growth index values ​​of the target wild tea tree. The growth mechanism model is a curve regression model that describes the relationship between tree height, crown width, diameter at breast height and ground diameter. The prediction model building module is used to construct a feature vector, which includes at least the local competition index, the theoretical growth index value, the altitude information, the canopy closure, and the growth environment type. Using the feature vector as input features and one or more actual growth morphology indicators of wild tea trees as prediction targets, a gradient boosting decision tree model is trained to establish a multi-factor growth state prediction model. The prediction output module is used to obtain the corresponding feature vector of the wild tea tree individual to be evaluated, input it into the trained multi-factor growth state prediction model, and output the predicted value of its growth state.

[0015] This invention integrates a cubic function growth mechanism model with a gradient boosting decision tree machine learning algorithm, and creatively introduces a local competition index to quantify intraspecific competition in the microenvironment, constructing a multi-factor collaborative model for predicting and diagnosing the growth status of wild tea trees. This method effectively overcomes the limitations of traditional regression models that ignore environmental heterogeneity and dynamic competition, significantly improving the prediction accuracy of key growth indicators such as tree height and diameter at breast height (DBH) at different altitudes. Furthermore, this invention achieves quantitative diagnosis of growth limiting factors through feature importance analysis, clearly identifying dominant factors such as competitive pressure and light conditions, and directly recommending precise conservation measures such as artificial thinning or forest canopy opening. This provides full-chain support from growth prediction to causal diagnosis and management decision-making, offering a scientific and technical solution for the scientific protection and sustainable utilization of wild tea tree resources. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A flowchart illustrating a method for constructing a growth model of wild tea trees on Hainan Island, as provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the GBDT model training process provided in an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of a system for constructing a growth model of wild tea trees on Hainan Island, provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] like Figure 1 As shown in the figure, this invention discloses a method 100 for constructing a growth model of wild tea trees on Hainan Island, including the following steps: S1. For the target wild tea tree individual, obtain and quantify its microenvironmental competition factors. The microenvironmental competition factors include at least the local competition index calculated based on the diameter distribution of all other wild tea trees within the surrounding fixed radius quadrats. S2, obtain the altitude information of the target wild tea tree, and based on the preset vegetation vertical zone to which the altitude information belongs, select the corresponding preset growth mechanism model to calculate one or more theoretical growth index values ​​of the target wild tea tree, wherein the growth mechanism model is a curve regression model describing the relationship between tree height, crown width, diameter at breast height and ground diameter. S3. Construct a feature vector, which includes at least the local competition index, the theoretical growth index value, the altitude information, the canopy closure, and the growth environment type; use the feature vector as the input feature, and use one or more actual growth morphology indicators of wild tea trees as the prediction target to train a gradient boosting decision tree model to establish a multi-factor growth state prediction model. S4. For the wild tea tree individual to be evaluated, obtain its corresponding feature vector, input it into the trained multi-factor growth state prediction model, and output its predicted growth state value.

[0022] In one embodiment, for step S1, the core distribution area of ​​wild tea trees on Hainan Island is determined to be six cities and counties: Wuzhishan, Baoting, Qiongzhong, Baisha, Ledong, and Lingshui. The geographical coordinates and altitude of each wild tea tree with a diameter at breast height (DBH) ≥ 5cm are recorded using positioning equipment, and its growth environment type and canopy density are also recorded, thereby establishing a spatial distribution database of wild tea tree resources. The growth environment types include secondary forests, tea gardens, roadsides, and artificial forests.

[0023] Preferably, the local competition index is calculated as follows: a circular quadrat with radius R is drawn with the target wild tea tree as the center; the diameter at breast height (DBH) of all other wild tea trees in the quadrat is counted, and a competition weight value is assigned to each surrounding tea tree according to a predetermined diameter weight table. The principle for setting the competition weight value is: the diameter of young trees in the rapid growth and intense resource competition stage is given the highest weight, while the weight of seedlings and large-diameter trees is relatively low.

[0024] Specifically, a circular quadrat with a radius of R is drawn on the ground, centered on the root collar of the target wild tea tree (diameter at breast height ≥ 5 cm). The radius R can be adjusted between 3 meters and 10 meters according to the stand density. In a preferred embodiment, R = 5 meters is set to effectively capture the competitive effect of neighboring individuals. In one embodiment, based on the diameter at breast height (D) data, wild tea trees are divided into six diameter grades: Grade I seedlings (D<1.0cm), Grade II saplings (1cm≤D<5cm), Grade III medium trees (5cm≤D<10cm), Grade IV medium trees (10cm≤D<15cm), Grade V large trees (15cm≤D<20cm), and Grade VI old trees (D≥20cm); The diameter at breast height (DBH) of all other wild tea trees within the sample plot was statistically analyzed, and competitive weights were assigned based on their DBH class. In some embodiments of this invention, the weights are not equal but are determined based on in-depth ecological analysis. Preferably, the diameter weight table is as follows: the weight W1 of Grade I seedlings ranges from 0.5 to 0.8; the weight W2 of Grade II saplings ranges from 0.9 to 1.2; the weight W3 of Grade III medium trees ranges from 0.6 to 0.9; and the weight W4 of Grade IV and above large trees ranges from 0.3 to 0.6, and satisfies W2>W1, W2>W3, W2>W4; For example, Grade II saplings (1cm≤DBH<5cm) in the rapid growth stage compete most fiercely for water, light, and nutrients, so they should be given the highest weight (e.g., set to 1.0); Grade I seedlings (DBH<1.0cm) have weaker competitive ability and a lower weight (e.g., 0.7); Grade III and above medium and large trees have a more stable canopy structure and relatively mild competition, so the weight decreases step by step (e.g., Grade III 0.8, Grade IV and above 0.5).

[0025] Furthermore, based on the determined weights, the Local Competition Index (CI) is calculated: CI is calculated using the weighted competition intensity per unit area, and the formula is as follows: Where CI represents the local competition index of the target wild tea tree; R represents the radius of the circular quadrat centered on the target wild tea tree (unit: meters); n represents the total number of other wild tea trees in the quadrat besides the target tea tree; and i represents the i-th competing tea tree in the quadrat. This represents the competitive weight value assigned based on the diameter at breast height (DBH) of the i-th tea tree; A high CI value indicates that the target tea tree lives in a microenvironment with high competitive pressure. The unit area is filled with other tea trees (especially those in their rapid growth phase), which fiercely compete for light, water, and nutrients, potentially inhibiting the growth of the target tea tree. A low CI value means that the target tea tree is in a microenvironment with less competitive pressure and relatively abundant resources, making it easier to realize its growth potential.

[0026] In some embodiments, for step S2, firstly, based on the vertical zonation of vegetation on Hainan Island, the altitude is divided into three intervals: h<700m (tropical lowland rainforest), 700m≤h<1300m (tropical mountain rainforest), and h≥1300m (mountain mossy dwarf forest); each altitude zone corresponds to a set of optimal growth mechanism models obtained through curve regression analysis.

[0027] Preferably, the growth mechanism model is a cubic function curve model based on the division of different altitude zones, specifically including: In the lowland rainforest zone where h < 700 meters, the tree height-crown width model is H = a1 + b1C + c1C² + d1C³. In the mountain rainforest zone where 700m ≤ h < 1300m, the diameter at breast height (DBH) - diameter at ground level (D) model is D = a² + b²d + c²d² + d²d³. In the mossy dwarf forest zone at the summit with h ≥ 1300 meters, the diameter at breast height (DBH) - diameter at ground level (D) model is D = a³ + b³d + c³d² + d³d³. Where H represents tree height, C represents crown width, d represents ground diameter, D represents diameter at breast height, and a1, b1, c1, d1, a2, b2, c2, d2, a3, b3, c3, and d3 are obtained by fitting statistical data.

[0028] Specifically, the model was constructed based on: ① Altitude zoning: Hainan vegetation was divided into three regions according to vertical zonation (h < 700m lowland rainforest, 700m ≤ h < 1300m mountain rainforest, and h ≥ 1300m mountaintop mossy dwarf forest); ② Data basis: Field measurements of tree height (H), crown width (C), diameter at breast height (D), and ground diameter (d) of wild tea trees in 6 cities and counties; ③ Model selection: Through fitting 9 curve models, the cubic function with the highest R² (coefficient of determination) was selected as the optimal model. Ecological adaptation strategies at different altitudes: ①h < 700m: With sufficient water and heat, tea trees prioritize height growth (HC model R² = 0.488), lateral branch expansion is limited, and photosynthetic products are preferentially allocated to the stem; ②700m ≤ h < 1300m: With suitable temperature and cloud cover, tea trees adopt a balanced growth strategy, with high synergy between diameter at breast height (DBH) and ground diameter (DBC) (R² = 0.664), and enhance frost resistance through basal xylem deposition; ③h ≥ 1300m: In harsh high-altitude environments, tea trees prioritize basal thickening (Dd model R² = 0.887) to enhance mechanical stability, while also dwarfing and shaping (lower average tree height) to improve stress resistance.

[0029] For example, data on the height, crown width, diameter at breast height (DBH), and ground diameter of wild tea trees at different altitudes were collected. Scatter plots were then created, and nine different curve models were used to establish functional models: ① a functional model with tree height as the dependent variable and crown width, DBH, and ground diameter as independent variables; ② a functional model with DBH as the dependent variable and ground diameter as the independent variable. The correlations between the morphological characteristics of wild tea trees varied at different altitudes.

[0030] The results are shown in Table 1. h For heights <700m, the optimal fit between wild tea tree height and crown width, and between tree height and diameter at breast height, is a cubic curve model, with the following expressions: H =0.614+0.014 C -0.258 C 2 +2.13 C 3 R 2 =0.488; H =-3.202+0.002 D -0.101 D 2 +1.749 D 3 R 2 =0.251. The optimal fit between the diameter at breast height (DBH) and ground diameter of wild tea trees is the "S" function model, expressed as In... D =2.705-6.534 (1 / d) R 2 =0.412. When 700m≤ h For heights <1300m, the optimal fit between wild tea tree height and crown width, height and diameter at breast height (DBH), and DBH and ground diameter is a cubic function model, with the following expressions: H =2.824-0.004 C +0.086 C 2 +0.789 C 3 R 2 =0.434; H =1.949-0.001 D +0.020 D 2 +0.248 D 3 R 2 =0.340; D =1.001+0.003 d 2 +0.691 d 3 R 2 =0.664. When h For heights ≥1300m, the optimal fit between wild tea tree height and crown width, tree height and diameter at breast height (DBH), and DBH and ground diameter is a cubic function model, with the following expressions: H =6.574-0.023 C +0.316 C 2 -0.796 C3 R 2 =0.256; H =0.853+0.001 D -0.054 D 2 +1.100 D 3 R 2 =0.522; D =-0.726+0.001 d -0.024 d 2 +1.026 d 3 R 2 =0.887.

[0031] Table 1 In some embodiments, for step S3, a multidimensional feature vector is constructed for each sample tea tree. This vector is formed by organically combining features that reflect different levels of ecological mechanisms, specifically including the following features: Dynamic competition factor: Specifically, it is the local competition index (CI), which is the CI value calculated directly using the aforementioned embodiment. This is a continuous numerical feature. Theoretical growth index values: such as D_theoretical (theoretical diameter at breast height) and H_theoretical (theoretical tree height). The measured diameter at breast height d (or crown width C) of the tea tree and its altitude zone are input into the corresponding preset cubic function growth model in S2, and the calculated output value is the theoretical value. Altitude h: Obtained directly from GPS measurement equipment, as a continuous numerical feature; Canopy closure (CD): A value between 0.0 and 1.0 recorded visually, used as a continuous numerical feature; Growing environment type (GT): This is a categorical feature (e.g., [“secondary forest”, “within a tea garden”, “by a village road”, “under artificial forest”]). To make it processable by the model, it is digitized. For example, this invention employs a one-hot encoding method. Specifically, a new binary vector is created for this feature, with the dimension equal to the number of all possible values ​​for the environment type. For a specific tea tree, the dimension corresponding to its type is labeled as 1, and the other dimensions are labeled as 0. For example, assuming there are four environment types, the encoded values ​​would be: "Secondary Forest" is [1,0,0,0]; "Within Tea Garden" is [1,0,0,0]; "By Village Road" is [0,1,0,0]; and "Under Artificial Forest" is [0,0,0,1]. Ultimately, each tea plant is represented as a structure consisting of CI + D_theoretical + H_theoretical + h A fixed-dimensional feature vector composed of sub-features such as +CD+[GT_columns after one-hot encoding].

[0032] In one embodiment, the constructed feature vector is used as input features, and the actual tree height (H), diameter at breast height (D), or crown width (C) measured precisely in the field is used as the prediction target (i.e., label). Supervised learning training is performed using a gradient boosting decision tree (GBDT model, such as LightGBM). The model iteratively learns the complex, non-linear mapping relationship between these features and the target value.

[0033] Specifically, the data from all surveyed wild tea tree samples (e.g., 6915 trees) were compiled. Each sample corresponds to a feature vector (X) constructed in the previous steps and its corresponding actual measurement value (y), i.e., a label (e.g., measured tree height H, diameter at breast height D, or crown width C). This forms a complete and standardized dataset.

[0034] To objectively evaluate the model's generalization ability and prevent overfitting, the dataset needs to be randomly divided into two parts: Training set: usually accounts for 70%-80%, used for parameter learning and rule building within the model.

[0035] Test set: Typically accounts for 20%-30%, used to simulate the model's performance on unseen data after training, in order to fairly evaluate its prediction accuracy.

[0036] like Figure 2 The diagram shown illustrates the GBDT model training process, which includes the following steps: S201, GBDT model initialization. The GBDT model starts with a simple initial prediction value, which is usually the average of the labels of all samples in the training set (e.g., the average tree height of all tea trees). S202, predict the prediction error of the current model. That is, the goal of the model to calculate the first decision tree is not to predict the label itself, but to predict the prediction error (i.e., residual) of the current model. For each sample, residual = true value - current predicted value (initially the mean). S203, The model constructs the first tree based on a greedy algorithm, including using the feature data (X) of the entire training set, but using the calculated residuals as the new learning target to train the first decision tree; Specifically, for example, the construction of this tree is carried out by a greedy algorithm, which selects the best split point among all features (e.g., "if the local competition index CI < 0.05, then split into one group"), so that the residual variance of the samples in each group is reduced to the greatest extent after splitting. This tree is usually restricted to a shallow depth (e.g., 4-6 layers) and is called a "weak learner". It learns a portion of the most significant error patterns in the data. S204, Update Predictions and Residuals, including multiplying the predictions from the first tree (the predicted residuals) by a small learning rate (e.g., 0.1) and adding it to the initial predictions to form a new, slightly improved overall prediction. Subsequently, based on this new prediction, recalculate the residuals for all samples. S205, using the newly calculated residuals as the target, train a second decision tree. This second tree learns the residual error patterns that the first tree failed to capture; S206, repeat the above process S201-S205, fit the current residual with the new tree -> update the overall prediction -> calculate the new residual - hundreds or thousands of times (for example, build 1000 trees), where each new tree is dedicated to correcting the errors accumulated by all the previous trees; This process enables the model to learn, in an incremental manner, extremely complex nonlinear relationships and interactions between features and the target. For example, it can automatically learn complex feature interaction rules such as "in a high canopy closure environment, the local competition index significantly enhances its inhibitory effect on tree height" without manual specification.

[0037] S207, the training process stops after reaching a preset number of iterations (number of trees), or terminates when the model no longer improves on an independent validation set (early stopping), to prevent overfitting.

[0038] After training, the final model is an ensemble model composed of all the trained trees. This model is saved and used for subsequent predictions on new samples. For a new input feature vector, the final prediction is the sum of the initial prediction and the predictions from all the trees.

[0039] In one embodiment, preferably, based on the feature importance ranking obtained from model training, the diagnostic results of the dominant factors affecting its growth status are output. Specifically, the feature importance ranking is performed by extracting and outputting the contribution ranking of all input features to the prediction target after the multi-factor growth status prediction model is trained, so as to quantitatively identify the dominant environmental factors and competing factors affecting the growth of wild tea trees.

[0040] Specifically, feature importance is not an independent statistic calculated separately after model training, but rather metadata that is naturally generated and recorded during model training. It is based on the working principle of the Gradient Boosting Decision Tree (GBDT) model and, for example, is primarily measured through the following two core metrics: 1. Importance based on splitting frequency Specifically, when constructing each decision tree, the model needs to select the best feature and split point at each node to partition the data. The more times a feature is selected as the splitting feature across all trees, the more important its role in making the correct decision is. After training, the entire model (all nodes of all trees) is traversed, and the total number of times each feature was used for splits is counted. This count is then normalized (e.g., converting the counts of all features to a percentage of the total counts) to obtain an importance score for each feature. A high importance score for a feature directly indicates that the model repeatedly relies on it to distinguish different growth conditions.

[0041] 2. The Importance of Information Gain Specifically, it measures not "whether it is used", but "how effective it is". Each time a feature is used for splitting, it will bring an increase in "purity", that is, how much the residual (or loss function) of the samples in the two child nodes after splitting is reduced compared to the parent node. This amount of reduction is "information gain". The model accumulates the total information gain of each feature across all splits. If a feature brings a large information gain each time it is used (i.e., it can very effectively reduce prediction error), then even if it is not used the most times, its cumulative gain may be very large, resulting in a high importance score.

[0042] Specifically, during model training, whenever the model creates a new decision tree, it traverses every split node in the tree. For each node: it records the feature selected for splitting; it also records the information gain brought by this split. This value quantifies how much the model's prediction error is reduced because of using this feature for splitting. The larger the information gain, the more effective the split is.

[0043] Once all trees have been built, perform a global aggregation: For each feature (e.g., "Local Competition Index CI"), find all the split nodes that use this feature, then sum the information gain values ​​corresponding to these nodes to obtain the total information gain of the feature. Finally, in order to obtain the importance score, normalize the total information gain of all features, that is, the importance score of each feature is expressed as the percentage of its total information gain to the sum of the total information gains of all features.

[0044] Preferably, the importance score of feature F can be simplified to the formula: Importance score of feature F = (sum of information gain of all nodes using feature F) / (sum of information gain of all nodes using all features) 100% As an example, suppose we train a model to predict the height of wild tea trees. The feature vector contains the following three key features: Feature A: Local Competition Index (CI) (Dynamic Competition Factor) Feature B: Theoretical tree height (H_theoretical) (mechanistic model output) Feature C: Canopy Closure (CD) (Traditional Habitat Indicator) During the model training process, the three features are used as follows: Feature A (CI): Used for 20 splits. These 20 splits are of high quality, each bringing a significant information gain, for example, because it clearly distinguishes between highly competitive and low-competition tea plants. Assume its total information gain is 150 units.

[0045] Feature B (H_theoretical): Used for 50 splits, it is the most frequently used feature. However, because it is a theoretical baseline, the model often uses it for minor, small-amplitude error corrections, so the information gain per split is not high. Assume its total information gain is 100 units.

[0046] Feature C (CD): Used for 10 splits. Although used infrequently, it resulted in very high information gain on several key splits (e.g., in distinguishing between "full illumination" and "strong shading"). Assume its total information gain is 50 units.

[0047] Calculate the overall importance: Total information gain of all features = 150(A) + 100(B) + 50(C) = 300 units Importance score of feature A (CI) = (150 / 300) 100% = 50.0% Importance score of feature B (H_theoretical) = (100 / 300) 100% = 33.3% Importance score of feature C (CD) = (50 / 300) 100% = 16.7% Therefore, if we only look at the split frequency, we might mistakenly assume that feature B (theoretical tree height) is the most important (50 times). However, through comprehensive scoring, feature A (local competition index) is the true primary dominant factor (50%), because it contributes greater value every time it is used and plays the most crucial role in reducing model error.

[0048] Preferably, when the diagnostic results of the dominant factor show that the local competition index is the key limiting factor, it is recommended to implement artificial thinning; when the canopy closure is the key limiting factor, it is recommended to implement forest gap thinning or light management.

[0049] Specifically, when the local competition index (CI) ranks highly in the characteristic importance ranking, it indicates that, under the current habitat, competition for light, water, and nutrients within the wild tea tree population is the primary cause of inhibited individual growth. Population density may be too high, especially in the sapling and mid-sized tree stages, where intense intraspecific competition leads to insufficient resources to meet the normal growth needs of individual trees. Artificial thinning refers to the selective felling of some competing individuals within the quadrat surrounding the target wild tea tree, especially those Class II saplings that are assigned a high competitive weight, in order to directly reduce local competitive pressure (CI value).

[0050] When "canopy closure (CD)" ranks highly in importance, it indicates that insufficient light due to canopy shading is the dominant environmental stress limiting the growth of wild tea trees in the understory. Excessive canopy closure restricts photosynthesis in tea trees, thereby directly inhibiting their growth and biomass accumulation.

[0051] It is recommended to reduce the canopy cover and increase the light intensity within the forest through artificial intervention. For example, in areas with extremely high canopy closure, a small number of non-target tree species or diseased or weak trees in the upper shade can be selectively removed to create small forest gaps; or, in cases where canopy closure is mainly caused by a few tall trees, these specific trees can be pruned appropriately, removing some of their lower branches, like "opening a skylight," allowing more diffused light to penetrate to the tea trees under the forest canopy.

[0052] Preferably, the method also includes tree age estimation. Specifically, for shrub-type wild tea trees whose diameter at breast height (DBH) cannot be directly measured, their ground diameter (DBH) is measured, and their theoretical DBH is calculated using the DBH-DBH growth mechanism model in high-altitude zones. Then, their tree age is estimated based on the DBH-age relationship model.

[0053] Specifically, in surveys of wild tea tree resources, many individuals have short trunks, low branching points, or grow in shrub-like formations, making it difficult to measure their diameter at breast height (DBH) using standard methods (at 1.3 meters above the ground). DBH is the most commonly used key independent variable in establishing tree age estimation models.

[0054] This embodiment utilizes a diameter at breast height (DBH)-diameter at ground level growth mechanism model established at high altitudes (h≥1300m) with an extremely high goodness of fit (R²=0.887). This model reveals a stable and predictable mathematical relationship between the base diameter at breast height (DBH) and the diameter at breast height (BHP) of tea trees in this habitat.

[0055] Specifically, the process includes the following steps: For wild tea shrubs whose age needs to be estimated, investigators use a diameter-at-breast-width (DBH) ruler or tape measure to measure their ground diameter (d). The standard for measuring ground diameter is the diameter of the root collar, 20 cm above the ground, including the bark.

[0056] Obtain the precise altitude (h) information of the tea tree's location. If it is confirmed to be located in a high-altitude zone (e.g., h ≥ 1300 m), then invoke a pre-set cubic diameter-diameter-ground function model applicable to that altitude zone, as described in the aforementioned embodiment. D=a3+b3d+c3d²+d3d³; among them, a3=-0.726, b3=0.001, c3=-0.024, d3=1.026; By substituting the measured diameter at breast height (d) into the above model, the theoretical diameter at breast height (D_theoretical) of the tea tree can be calculated. By collecting wild tea tree samples of different diameter-at-breast height (DBH) levels (usually requiring methods such as drilling wood cores with a growth cone to obtain true age data), a regression model of the DBH (D)-age (A) relationship of wild tea trees on Hainan Island was established. This model can be linear, nonlinear, or an empirical equation, such as A=f(D), for example, A = e0 + e1D_theoretical + e2D_theoretical². Based on statistical indicators such as the coefficient of determination (R²) and residual distribution of different statistical data, the optimal model was selected as the final DBH-age relationship model A=f(D), which will not be elaborated further here.

[0057] Substitute the calculated theoretical diameter at breast height (DBH) value D_theoretical into the established DBH-age relationship model A=f(D); then the estimated age A of the tea tree can be calculated and output.

[0058] Figure 3 A system 300 for constructing a growth model of wild tea trees on Hainan Island is shown. Embodiments of this system are... Figure 1 Corresponding to the illustrated method embodiments, the specific methods include: The local competition index acquisition module 301 is used to acquire and quantify the microenvironmental competition factors for a target wild tea tree individual. The microenvironmental competition factors include at least the local competition index calculated based on the diameter distribution of all other wild tea trees within a surrounding fixed radius quadrat. The curve regression model generation module 302 is used to obtain the altitude information of the target wild tea tree, and based on the preset vegetation vertical zone to which the altitude information belongs, select the corresponding preset growth mechanism model to calculate one or more theoretical growth index values ​​of the target wild tea tree. The growth mechanism model is a curve regression model that describes the relationship between tree height, crown width, diameter at breast height and ground diameter. The prediction model building module 303 is used to construct a feature vector, which includes at least the local competition index, the theoretical growth index value, the altitude information, the canopy closure, and the growth environment type; using the feature vector as input features and one or more actual growth morphology indicators of wild tea trees as prediction targets, a gradient boosting decision tree model is trained to establish a multi-factor growth state prediction model. The prediction output module 304 is used to obtain the corresponding feature vector of the wild tea tree individual to be evaluated, input it into the trained multi-factor growth state prediction model, and output the predicted value of its growth state.

[0059] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing a growth model of wild tea trees on Hainan Island, characterized in that, Includes the following steps: For a target wild tea tree individual, its microenvironmental competition factors are obtained and quantified. The microenvironmental competition factors include at least the local competition index calculated based on the diameter distribution of all other wild tea trees within a surrounding fixed radius quadrat. Obtain the altitude information of the target wild tea tree, and based on the preset vegetation vertical zone to which the altitude information belongs, select the corresponding preset growth mechanism model to calculate one or more theoretical growth index values ​​of the target wild tea tree. The growth mechanism model is a curve regression model that describes the relationship between tree height, crown width, diameter at breast height and ground diameter. Construct a feature vector, which includes at least the local competition index, the theoretical growth index value, the altitude information, the canopy closure, and the growth environment type; use the feature vector as input features, and use one or more actual growth morphology indicators of wild tea trees as prediction targets, train a gradient boosting decision tree model to establish a multi-factor growth state prediction model. For each wild tea tree individual to be evaluated, its corresponding feature vector is obtained and input into the trained multi-factor growth state prediction model, which then outputs the predicted value of its growth state.

2. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 1, characterized in that, The local competition index is calculated as follows: A circular quadrat with radius R is drawn with the target wild tea tree as the center; The diameter at breast height (DBH) of all other wild tea trees within the sample plot was counted, and each surrounding tea tree was assigned a competitive weight value according to a predetermined diameter weight table. The principle for setting the competitive weight value was that the diameter of young trees in the rapid growth and resource competition stage was given the highest weight, while the weight of seedlings and large-diameter trees was relatively low.

3. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 1, characterized in that, The specific weighting table for diameter class is as follows: the weight W1 for Class I seedlings ranges from 0.5 to 0.8; the weight W2 for Class II saplings ranges from 0.9 to 1.2; the weight W3 for Class III medium-sized trees ranges from 0.6 to 0.9; and the weight W4 for Class IV and above large trees ranges from 0.3 to 0.6, and satisfies W2>W1, W2>W3, and W2>W4.

4. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 1, characterized in that, include: The growth mechanism model is a cubic function curve model based on different altitude zones h, specifically including: In the lowland rainforest zone where h < 700 meters, the tree height-crown width model is H = a1 + b1C + c1C² + d1C³. In the mountain rainforest zone where 700m ≤ h < 1300m, the diameter at breast height (DBH) - diameter at ground level (D) model is D = a² + b²d + c²d² + d²d³. In the mossy dwarf forest zone at the summit with h ≥ 1300 meters, the diameter at breast height (DBH) - diameter at ground level (D) model is D = a³ + b³d + c³d² + d³d³. Where H represents tree height, C represents crown width, d represents ground diameter, D represents diameter at breast height, and a1, b1, c1, d1, a2, b2, c2, d2, a3, b3, c3, and d3 are obtained by fitting statistical data.

5. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 4, characterized in that, The multi-factor growth state prediction model is established by combining the theoretical growth index values ​​with the corresponding actual measured values ​​as input features. The theoretical growth index values ​​include at least the theoretical diameter at breast height (DBH) value calculated from the ground diameter using the DBH-ground diameter model, and / or the theoretical tree height value calculated from the crown width using the tree height-crown width model.

6. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 1, characterized in that, The growth environment type is a categorical variable, which is digitized through unique thermal coding. Its types include at least: secondary forest, tea garden, village roadside, and artificial forest.

7. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 1, characterized in that, It also includes tree age estimation. Specifically, for shrub-type wild tea trees whose diameter at breast height (DBH) cannot be directly measured, their ground diameter (DDB) is measured, and their theoretical DBH is calculated using the DBH-DDB growth mechanism model in high-altitude zones. Then, their tree age is estimated based on the DBH-age relationship model.

8. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 1, characterized in that, It also includes outputting diagnostic results of dominant factors affecting the growth status based on the feature importance ranking obtained from model training. Specifically, the feature importance ranking is to extract and output the contribution ranking of all input features to the prediction target after the multi-factor growth status prediction model is trained, so as to quantitatively identify the dominant environmental factors and competing factors affecting the growth of wild tea trees.

9. The method for constructing a growth model of wild tea trees on Hainan Island according to claim 8, characterized in that, When the diagnostic results of the dominant factor show that the local competition index is the key limiting factor, artificial thinning is recommended; when the canopy closure is the key limiting factor, forest gap thinning or light management is recommended.

10. A system for constructing a growth model of wild tea trees on Hainan Island, characterized in that, include: The local competition index acquisition module is used to acquire and quantify the microenvironmental competition factors for a target wild tea tree individual. The microenvironmental competition factors include at least the local competition index calculated based on the diameter distribution of all other wild tea trees within a surrounding fixed radius quadrat. The curve regression model generation module is used to obtain the altitude information of the target wild tea tree, and based on the preset vegetation vertical zone to which the altitude information belongs, select the corresponding preset growth mechanism model to calculate one or more theoretical growth index values ​​of the target wild tea tree. The growth mechanism model is a curve regression model that describes the relationship between tree height, crown width, diameter at breast height and ground diameter. The prediction model building module is used to construct a feature vector, which includes at least the local competition index, the theoretical growth index value, the altitude information, the canopy closure, and the growth environment type. Using the feature vector as input features and one or more actual growth morphology indicators of wild tea trees as prediction targets, a gradient boosting decision tree model is trained to establish a multi-factor growth state prediction model. The prediction output module is used to obtain the corresponding feature vector of the wild tea tree individual to be evaluated, input it into the trained multi-factor growth state prediction model, and output the predicted value of its growth state.

Images