Litchi flowering induction estimation method based on short, medium and long term forecasting data

Abstract

The invention relates to the technical field of litchi flower formation prediction, and discloses a litchi flower formation induction estimation method based on short, medium and long term prediction data, and the method comprises the following steps: constructing a static multi-source heterogeneous data set; constructing a climate data set; and acquiring satellite remote sensing data of the target area as a remote sensing data set. Inputting the static multi-source heterogeneous data set and the climate data set into a pre-trained XGBoost model to obtain a litchi flowering induction result of the target area; inputting the climate data set into a pre-trained LSTM model to obtain a florescence climate index of the target area; and obtaining a vegetation index from the remote sensing data set, fusing the vegetation index to obtain a first flowering index, fusing the first flowering index with the terrain index to obtain a second flowering index, and evaluating the flowering stability condition of the target area based on the second flowering index.

Description

Technical Field

[0001] The present invention relates to the technical field of litchi flowering prediction, and particularly to a method for predicting litchi flowering induction based on short, medium and long-term forecast data. Background Art

[0002] As an important economic fruit tree in tropical and subtropical regions, the flowering induction mechanism of litchi has typical biphasic regulation characteristics. It requires sufficient low-temperature accumulation and is easily inhibited by high temperatures. Traditional research mostly analyzes the flowering mechanism based on short-time series meteorological observations and artificial temperature control experiments, lacking systematic prediction of the flowering stability based on short, medium and long-term forecast data. Under the background of global warming, the frequent occurrence of insufficient winter low temperatures and extreme high temperatures has led to a continuous decline in the flowering rate of mid-late maturing litchi in the coastal areas of South China. The existing technology fails to effectively integrate the coupling analysis framework of multi-source data such as phenological periods, climate data and machine learning models, and it is difficult to quantitatively evaluate the impact of short-term meteorology and medium and long-term climate on flowering induction. Therefore, a systematic quantitative evaluation of litchi flowering based on short, medium and long-term forecasts not only has important academic value but also has extremely high practical significance. By deeply analyzing the influence of different meteorological factors and combining machine learning models for accurate prediction of flowering, producers can grasp the future flowering stability in advance and take precise and effective management measures according to the prediction results to ensure a high flowering rate. This can not only effectively improve the stability of litchi production but also effectively avoid the risks brought by adverse meteorological and climate conditions, and promote the sustainable development of the litchi industry. Summary of the Invention

[0003] Aiming at the above-mentioned prior art, the present invention provides a method for predicting litchi flowering induction based on short, medium and long-term forecast data, mainly solving the technical problems existing in the above background art.

[0004] To achieve the above object, the technical solution of the embodiment of the present invention is realized as follows:

[0005] A method for predicting litchi flowering induction based on short, medium and long-term forecast data, the method comprising the following steps:

[0006] Collect litchi phenological data, soil humidity data, light intensity data and litchi physiological parameter data of the target area, and construct a static multi-source heterogeneous data set;

[0007] Obtain the daily historical climate data and short, medium and long-term forecast data of the target area, construct a climate data set, and obtain the satellite remote sensing data of the target area as a remote sensing data set;

[0008] Input the static multi-source heterogeneous data set and the climate data set into a pre-trained XGBoost model to obtain the litchi flowering induction result E of the target areaXGBoost ;

[0009] Input the climate dataset into the pre-trained LSTM model to obtain the flowering climate index E of the target area LSTM ;

[0010] Obtain the vegetation index from the remote sensing dataset and fuse it with E XGBoost 、E LSTM and the vegetation index to obtain the first flowering index;

[0011] The first flowering index is fused with the terrain index to obtain the second flowering index, and the flowering stability of the target area is evaluated based on the second flowering index.

[0012] Optionally, input the static multi-source heterogeneous dataset and the climate dataset into the pre-trained XGBoost model, specifically including:

[0013] Encode and convert the non-numerical features in the static multi-source heterogeneous dataset and the climate dataset into numerical features, and normalize the numerical features to obtain the preprocessed static multi-source heterogeneous dataset and climate dataset;

[0014] Use the preprocessed static multi-source heterogeneous dataset and climate dataset as input samples and input them into the pre-trained XGBoost model to obtain the litchi flowering induction result E of the target area XGBoost , and its output result is expressed as: where K represents the number of decision trees in the XGBoost model, k represents the kth decision tree, ω k represents the weight of the kth decision tree, and T k (x) represents the prediction output of the kth decision tree for the input sample x.

[0015] Optionally, to obtain the flowering climate index E of the target area LSTM , specifically including:

[0016] Set a sliding window with a window size of W and a window step of S, and slice the climate dataset into time series according to the sliding window;

[0017] Input the time series into the LSTM model. The LSTM model updates the hidden state and the cell state through the calculations of the forget gate, input gate, cell state, and output gate, and performs weighted summation on the hidden states of all time steps through the attention mechanism to obtain the predicted values of the low-temperature effective accumulated cold amount and the high-temperature effective accumulated heat amount respectively;

[0018] Based on the predicted values of the low-temperature effective accumulated cold amount and the high-temperature effective accumulated heat amount, calculate the flowering climate index E through the following formula LSTM :

[0019]

[0020] Among them, LC is the predicted value of the low-temperature effective accumulated chilling degree with the daily minimum temperature ranging from 5 to 15 °C, and HC is the predicted value of the high-temperature effective accumulated heat quantity with the daily maximum temperature ≥ 28 °C.

[0021] Optionally, obtain the vegetation index from the remote sensing dataset, specifically including: extracting the red band and near-infrared band data of the satellite remote sensing data from the remote sensing dataset, and calculating the vegetation index based on the following formula:

[0022]

[0023] Among them, NDVI represents the vegetation index, NIR represents the reflection characteristics of the near-red band in the target area, and Red represents the reflection characteristics of the red band in the target area.

[0024] Optionally, obtain the terrain index from the remote sensing dataset, specifically including: resampling the DEM data in the remote sensing dataset to the same spatial resolution as NDVI, and using GIS software to calculate the elevation, aspect, and slope in the DEM data, and normalizing the calculation results as terrain features to obtain the terrain index.

[0025] Optionally, calculate the fused first flowering index E XGBoost 、E LSTM and the vegetation index,

[0026] FCI * =α*E XGBoost +β*E LSTM +γ*NDVI

[0027] Among them, α, β, and γ are all weight coefficients.

[0028] Optionally, obtain the second flowering index by fusing the first flowering index and the terrain index through the following formula:

[0029]

[0030] Among them, k is an empirical coefficient, and E 地形 represents the terrain index.

[0031] Optionally, evaluate the flowering stability of the target area based on the second flowering index, specifically including: when , the target area is an unstable flowering area, when , the target area is a relatively stable flowering area, and when , the target area is a stable flowering area.

[0032] The beneficial effects of the present invention are as follows: By integrating short-, medium-, and long-term forecast and prediction data, phenological observation data, and satellite remote sensing data, and combining the XGBoost and LSTM models to capture static features and time series patterns respectively, the spatio-temporal resolution of the flowering induction prediction is significantly improved. By constructing a dynamic fusion mechanism of vegetation indices and topographic indices, the comprehensive impact of environmental factors on flowering stability is quantified, solving the limitations of a single model in response to complex climates. The flowering index grading and evaluation system based on short-, medium-, and long-term forecast and prediction data can not only effectively improve the prediction accuracy of litchi flowering rate, but also provide scientific decision-making support for precision agriculture management, promoting the development of the litchi industry towards a more efficient, stable, and sustainable direction. Brief Description of the Drawings

[0033] Figure 1 It is a schematic flowchart of the litchi flowering induction estimation method based on short-, medium-, and long-term forecast and prediction data in the embodiment of the present application. Detailed Embodiments

[0034] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings of the specification and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. In the following description, the expression "some embodiments" is used, which describes a subset of all possible embodiments. However, it should be understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict.

[0035] In the following description, a large number of specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, some well-known technical features are not described to avoid confusion with the present invention.

[0036] It should be understood that the present invention can be implemented in different forms and should not be construed as limited to the embodiments presented herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the present invention to those skilled in the art. And the purpose of the terms used herein is only to describe specific embodiments and is not a limitation of the present invention. As used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the terms "comprising" and / or "including", when used in this specification, determine the presence of the described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups. As used herein, the term "and / or" includes any and all combinations of the related listed items.

[0037] It should be further noted that when an element is referred to as being "fixed to" another element, it can be directly on the other element or there can also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "inner", "outer", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation.

[0038] To thoroughly understand the present invention, detailed structures will be presented in the following description to illustrate the technical solutions proposed by the present invention. The optional embodiments of the present invention are described in detail below. However, in addition to these detailed descriptions, the present invention can also have other implementations.

[0039] Please refer to the attached Figure 1 , this application provides a method for predicting litchi flower bud induction based on short, medium and long-term forecast data, and the method includes the following steps:

[0040] S1. Collect litchi phenological data, soil humidity data, light intensity data and litchi physiological parameter data in the target area, and construct a static multi-source heterogeneous data set;

[0041] S2. Obtain the daily historical climate data and short, medium and long-term forecast data in the target area, construct a climate data set, and obtain the satellite remote sensing data in the target area as a remote sensing data set;

[0042] S3. Input the static multi-source heterogeneous data set and the climate data set into a pre-trained XGBoost model to obtain the litchi flower bud induction result E XGBoost ;

[0043] S4. Input the climate data set into a pre-trained LSTM model to obtain the flower-forming climate index E of the target areaLSTM ;

[0044] S5. Obtain the vegetation index from the remote sensing dataset, and fuse E XGBoost and E LSTM with the vegetation index to obtain the first flowering index;

[0045] S6. Fuse the first flowering index with the terrain index to obtain the second flowering index, and evaluate the flowering stability of the target area based on the second flowering index.

[0046] Specifically, in step S1, the litchi phenological data refers to the key time nodes of litchi at different growth stages, including but not limited to the last autumn shoot maturity period, the physiological differentiation period of flower buds, the morphological differentiation period of flower buds, the flowering period, etc. These data are crucial for understanding the litchi flowering induction process and are usually collected by means of field observations. That is, representative litchi planting plots are selected in the target area, and field observations are carried out regularly, for example, once a week, to record the growth and development stages of litchi. The observation content includes but not limited to the last autumn shoot maturity time, the physiological differentiation time of flower buds, the morphological differentiation time of flower buds, the flowering time, etc. The collected phenological data are recorded in a special table or database.

[0047] The soil moisture data refers to the water content of the soil in the target area, usually expressed as volumetric water content or gravimetric water content. Soil moisture has an important impact on the growth and flowering induction of litchi and is usually collected by installing soil moisture sensors in the target area. These sensors can monitor the soil moisture changes in real time. The sensors should be arranged at different depths, such as 0 - 20 cm, 20 - 40 cm, etc., to obtain soil moisture data at different levels.

[0048] The light intensity data refers to the intensity of solar radiation in the target area, usually expressed as sunshine hours and light intensity. Light intensity has an important impact on the photosynthesis and flowering induction of litchi. Usually, light sensors are installed in the target area. These sensors can monitor the light intensity changes in real time. The sensors should be arranged at different heights, such as the canopy layer, the ground layer, etc., to obtain light data at different levels. At the same time, collect the daily sunshine hours and light intensity data provided by the meteorological station to supplement the sensor data. These data will help to evaluate the impact of light conditions on litchi flowering induction.

[0049] The physiological parameter data of litchi refers to the physiological indicators related to the growth and development of litchi, including but not limited to leaf water content, leaf temperature, leaf chlorophyll content, etc. These parameters can reflect the physiological state of litchi and the process of flower bud induction. For the leaf water content data, litchi leaf samples are regularly collected from the target area, and the leaf water content is measured in the laboratory; for the leaf temperature data, an infrared temperature sensor is installed in the target area to monitor the temperature change of litchi leaves in real time; for the leaf chlorophyll content: a portable chlorophyll meter is used to regularly measure the chlorophyll content of litchi leaves.

[0050] In an optional implementation manner, in step S2, it is first necessary to obtain the daily historical climate data and short, medium, and long-term forecast data of the target area. The daily historical climate data usually includes meteorological elements such as air temperature, precipitation, humidity, wind speed, etc., which can be obtained through meteorological station observations or meteorological data service platforms.

[0051] The short, medium, and long-term forecast data usually includes 1-3 days (short-term), 4-10 days (medium-term), dekad and month (long-term) updated daily, and it also comes from third-party platforms such as the meteorological bureau. These climate data can provide meteorological and climate information for different future periods, which helps to analyze the potential impact of future meteorological and climate changes on litchi flower bud induction. At the same time, satellite remote sensing data of the target area is collected. These data include information such as vegetation index, soil humidity, surface temperature, etc., which can reflect the vegetation status and environmental conditions of the litchi planting area. By integrating these data, a comprehensive climate dataset and remote sensing dataset are constructed to provide basic data support for the subsequent estimation of litchi flower bud induction. These datasets will be used to analyze the impact of future meteorological and climate changes on litchi flower bud induction to evaluate the stability and risk of litchi flower bud formation at different times.

[0052] In an optional implementation manner, in step S3, the static multi-source heterogeneous dataset and the climate dataset are input into the pre-trained XGBoost model, specifically including:

[0053] Encode and convert the non-numerical features in the static multi-source heterogeneous dataset and the climate dataset into numerical features, and normalize the numerical features to obtain the preprocessed static multi-source heterogeneous dataset and climate dataset;

[0054] Take the preprocessed static multi-source heterogeneous dataset and climate dataset as input samples and input them into the pre-trained XGBoost model to obtain the litchi flower bud induction result E of the target area XGBoost , and its output result is expressed as: where K represents the number of decision trees in the XGBoost model, k represents the kth decision tree, ω k represents the weight of the kth decision tree, and T k(x) represents the predicted output of the k-th decision tree for the input sample x.

[0055] Specifically, the XGBoost model is an ensemble learning algorithm based on the gradient boosting framework, which improves the prediction accuracy by constructing multiple decision trees. During the model training process, the static multi-source heterogeneous dataset and the climate dataset are first divided into a training set and a prediction set. The XGBoost model first initializes a basic model and inputs the training set data, and then gradually adds new decision trees in an iterative manner. Each tree is optimized based on the previous tree to minimize the prediction error. Specifically, the XGBoost model determines the splitting direction and splitting point of each tree by calculating the gradient of the loss function, so as to achieve efficient fitting of the data. For those skilled in the art, the training process of the XGBoost model and the mathematical formulas used are common knowledge, and will not be elaborated in detail in this embodiment.

[0056] In the prediction stage, the input sample will be predicted by each decision tree in turn, and each tree will give a prediction value according to its own structure and weight. The final prediction result is the weighted sum of the prediction values of all decision trees, where the weight of each tree reflects its importance in the model. In this way, the XGBoost model can comprehensively consider the prediction results of multiple decision trees, thereby improving the prediction accuracy and stability. In this embodiment, the output result of the XGBoost model is represented as a numerical value, which represents the litchi flower induction result of the target area and can be used to evaluate the possibility and stability of litchi flowering.

[0057] When training with short-term data, assuming it is December 10th now, the actual flowering results from December 8th to 10th are used as labels, and the historical meteorological data from December 1st to December 7th, the data predicted by the meteorological bureau for December 8th to 10th on December 7th, and the static dataset with optional time are input into the XGBoost model for training, and the flowering result prediction for December 8th to 10th is obtained during the training process.

[0058] When using medium- and long-term prediction data, assuming it is December now, the data from September to November belong to historical meteorological data, and the data after December belong to prediction data. The training process is to input the historical climate data from September to October, the prediction data made by the meteorological bureau for November in October, and the static dataset with optional time into the XGBoost model, and use the flowering result in December as the label. In the model, the non-linear relationship between the above static factors and the static characteristics of climate data is transformed into a linear result that can be understood by humans. Then in actual prediction, the input data should be historical data + prediction data + static characteristics of climate data to obtain a preliminary prediction result.

[0059] In the actual prediction application scenario, when predicting the flower bud induction results in January and subsequent months, the data input into the model should be historical meteorological data, such as data from September to November, prediction data, such as data after December, and static datasets. Based on these data, the model can obtain preliminary prediction results, thus providing a scientific basis for the estimation of litchi flower bud induction.

[0060] In an optional implementation, obtaining the flower formation climate index E of the target area LSTM , specifically includes:

[0061] Set a sliding window with a window size W and a window step size S, and slice the climate dataset into time series according to the sliding window;

[0062] Input the time series into the LSTM model. The LSTM model updates the hidden state and cell state through the calculations of the forget gate, input gate, cell state, and output gate, and performs weighted summation on the hidden states of all time steps through the attention mechanism to obtain the predicted values of the effective accumulated low temperature and the effective accumulated high temperature respectively;

[0063] Based on the predicted values of the effective accumulated low temperature and the effective accumulated high temperature, calculate the flower formation climate index E through the following formula LSTM :

[0064]

[0065] where LC is the predicted value of the effective accumulated low temperature with the daily minimum temperature between 5 - 15°C, and HC is the predicted value of the effective accumulated high temperature with the daily maximum temperature ≥ 28°C

[0066] It can be seen from the logarithmic model that the more the effective accumulated low temperature and the less the effective accumulated high temperature, the larger the FCI value, and the easier it is for litchi to form flower buds.

[0067] Specifically, in step S4, in order to obtain the flower formation climate index of the target area, first, a sliding window with a window size W and a window step size S needs to be set. The function of this sliding window is to slice the climate dataset into multiple time series segments. For example, if the climate data is daily data, the window size W can be set to 30 days, and the window step size S can be set to 1 day or 5 days, etc. The specific values need to be determined according to the actual situation and experience. Through the sliding window technology, the original climate data can be divided into multiple continuous time series, and these time series will be used as the input of the LSTM model.

[0068] The LSTM model is a variant of the Recurrent Neural Network (RNN) and is particularly good at processing time series data. The LSTM model updates the hidden state and cell state through the calculations of the forget gate, input gate, cell state, and output gate. The forget gate determines which information needs to be discarded, the input gate controls which new information needs to be stored, and the output gate determines which information needs to be output. In this process, the LSTM model can capture the dynamic changes and long-term dependencies in the time series.

[0069] In the LSTM model, the attention mechanism is used to perform a weighted sum of the hidden states at all time steps. The attention mechanism can automatically learn which hidden states at which time steps are more important for the prediction result. By performing a weighted sum of the hidden states, the LSTM model can obtain a more meaningful feature representation, thereby improving the prediction accuracy.

[0070] Finally, the LSTM model will respectively predict the low-temperature effective accumulated temperature and high-temperature effective accumulated heat based on the input time series. These predicted values will be used to calculate the flowering climate index, which can reflect the stability of the flowering situation in the target area. In this way, the impact of climate change on litchi flowering induction can be evaluated, providing a scientific basis for litchi planting and management.

[0071] During the training process, optimization algorithms such as Stochastic Gradient Descent or Adam are used to update the weights and biases of the model. The loss function usually adopts the mean squared error or root mean squared error to measure the difference between the predicted value and the true value. Through the backpropagation algorithm, the model can automatically adjust the parameters to minimize the loss function. For those skilled in the art, the training process of the LSTM model and the mathematical formulas used are common knowledge, and thus will not be elaborated in detail in this embodiment.

[0072] In an optional implementation manner, obtaining the vegetation index from the remote sensing dataset specifically includes: extracting the red band and near-infrared band data of the satellite remote sensing data from the remote sensing dataset, and calculating the vegetation index based on the following formula:

[0073]

[0074] where NDVI represents the vegetation index, NIR represents the near-red band reflection characteristics of the target area, and Red represents the red band reflection characteristics of the target area.

[0075] In an optional implementation manner, obtaining the terrain index from the remote sensing dataset specifically includes: resampling the DEM data in the remote sensing dataset to the same spatial resolution as the NDVI, and using GIS software to calculate the elevation, aspect, and slope in the DEM data, and normalizing the calculation results as terrain features to obtain the terrain index.

[0076] Specifically, in this embodiment, first, digital elevation model (DEM) data needs to be obtained from the remote sensing dataset. The DEM data contains topographic information of the target area, such as altitude, slope, and aspect. To make the DEM data have the same spatial resolution as the normalized difference vegetation index (NDVI), the DEM data needs to be resampled. The purpose of resampling is to ensure that the spatial resolutions of different data sources are consistent, so that they can be accurately corresponded and compared in subsequent analyses.

[0077] After resampling, a geographic information system software is used to analyze the DEM data to calculate the altitude, aspect, and slope of each pixel point. Altitude represents the height of the terrain, aspect represents the orientation of the terrain, and slope represents the steepness of the terrain. These topographic features are of great significance for understanding the microclimate conditions during the litchi flower induction process.

[0078] After calculating the altitude, aspect, and slope, these topographic features are normalized. The purpose of normalization is to convert numerical features with different ranges to the same scale for subsequent analysis and modeling. Normalization methods can include min-max normalization or Z-score normalization, etc.

[0079] Finally, the normalized altitude, aspect, and slope are used as topographic indices. The topographic index comprehensively reflects the topographic features of the target area and can be used to evaluate the impact of topography on litchi flower induction. For example, areas with higher altitudes may have lower temperatures, areas with different aspects may be affected by different lighting conditions, and areas with larger slopes may affect the retention of water and nutrients. Through the topographic index, the impact of the topographic conditions of the target area on litchi flower induction can be understood more comprehensively.

[0080] In an alternative embodiment, the significance of calculating the first flower induction index is to comprehensively evaluate the impact of the vegetation index on litchi flower induction. Specifically, the vegetation index (such as NDVI) reflects the vegetation growth status and photosynthesis efficiency in the litchi planting area, and these factors will affect the growth environment and flower induction conditions of litchi. By introducing the vegetation index and assigning different weight coefficients, the first flower induction index after fusing E XGBoost 、E LSTM and the vegetation index can be calculated by the following formula.

[0081] FCI * =α*E XGBoost +β*E LSTM +γ*NDVI

[0082] where α, β, and γ are all weight coefficients.

[0083] It should be noted that α, β, and γ are all obtained through a genetic algorithm, and the optimization objective of the genetic algorithm is to maximize the Pearson correlation coefficient between the fused first flowering index FCI * and the measured flowering rate. During the optimization process, a random population is generated, and each individual represents a set of possible weight coefficients (α, β, γ). Assuming the population size is N, each individual i can be represented as (α i , β i , γ i ). Then, new weight coefficients (α, β, γ) are continuously obtained through the selection - crossover - mutation - replacement process until the new weight coefficients (α, β, γ) satisfy the termination condition of maximizing the Pearson correlation coefficient between the fused first flowering index FCI * and the measured flowering rate.

[0084] In an alternative embodiment, the first flowering index incorporates vegetation indices. However, to more comprehensively evaluate the stability of litchi flowering induction, the terrain index also needs to be taken into account. The terrain index reflects the terrain characteristics of the litchi planting area, such as altitude, slope, and aspect. These factors have important impacts on the growth environment and flowering conditions of litchi. Therefore, by fusing the first flowering index with the terrain index, a second flowering index can be obtained, which can more comprehensively reflect the potential influencing factors of litchi flowering induction. The second flowering index is obtained by fusing the first flowering index with the terrain index through the following formula:

[0085]

[0086] where k is an empirical coefficient, and E 地形 represents the terrain index.

[0087] In an alternative embodiment, based on the second flowering index, the flowering stability of the target area is evaluated, specifically including: when , the target area is an unstable flowering area; when , the target area is a relatively stable flowering area; when , the target area is a stable flowering area.

[0088] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should be covered by the protection scope of the present invention. The protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for estimating litchi flowering induction based on short-, medium- and long-term forecast data, characterized in that: The method comprises the following steps: Collect litchi phenological period data, soil moisture data, light intensity data and litchi physiological parameter data in the target area to construct a static multi-source heterogeneous data set; Obtain daily historical climate data and short-, medium- and long-term forecast data for the target area, build a climate data set, and obtain satellite remote sensing data for the target area as a remote sensing data set; The static multi-source heterogeneous dataset and the climate dataset were input into the pre-trained XGBoost model to obtain the litchi flowering induction results E in the target area. XGBoost ; The climate dataset is input into the pre-trained LSTM model to obtain the flowering climate index E of the target area. LSTM ; The vegetation index is obtained from the remote sensing dataset and fused with E XGBoost 、E LSTM The first flowering index was obtained with the vegetation index; The first flowering index is fused with the terrain index to obtain the second flowering index, and the flowering stability of the target area is evaluated based on the second flowering index.

2. The method for estimating litchi flowering induction based on short-, medium- and long-term forecast data according to claim 1, characterized in that: The static multi-source heterogeneous dataset and the climate dataset are input into the pre-trained XGBoost model, including: The non-numerical features in the static multi-source heterogeneous data set and the climate data set are encoded and converted into numerical features, and the numerical features are normalized to obtain the preprocessed static multi-source heterogeneous data set and the climate data set; The preprocessed static multi-source heterogeneous dataset and climate dataset were used as input samples and input into the pre-trained XGBoost model to obtain the litchi flowering induction results E in the target area. XGBoost , the output is expressed as: Where K represents the number of decision trees in the XGBoost model, k represents the kth decision tree, and ω k represents the weight of the kth decision tree, T k (x) represents the predicted output of the kth decision tree for the input sample x.

3. A litchi flowering induction estimation method based on short-, medium- and long-term forecast data according to claim 2, characterized in that: Obtain the flowering climate index E of the target area LSTM , specifically including: Set a sliding window with a window size W and a window step size S, and divide the climate data set into time series according to the sliding window; The time series is input into the LSTM model. The LSTM model updates the hidden state and the cell state through the calculation of the forget gate, the input gate, the cell state and the output gate, and performs weighted summation of the hidden states of all time steps through the attention mechanism to obtain the predicted values ​​of the effective accumulated cold amount at low temperature and the effective accumulated heat amount at high temperature respectively. Based on the predicted values ​​of low-temperature effective accumulated cold and high-temperature effective accumulated heat, the flowering climate index E is calculated by the following formula: LSTM : Among them, LC is the predicted value of low-temperature accumulated cold when the daily minimum temperature is between 5-15℃, and HC is the predicted value of high-temperature accumulated heat when the daily maximum temperature is ≥28℃.

4. The method for estimating litchi flowering induction based on short-, medium- and long-term forecast data according to claim 3, characterized in that: Obtaining the vegetation index from the remote sensing data set specifically includes: extracting red light band and near infrared band data of satellite remote sensing data from the remote sensing data set, and calculating the vegetation index based on the following formula: Among them, NDVI represents the vegetation index, NIR represents the near-infrared light band reflectance characteristics of the target area, and Red represents the red light band reflectance characteristics of the target area.

5. The method for estimating litchi flowering induction based on short-, medium- and long-term forecast data according to claim 4, characterized in that: Obtaining a terrain index from the remote sensing data set specifically includes: resampling the DEM data in the remote sensing data set to the same spatial resolution as the NDVI, calculating the altitude, aspect and slope in the DEM data using GIS software, and normalizing the calculation results as terrain features to obtain a terrain index.

6. A litchi flowering induction estimation method based on short-, medium- and long-term forecast data according to claim 5, characterized in that: The fusion E is calculated by the following formula XGBoost 、E LSTM The first flowering index after the vegetation index, FCI * =α*E XGBoost +β*E LSTM +γ*NDVI Among them, α, β, and γ are all weight coefficients.

7. A litchi flowering induction prediction method based on short-, medium- and long-term forecast data according to claim 6, characterized in that: The second flowering index is obtained by combining the first flowering index with the terrain index through the following formula: Among them, k is the empirical coefficient, E 地形 Represents the terrain index.

8. The method for estimating litchi flowering induction based on short-, medium- and long-term forecast data according to claim 7, characterized in that: The second flowering index is used to evaluate the flowering stability of the target area, including: When When When , the target area is the flowering stable area.

Images