Large-range grassland phenological period prediction method based on hydrothermal optical dynamic threshold value method

Abstract

The invention discloses a large-range grassland phenological period prediction method based on a hydrothermal optical dynamic threshold value method. The method comprises the following steps: acquiring remote sensing data and hydrothermal optical factor data of vegetation in a research area; performing reconstruction fitting on the vegetation time sequence curve; respectively carrying out a cumulative average value method and cumulative processing on each hydrothermal light factor in two months before the start of the key phenological period; the explanatory variables are scored by using a Knock off method; carrying out random forest modeling again by utilizing the predictive factor and the key phenological period; and determining a multi-hydrothermal thermo-optical factor demand dynamic threshold in a large-area range, and performing grassland phenological period prediction according to a plant phenological model. According to the method, the effect of the hydrothermal-optical factors on the plant in the important process of plant growth is decoupled, the threshold value of the multiple hydrothermal-optical factors is quantified, the dynamic threshold value space-time distribution condition of the multiple hydrothermal-optical factors is determined, and the problem that a large-area range of a specific environment factor threshold value is not suitable for an existing plant phenology model is solved.

Description

Technical Field

[0001] The invention relates to the field of vegetation phenology, and in particular to a large-scale grassland phenological period prediction method based on a hydrothermal-photodynamic threshold method. Background Art

[0002] Vegetation phenology refers to the natural phenomenon of vegetation in an annual cycle due to the influence of climate and other environmental factors. It is an important parameter for vegetation to respond to interannual climate change. Accurate monitoring of vegetation phenology is crucial for simulating the carbon cycle of terrestrial ecosystems and understanding the response and feedback of terrestrial ecosystems to climate change. At present, the monitoring technologies for plant phenology are mainly divided into ground observations such as manual recording, camera photography and ground flux station observations, satellite remote sensing monitoring and plant phenology models.

[0003] Ground observations of plant phenology are mainly based on field observations to explore growth events at the scale of a plant or species, such as germination, leafing, flowering, and fruiting. The relevant vegetation data obtained by satellite remote sensing can be effectively applied to regional-scale research, mainly by obtaining phenological indicators through the changing characteristics of vegetation parameter time series data. However, since ground observations are mainly concentrated on site-scale research and are discontinuous in spatial scale, there are many limitations in phenological monitoring work in large areas and long time series. Satellite remote sensing technology can effectively realize the conversion of spatial scales from point to surface for phenological monitoring, but the temporal and spatial resolution of remote sensing image data is low, and there is a large uncertainty in the accuracy of identifying the phenology of different vegetation types, and it cannot reflect the physiological process mechanism of plant phenological changes.

[0004] Plant phenology model is an important research method for simulating and predicting vegetation phenology. It mainly studies the physiological mechanism, environmental threshold and driving factors of plant phenology based on the demand or response of plant growth and development to environmental factors such as water, heat and light. Existing research has constructed a large number of plant phenology models, but most of them only consider a few environmental factors and set specific environmental thresholds to simulate and predict plant phenology. For example, the accumulated temperature model assumes that plants begin to turn green when the accumulated temperature reaches 0℃, but the differences in the sensitivity of different plant growth to factors such as water, heat and light and their spatial heterogeneity limit its application at the regional scale. In addition, the current plant phenology models consider many parameters, which leads to great uncertainty in the simulation and prediction results. Summary of the invention

[0005] Purpose of the invention: In view of the above problems, the purpose of the present invention is to provide a large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method.

[0006] Technical solution: The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method of the present invention comprises: Obtain remote sensing data and water, heat and light factor related data of vegetation in the study area, interpolate the remote sensing data and water, heat and light factor related data respectively, and unify the resolution to obtain the vegetation time series curve; Reconstruct and fit the vegetation time series curve, and extract the key phenological period of the vegetation time series curve; The water, heat and light factors within two months before the beginning of the key phenological period were processed by cumulative average method and accumulation, and all the average values ​​and accumulated values ​​were used as explanatory variables. The Knock off method is used to score the explanatory variables based on the random forest model, and the explanatory variables with scores higher than the scoring threshold are selected as predictors; The random forest model was re-conducted using the predictors and key phenological periods to obtain the plant phenology model; All prediction factors are input into the plant phenology model to determine the dynamic thresholds of the multiple water, heat and light factor requirements in the study area, and then the grassland phenology period is predicted based on the plant phenology model.

[0007] Furthermore, the vegetation time series curve is reconstructed, and the key phenological periods of vegetation are extracted, including: The SG filter was used to reconstruct the vegetation time series curve, and the curvature curve method was used to extract the key phenological period of the reconstructed vegetation time series curve; the key phenological period refers to the time from the beginning of the vegetation greening period to the end of the vegetation yellowing period.

[0008] Furthermore, the random forest model is used to score the explanatory variables, and the explanatory variables with scores higher than the scoring threshold are selected as predictors, including: The first j The explanatory variables are denoted as X j , construct the counterfeit factors corresponding to each explanatory variable ,Will X j and Input into the random forest model for factor importance scoring, and the obtained factor importance scores are recorded as Y j and ; Calculate the competitive scores of explanatory variables based on factor importance scores ; For a given error rate level , based on competitive ratings Calculate the score threshold for the rejection threshold T , the expression is: , In the formula, the set It means that if the explanatory variable is invalid, the competitiveness score of the variable is independent and follows a symmetrical distribution centered on 0; Represents a candidate competitive scoring threshold, belonging to the set , and is less than or equal to .

[0009] Furthermore, all the prediction factors were input into the plant phenology model to determine the dynamic thresholds of multiple water, heat and light factor requirements in the study area, including: According to all the predictors and plant phenology models, the best phenological information is obtained. v predictors, among which v is less than the number of all predictors, quantifying the v The numerical value of each predictor serves as the dynamic threshold of the demand for multiple water, heat and light factors.

[0010] Furthermore, the prediction of grassland phenology based on the plant phenology model includes: The dynamic thresholds of multiple water, heat and light factor requirements in the same period of history are input into the plant phenology model to predict the grassland phenology in a certain period in the future.

[0011] Furthermore, the key phenological periods of the reconstructed vegetation time series curve extracted using the curvature curve method include: Calculate the curvature of the reconstructed vegetation time series curve , the calculation formula is: , In the formula, and They represent the first and second derivatives of the reconstructed vegetation time series curve respectively; The curvature The horizontal axis corresponding to the first local maximum value in the curve change rate is defined as the beginning of the greening period, and the horizontal axis corresponding to the last local minimum value is defined as the end of the yellowing period.

[0012] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention combines the plant phenological growth process mechanism and the random forest model to construct a plant phenological model, decouples the effects of water, heat and light factors on plants in the important process of plant growth, and quantifies the thresholds of multiple water, heat and light factors; then simulates plant phenology in a large area, determines the spatiotemporal distribution of the dynamic thresholds of multiple water, heat and light factors, and solves the problem of the inapplicability of previous plant phenological models that use specific environmental factor thresholds in large areas; and improves the accuracy of the results of the plant phenological model through machine learning methods, which has important reference significance for the stability of the entire alpine grassland ecosystem and the development trend of grassland and animal husbandry. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flowchart of the large-scale grassland phenological period prediction method based on the water-heat-light dynamic threshold method; Figure 2 This is a comparison chart before and after reconstruction of the vegetation time series curve. DETAILED DESCRIPTION

[0014] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0015] The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method described in this embodiment is as follows: Figure 1 As shown, the following steps are included: Step 1, obtain remote sensing data and water, heat and light factor related data of vegetation in the study area, interpolate the remote sensing data and water, heat and light factor related data respectively, and unify the resolution to obtain the vegetation time series curve; Step 2, reconstruct and fit the vegetation time series curve, and extract the key phenological period of the vegetation time series curve; Step 3, the water, heat and light factors within two months before the beginning of the key phenological period are processed by cumulative average method and accumulation, and all the average values ​​and accumulated values ​​are used as explanatory variables; Step 4: Based on the random forest model, the explanatory variables are scored using the knock off method, and the explanatory variables with scores higher than the scoring threshold are selected as predictors; Step 5, re-conduct random forest modeling using the prediction factors and key phenological periods to obtain a plant phenology model; Step 6: Input all prediction factors into the plant phenology model to determine the dynamic thresholds of the multiple water, heat and light factor requirements in the study area, and then predict the grassland phenology period based on the plant phenology model.

[0016] The present invention aims at the problems existing in ground observation, satellite remote sensing monitoring and plant phenology models, comprehensively considers the demand and response to water, heat and light factors in the important process of plant growth, constructs a plant phenology model in combination with the plant phenology growth process mechanism and the random deep forest model, decouples the effects of water, heat and light factors on plants in the important process of plant growth and quantifies the thresholds of multiple water, heat and light factors, then simulates plant phenology in a large area, determines the spatiotemporal distribution of dynamic thresholds of multiple water, heat and light factors, solves the inapplicability of previous plant phenology models that use specific environmental factor thresholds in large areas, and also improves the accuracy of the results of the plant phenology model through machine learning methods, which has important reference significance for the stability of the entire alpine grassland ecosystem and the development trend of grassland and animal husbandry.

[0017] In one example, the corresponding remote sensing data and water, heat and light factor related data in the study area are obtained, where the remote sensing data comes from the MODIS satellite remote sensing data source, and the remote sensing data includes the normalized difference vegetation index (NDVI) and alpine grassland data; the NDVI data comes from the MOD13Q1 product data provided by Terra / MODIS, and the alpine grassland data comes from the MOD12Q1 land cover data provided by MODIS Terra+Aqua. Water, heat and light factors include precipitation, air temperature, soil temperature, soil moisture, saturated water pressure difference and photosynthetically active radiation related data. Among them, the precipitation and air temperature data come from the China Regional Meteorological Element Driven Dataset, the soil moisture data comes from the China Soil Moisture or Soil Moisture Dataset, the soil temperature data and saturated water vapor pressure difference data are both derived from the ERA5-Land reanalysis data, and the photosynthetically active radiation data comes from the global long-sequence high-resolution photosynthetically active radiation data set. The spatial resolution of the above-mentioned related data is unified to 0.1 degree × 0.1 degree using the linear interpolation method, and the time length is selected from 2001 to 2018. The NDVI time series growth curve is constructed based on the NDVI data, such as Figure 2 As shown, the horizontal axis is the time series and the vertical axis is the NDVI value.

[0018] Furthermore, the vegetation time series curve is reconstructed, and the key phenological periods of vegetation are extracted, including: The SG filter was used to reconstruct the vegetation time series curve, and the curvature curve method was used to extract the key phenological period of the reconstructed vegetation time series curve; the key phenological period refers to the time from the beginning of the vegetation greening period to the end of the yellowing period.

[0019] NDVI data is affected by clouds, aerosols, and water vapor, so its time series needs to be reconstructed and fitted to obtain a growth process line that conforms to vegetation growth. In this example, a suitable window width is set in the SG filter, and the NDVI time series curve is reconstructed and fitted through the SG filter. The calculation formula of the SG filter is as follows: , In the formula, Indicates j +1 original value of NDVI data, Indicates j The fitted value of NDVI, Indicates i NDVI value filter coefficient, m Represents the width of the window; N represents the number of convolutions, which is equal to the width of the sliding array 2 m +1.

[0020] In this example, the width of the SG filter window is set to 9, and the reconstructed time series curve is as follows: Figure 2 shown.

[0021] Furthermore, the key phenological periods of the reconstructed vegetation time series curve extracted using the curvature curve method include: Calculate the curvature of the reconstructed vegetation time series curve , the calculation formula is: , In the formula, and They represent the first and second derivatives of the reconstructed vegetation time series curve respectively; The curvature The horizontal axis corresponding to the first local maximum value in the curve change rate is defined as the beginning of the greening period, and the horizontal axis corresponding to the last local minimum value is defined as the end of the yellowing period.

[0022] The vegetation growth process is affected by the comprehensive influence of water, heat and light factors in the period before the critical phenological period. Therefore, in this example, the different water, heat and light factors of the first two months are processed by the cumulative average method and accumulation from the critical phenological period to the previous two months. Taking the precipitation data of the previous two months before the critical phenological period as an example, the precipitation of the previous day is the precipitation of the previous day, and the precipitation of the previous two days is the average of the precipitation of the previous day and the previous day. If we push forward 60 days in sequence, we can get 60 cumulative average precipitation factors; similarly, if we push forward 60 days for the five water, heat and light factors such as temperature, soil temperature, soil moisture, saturated water vapor pressure difference, and photosynthetically active radiation, we can get a total of 6×60=360 cumulative average water, heat and light factors related to phenology. The calculation formula of the cumulative average method is: , In the formula, Indicates the first n The cumulative average of the day, in this example, is 60; Indicates the first n The specific value of the water, heat and light factor of the day.

[0023] Similarly, the calculation of the cumulative value takes the precipitation data of the two months before the key phenological period as an example. The precipitation of the previous day is the precipitation of the previous day, and the precipitation of the previous two days is the cumulative value of the precipitation of the previous day and the previous day. If we push forward 60 days in sequence, we can get 60 cumulative precipitation factors. Similarly, if we push forward 60 days, we can get 60 cumulative temperature factors. In this example, only 120 cumulative factors are selected, and finally 360+120=480 water, heat and light prediction factors are obtained, and these 480 water, heat and light factors are used as explanatory variables.

[0024] Furthermore, the Knock off method is used to score the explanatory variables based on the random forest model, and the explanatory variables with scores higher than the scoring threshold are selected as predictors, including: The first j The explanatory variables are denoted as X j , construct the counterfeit factors corresponding to each explanatory variable ,Will X j and Input into the random forest model for factor importance scoring, and the obtained factor importance scores are recorded as Y j and ; The counterfeit factor can be constructed by the knockpy function in python; Calculate the competitive scores of explanatory variables based on factor importance scores ; For a given error rate level , based on competitive ratings Calculate the score threshold for the rejection threshold T , the expression is: , In the formula, the set It means that if the explanatory variable is invalid, the competitiveness score of the variable is independent and follows a symmetrical distribution centered on 0; represents a candidate competitive scoring threshold, which belongs to the set , and needs to meet certain error rate conditions, that is, less than or equal to Finally, the predictors with scores higher than T are selected, and the false discovery rate FDR of the screening factors can be strictly controlled to be no higher than .

[0025] Furthermore, the screened predictors were input into the plant phenology model to determine the dynamic thresholds of multiple water, heat and light factor requirements in the study area, including: According to all the predictors and plant phenology models, the best phenological information is obtained. v predictors, among which v is less than the number of all predictors, quantifying the v The numerical value of each predictor serves as the dynamic threshold of the demand for multiple water, heat and light factors.

[0026] In this example, the Knock off method is used based on random forest to screen out explanatory variables with importance scores higher than the scoring threshold as predictors, and the screened predictors and the key phenological periods of the corresponding grids are used to re-perform random forest modeling. That is, the daily cumulative average method and the cumulative value of each water, heat and light factor in the first two months of the key phenological period are used as predictors, and the corresponding key phenological period is used as the dependent variable to reconstruct the sample set for random forest modeling. The random forest model is trained with the training samples, and the trained random forest model is used as the plant phenology model. The training samples consist of the screened predictors and the corresponding dependent variables. By inputting all the predictors into the plant phenology model, the one that best affects the phenological information can be determined. v predictors, v You can assign values ​​according to actual needs, such as in this example v =3, that is, 3 prediction factors are finally retained, and the actual values ​​of the 3 prediction factors are quantified as the dynamic thresholds of the multi-water, thermal and light factor requirements within the current study area.

[0027] Furthermore, the prediction of grassland phenology based on the plant phenology model includes: The dynamic thresholds of multiple water, heat and light factor requirements in the same period of history are input into the plant phenology model to predict the grassland phenology in a certain period in the future.

Claims

1. A large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method, characterized in that: include: Obtain remote sensing data and water, heat and light factor related data of vegetation in the study area, interpolate the remote sensing data and water, heat and light factor related data respectively, and unify the resolution to obtain the vegetation time series curve; Reconstruct and fit the vegetation time series curve, and extract the key phenological period of the vegetation time series curve; The water, heat and light factors within two months before the beginning of the key phenological period were processed by cumulative average method and accumulation, and all the average values ​​and accumulated values ​​were used as explanatory variables. The Knock off method is used to score the explanatory variables based on the random forest model, and the explanatory variables with scores higher than the scoring threshold are selected as predictors; The random forest model was re-conducted using the predictors and key phenological periods to obtain the plant phenology model; All prediction factors are input into the plant phenology model to determine the dynamic thresholds of the multiple water, heat and light factor requirements in the study area, and then the grassland phenology period is predicted based on the plant phenology model.

2. The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method according to claim 1 is characterized in that: The vegetation time series curve is reconstructed and the key phenological periods of vegetation are extracted, including: The SG filter was used to reconstruct the vegetation time series curve, and the curvature curve method was used to extract the key phenological period of the reconstructed vegetation time series curve; the key phenological period refers to the time from the beginning of the vegetation greening period to the end of the vegetation yellowing period.

3. The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method according to claim 2 is characterized in that: Based on the random forest model, the knock off method is used to score the explanatory variables, and the explanatory variables with scores higher than the scoring threshold are selected as predictors, including: The first j The explanatory variables are denoted as X j , construct the counterfeit factors corresponding to each explanatory variable ,Will X j and Input into the random forest model for factor importance scoring, and the obtained factor importance scores are recorded as Y j and ; Calculate the competitive scores of explanatory variables based on factor importance scores ; For a given error rate level , based on competitive ratings Calculate the score threshold for the rejection threshold T , the expression is: ， In the formula, the set It means that if the explanatory variable is invalid, the competitiveness score of the variable is independent and follows a symmetrical distribution centered on 0; Represents a candidate competitive scoring threshold, belonging to the set , and is less than or equal to .

4. The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method according to claim 3 is characterized in that: All prediction factors were input into the plant phenology model to determine the dynamic thresholds of multiple water, heat and light factor requirements in the study area, including: According to all the predictors and plant phenology models, the best phenological information is obtained. v predictors, among which v is less than the number of all predictors, quantifying the v The numerical value of each predictor serves as the dynamic threshold of the demand for multiple water, heat and light factors.

5. The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method according to claim 4 is characterized in that: The prediction of grassland phenology based on plant phenology model includes: The dynamic thresholds of multiple water, heat and light factor requirements in the same period of history are input into the plant phenology model to predict the grassland phenology in a certain period in the future.

6. The large-scale grassland phenological period prediction method based on the hydrothermal-photodynamic threshold method according to claim 5 is characterized in that: The key phenological periods of the vegetation time series curve extracted and reconstructed using the curvature curve method include: Calculate the curvature of the reconstructed vegetation time series curve , the calculation formula is: ， In the formula, and They represent the first and second derivatives of the reconstructed vegetation time series curve respectively; The curvature The horizontal axis corresponding to the first local maximum value in the curve change rate is defined as the beginning of the greening period, and the horizontal axis corresponding to the last local minimum value is defined as the end of the yellowing period.

Images