Large-area medium-spatial-resolution time sequence vegetation leaf area index estimation method, device and equipment cooperating with satellite-borne laser radar and multispectral remote sensing and medium

Abstract

The invention discloses a large-area medium-spatial-resolution time sequence vegetation leaf area index estimation method and device cooperating with satellite-borne laser radar and multispectral remote sensing, equipment and a medium, and relates to the field of vegetation structure parameter estimation. The method comprises the following steps: acquiring multispectral reflectivity data and satellite-borne laser radar data of a to-be-measured area in a to-be-measured time period; constructing a multi-feature data set based on the multi-spectral reflectivity data; performing space and time matching on the multi-feature data set and the satellite-borne laser radar data to obtain matched data; inputting the matching data and the satellite-borne laser radar data into the estimation model to obtain an LAI estimation result; the estimation model is obtained by training a deep neural network; the estimation model is used for determining a mapping relation between the multi-feature data set subjected to space and time matching and the satellite-borne laser radar data, and further determining an LAI estimation result. According to the invention, the LAI estimation precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of vegetation structure parameter estimation, and in particular to a method, device, equipment and medium for estimating a time-series vegetation leaf area index with medium spatial resolution in a large area by coordinating satellite-borne laser radar and multispectral remote sensing. Background Art

[0002] The Sentinel-2Level 2Prototype Processor (SL2P) tool can achieve time-series leaf area index (LAI) inversion with medium spatial resolution, but the accuracy is poor for forest areas. The leaf area index is half of the total leaf area per unit ground area and is an important vegetation structure parameter. From early one-dimensional radiation transfer models of vegetation canopies such as Suits and SAIL to the spectral invariance theory that represents the development direction of forward models, the leaf area index has always been the key to parameterizing complex radiation transfer processes. Forward simulation-backward inversion is a classic paradigm for obtaining leaf area index using satellite data. MODIS LAI products, VIIRS LAI products and SL2P tools produced based on this paradigm have promoted the rapid development of fields such as carbon cycle, land-atmosphere interaction and vegetation growth monitoring.

[0003] Among them, the SL2P tool uses PROSAIL forward simulation combined with a neural network to establish a lookup table, and combined with the characteristics of the Sentinel-2A and Sentinel-2B dual-satellite network, which can achieve a 5-day revisit period, it can quickly achieve a 20-meter spatial resolution of nearly 5-day revisit period of time series LAI inversion. However, many studies have pointed out that although the LAI produced by S2LP (hereinafter referred to as SL2P LAI product) can meet the accuracy requirements of applications in crop areas, the PROSAIL model used by SL2P is not suitable for spatially non-continuous forests, resulting in large errors in forest areas.

[0004] The Global Ecosystem Dynamics Investigation (GEDI) data has a 4-day orbital revisit period and has accuracy advantages in forest LAI estimation, but it is spatially discontinuous, has obvious position errors, and has quality differences between spots.

[0005] GEDI is the first satellite-borne large-spot full-waveform lidar data for global forest structure monitoring. Its 25-meter diameter spot has a similar spatial scale to Sentinel-2. Its 4-day orbital revisit period has the potential to reflect the temporal dynamics of forest LAI. Its 60-meter spot spacing along the track can provide high-density spatial sampling. At the same time, GEDI's direct observation capability of vegetation canopy structure enables it to effectively overcome the saturation effect that is common in passive optical remote sensing data. The near-global spot LAI product produced by GEDI based on the GORT model has an accuracy advantage in forest areas compared to existing passive optical-based inversion results. However, GEDI is a spatially discrete sampling observation, and it is impossible to estimate spatially continuous LAI alone. At the same time, GEDI has a more obvious position error and there are quality differences between different spots. Summary of the invention

[0006] The purpose of this application is to provide a method, device, equipment and medium for estimating the time-series vegetation leaf area index of a large area with medium spatial resolution by coordinating space-borne laser radar and multispectral remote sensing, which can improve the accuracy of LAI estimation.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for estimating leaf area index of vegetation in a large area with medium spatial resolution in a coordinated manner with spaceborne laser radar and multispectral remote sensing, comprising:

[0009] Obtain multispectral reflectance data and satellite-borne lidar data for the area to be measured during the period to be measured; the multispectral reflectance data is obtained using Sentinel-2, based on the high spatiotemporal resolution characteristics of a 20-meter spatial resolution and a 5-day revisit cycle; the satellite-borne lidar data is based on the characteristics of a 4-day orbital revisit cycle, and all GEDI data for the Global Ecosystem Dynamics Survey are obtained;

[0010] constructing a multi-feature data set based on the multispectral reflectance data;

[0011] Matching the multi-feature data set with the satellite-borne laser radar data in space and time to obtain matching data;

[0012] The matching data and the satellite-borne laser radar data are input into an estimation model to obtain an LAI estimation result; the estimation model is trained using a deep neural network; the estimation model is used to determine a mapping relationship between a multi-feature data set matched in space and time and the satellite-borne laser radar data, and then determine the LAI estimation result.

[0013] Optionally, constructing a multi-feature data set based on the multispectral reflectance data specifically includes:

[0014] Based on the multispectral reflectance data, effective pixel extraction processing is performed to obtain effective pixel data; wherein, cloud shadow scenes are extracted with SCL=3, and cloud scenes are extracted with SCL=10; effective pixel data are extracted with MSK_CLDPRB less than 5%, MSK_SNWPRB less than 5% and non-cloud shadow scenes and non-cloud scenes; SCL is the scene classification band; MSK_CLDPRB is the cloud probability; MSK_SNWPRB is the snow probability;

[0015] The multi-feature data set is determined according to the effective pixel data.

[0016] Optionally, the multi-feature data set includes: reflectivity, multiple vegetation indices, solar zenith angle, and geospatial information;

[0017] The reflectivity includes visible light, near infrared and short-wave infrared; the geospatial information is used to measure spatial proximity; the geospatial information includes:

[0018]

[0019] Among them, Spatial is the geographic spatial information; R is a constant, set to 1; θ is the longitude of the valid pixel data; is the latitude of the valid pixel data.

[0020] Optionally, spatially and temporally matching the multi-feature data set and the spaceborne laser radar data to obtain matching data specifically includes:

[0021] Based on the set screening rules, the GEDI data is screened to obtain the screened GEDI LAI spots;

[0022] For each valid pixel data in the multi-feature data set, the GEDI LAI light spot is searched for by valid pixel data one by one, based on the interval time before and after the observation date of the valid pixel data at the same position within the set time period, so as to time match the multi-feature data set and the GEDI LAI light spot, and obtain the searched GEDI LAI light spot;

[0023] Centered on the pixel where the searched GEDI LAI spot is located, a search is performed in the multi-feature dataset based on a 3×3 pixel block to spatially match the multi-feature dataset and the GEDI LAI spot, and obtain near-synchronous observation samples of the multi-feature set based on pixel blocks;

[0024] The searched GEDI LAI spots and multi-feature set near-synchronous observation samples are identified as matching data.

[0025] Optionally, the method for determining the estimation model specifically includes:

[0026] Build deep neural networks;

[0027] Inputting historical matching data, historical satellite-borne lidar data and corresponding LAI results into a deep neural network, training the network parameters of the deep neural network with the goal of minimizing the mean square error between the output of the deep neural network and the LAI result to obtain a trained deep neural network; the LAI result is the leaf area index of the GEDI L2B product in the GEDI data;

[0028] The trained deep neural network is determined as the estimation model.

[0029] Optionally, the deep neural network adopts an encoding-decoding architecture; wherein the encoding part is composed of a series connection of Residuals with convolution kernels of 64, 128, and 256 respectively; and the decoding part is composed of a series connection of Residuals with convolution kernels of 128 and 64 respectively.

[0030] Optionally, the mathematical expression for training the network parameters is:

[0031]

[0032] Among them, Φ is the network parameter; N is the total number of samples in the nearly synchronous observation samples in the multi-feature dataset; ω i is the weight of the i-th sample; is the output value corresponding to the i-th sample in the deep neural network; is the LAI result corresponding to the i-th sample.

[0033] In a second aspect, the present application provides a large-area medium spatial resolution time series vegetation leaf area index estimation device that cooperates with space-borne laser radar and multispectral remote sensing, comprising:

[0034] The data acquisition module is used to obtain multispectral reflectance data and satellite-borne laser radar data of the area to be measured during the period to be measured; the multispectral reflectance data is obtained by using Sentinel-2, based on the high spatiotemporal resolution characteristics of 20-meter spatial resolution and 5-day revisit cycle; the satellite-borne laser radar data is all GEDI data of the Global Ecosystem Dynamics Survey obtained based on the characteristics of the 4-day orbital revisit cycle;

[0035] A data set construction module, used to construct a multi-feature data set based on the multispectral reflectance data;

[0036] A matching module, used for matching the multi-feature data set with the space-borne laser radar data in space and time to obtain matching data;

[0037] An estimation module is used to input the matching data and the satellite-borne lidar data into an estimation model to obtain an LAI estimation result; the estimation model is trained using a deep neural network; the estimation model is used to determine the mapping relationship between a multi-feature data set matched in space and time and the GEDI data, and then determine the LAI estimation result.

[0038] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for estimating vegetation leaf area index in a large area with medium spatial resolution by cooperating with satellite-borne lidar and multispectral remote sensing.

[0039] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for estimating vegetation leaf area index in a large area with medium spatial resolution by cooperating with satellite-borne lidar and multispectral remote sensing.

[0040] According to the specific embodiments provided in this application, this application has the following technical effects:

[0041] The present application provides a method, device, equipment and medium for estimating the time-series vegetation leaf area index of a large area with medium spatial resolution by coordinating satellite-borne laser radar and multispectral remote sensing, by obtaining multispectral reflectance data of the area to be measured during the period to be measured; constructing a multi-feature data set based on the reflectance data; obtaining satellite-borne laser radar data of the area to be measured during the period to be measured; inputting the multi-feature data set matched in space and time and the satellite-borne laser radar data into the estimation model to obtain the LAI estimation result. Through the point-surface fusion of GEDI LAI and Sentinel-2, the characteristic of GEDI in overcoming the saturation effect that is common in passive optical remote sensing data can be brought into play, so that the estimation accuracy is higher than the accuracy based on passive optical remote sensing inversion. In addition, the estimation model is obtained by training with a deep neural network, has a stronger generalization ability, can fully mine the effective information of the Sentinel-2 multi-feature set, and can utilize the time-space dense sampling characteristics of GEDI LAI; therefore, the present application can improve the accuracy of LAI estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1A flowchart of the method for estimating the leaf area index of vegetation in a large area with medium spatial resolution using spaceborne lidar and multispectral remote sensing.

[0044] Figure 2 Schematic diagram of the network structure of a deep neural network;

[0045] Figure 3 This is a schematic diagram of the Residual structure;

[0046] Figure 4 It is a schematic diagram of the ECA structure;

[0047] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] This application intends to take advantage of the high spatiotemporal resolution characteristics of Sentinel-2 data, the 4-day orbit revisit period of GEDI and the accuracy of forest LAI estimation. Taking into account the quality factors of GEDI data, a medium spatial resolution time series vegetation leaf area index estimation method for large areas that coordinates spaceborne lidar and multispectral remote sensing is proposed to improve the accuracy of existing SL2P tools in forest areas.

[0050] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0051] In an exemplary embodiment, Figure 1 As shown, a method for estimating the time series vegetation leaf area index of a large area with medium spatial resolution in coordination with spaceborne laser radar and multispectral remote sensing is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to a server as an example for explanation, and includes the following steps.

[0052] Step 100: Obtain multispectral reflectance data and satellite-borne lidar data of the area to be measured during the period to be measured. The multispectral reflectance data is obtained using Sentinel-2, based on the high spatiotemporal resolution characteristics of a 20-meter spatial resolution and a 5-day revisit period; the satellite-borne lidar data is based on the characteristics of a 4-day orbital revisit period, and all GEDI data of the Global Ecosystem Dynamics Survey are obtained.

[0053] Step 200: construct a multi-feature dataset based on multispectral reflectance data.

[0054] Step 300: Match the multi-feature data set and the satellite-borne lidar data in space and time to obtain matching data.

[0055] Step 400: Input the matching data and the satellite-borne laser radar data into the estimation model to obtain the LAI estimation result. The estimation model is trained using a deep neural network; the estimation model is used to determine the mapping relationship between the multi-feature data set matched in space and time and the satellite-borne laser radar data, and then determine the LAI estimation result.

[0056] In one embodiment, constructing a multi-feature data set based on multispectral reflectance data specifically includes:

[0057] Based on the multispectral reflectance data, effective pixel extraction and processing are performed to obtain effective pixel data; among them, cloud shadow scenes are extracted with SCL=3, and cloud scenes are extracted with SCL=10; effective pixel data are extracted when MSK_CLDPRB is less than 5%, MSK_SNWPRB is less than 5% and non-cloud shadow scenes and non-cloud scenes; SCL is the scene classification band; MSK_CLDPRB is the cloud probability; MSK_SNWPRB is the snow probability.

[0058] A multi-feature data set is determined based on the effective pixel data.

[0059] The multi-feature dataset includes reflectance, multiple vegetation indices, solar zenith angle, and geospatial information; reflectance includes visible light, near infrared, and short-wave infrared; geospatial information is used to measure spatial proximity. Geospatial information includes:

[0060]

[0061] Among them, Spatial is the geographic spatial information; R is a constant, set to 1; θ is the longitude of the valid pixel data; is the latitude of the valid pixel data.

[0062] In one embodiment, spatially and temporally matching the multi-feature data set and the spaceborne laser radar data to obtain matching data specifically includes:

[0063] Based on the set screening rules, the GEDI data are screened to obtain the screened GEDI LAI spots.

[0064] For each valid pixel data in the multi-feature data set, the GEDI LAI spot is searched for each valid pixel data with the same position and the interval time before and after the observation date of the valid pixel data within the set time period as a benchmark, so as to perform time matching between the multi-feature data set and the GEDI LAI spot to obtain the searched GEDI LAI spot.

[0065] Centered on the pixel where the searched GEDI LAI spot is located, a search is performed based on a 3×3 pixel block in the multi-feature dataset to spatially match the multi-feature dataset and the GEDI LAI spot, and a near-synchronous observation sample of the multi-feature set based on the pixel block is obtained.

[0066] The searched GEDI LAI spots and multi-feature set near-synchronous observation samples are identified as matching data.

[0067] In one embodiment, the method for determining the estimation model specifically includes:

[0068] Construct a deep neural network; input historical matching data, historical satellite-borne lidar data, and corresponding LAI results into the deep neural network, and train the network parameters of the deep neural network with the goal of minimizing the mean square error between the output of the deep neural network and the LAI result to obtain a trained deep neural network. The LAI result is the leaf area index of the GEDI L2B product in the GEDI data.

[0069] The trained deep neural network is determined as the estimation model.

[0070] The deep neural network adopts an encoding-decoding architecture; the encoding part is composed of a series of Residual networks with 64, 128, and 256 convolution kernels respectively; the decoding part is composed of a series of Residual networks with 128 and 64 convolution kernels respectively.

[0071] In one embodiment, the mathematical expression for training the network parameters is:

[0072]

[0073] Among them, Φ is the network parameter; N is the total number of samples in the nearly synchronous observation samples in the multi-feature dataset; ω i is the weight of the i-th sample; is the output value corresponding to the i-th sample in the deep neural network; is the LAI result corresponding to the i-th sample.

[0074] In practical applications, the specific operating steps of the method mentioned in this application can be as follows:

[0075] Step 1: Obtain all Sentinel-2 reflectance data for the estimated period in the estimated area, extract all valid pixels of Sentinel-2 according to the MSK_CLDPRB (cloud probability) and MSK_SNWPRB (snow probability) and SCL (scene classification) bands of Sentinel-2, and obtain valid pixel data. Specifically, first extract cloud shadows and clouds with SCL=3 and SCL=10 respectively. Then extract valid pixels of Sentinel-2 with MSK_CLDPRB less than 5%, MSK_SNWPRB less than 5% and non-cloud shadow and cloud scenes.

[0076] Step 2: Construct a Sentinel-2 multi-feature dataset based on Sentinel-2 reflectance data. The Sentinel-2 multi-feature dataset includes 51 features, including Sentinel-2 reflectance (Bandx), various vegetation indices (VIx), solar zenith angle (SZA), and geospatial information (Spatial). Among them, reflectance Bandx includes visible light (Band2, Band3, Band4), near infrared (Band5, Band6, Band7, Band8, Band9, Band10, Band11, Band12, Band13, Band14, Band15, Band16, Band17, Band18, Band19, Band20, Band21, Band22, Band33, Band44, Band55, Band6, Band7, Band8, Band19, Band23, Band14, Band24, Band25, Band26, Band27, Band28, Band29, Band30, Band31, Band32, Band33, Band4 8A ) and short-wave infrared (Band 11 、Band 12 ). The spatial information contains three features, which are used to measure spatial proximity. The expression of the spatial information is:

[0077]

[0078] Among them, Spatial is the geographic spatial information; R is a constant, set to 1; θ is the longitude of the valid pixel data; is the latitude of the valid pixel data.

[0079] Thanks to the unique red edge band of Sentinel-2, the effectiveness of Sentinel-2 vegetation index for LAI inversion has been confirmed. Therefore, multiple vegetation indices VIx of Sentinel-2 are added to the multi-feature set. The calculation formula of VIx is shown in Table 1.

[0080] Table 1 Calculation formulas for vegetation indices involved in VIx

[0081]

[0082]

[0083] Wherein, Bx in Table 1 represents the xth reflectivity band of Sentinel-2.

[0084] Step 3: Obtain all the spots of the Global Ecosystem Dynamics Investigation (GEDI) L2B products from 2019 to 2022 in the estimated area, and extract the leaf area index (LAI) of the GEDI L2B products, hereinafter referred to as GEDI LAI. Use the rules shown in Table 2 to select high-quality GEDI LAI spots.

[0085] Table 2 Screening rules for high-quality GEDI LAI spots

[0086]

[0087] Step 4: Search for the corresponding GEDI LAI light spot for each valid pixel data of Sentinel-2. In order to ensure the temporal consistency of GEDI and Sentinel-2 data, theoretically, we should search for GEDI light spots with the same observation date as the Sentinel-2 multi-feature set (synchronous observation). However, Sentinel-2 is affected by clouds, rain, etc., resulting in limited data for synchronous observations of Sentinel-2 and GEDI.

[0088] To this end, based on the synchronous observation of Sentinel-2 and GEDI, GEDI data with an observation time interval of less than or equal to two days between Sentinel-2 and GEDI are added to ensure the representativeness and quantity of subsequent samples. Specifically, for each valid Sentinel-2 pixel, that is, each valid pixel data in the multi-feature dataset, search for GEDI spots with an observation time interval of less than two days (including 2 days) before and after the observation date of the same location as the valid Sentinel-2 pixel. In the case of multiple GEDI spots within a 2-day time interval, the GEDI spot with the smaller time interval is retained.

[0089] Step 5: Construct a near-synchronous observation sample of the GEDI-Sentinel-2 multi-feature set based on pixel blocks. GEDI has a position error of 10.2 meters, and there is a certain spatial scale difference between the 25-meter diameter spot size of GEDI and the 20-meter spatial resolution of Sentinel-2. In order to reduce the GEDI position error, while considering the spatial scale difference between GEDI and Sentinel-2 and the 60-meter along-track spacing of GEDI, the Sentinel-2 multi-feature set with a size of 3×3 pixel blocks is searched with the Sentinel-2 pixel where the searched GEDI spot is located as the center, and then the near-synchronous observation sample of the GEDI-Sentinel-2 multi-feature set based on pixel blocks is formed.

[0090] Step 6: Use deep learning to construct a mapping between GEDI and its corresponding Sentinel-2 multi-feature set based on 3×3 pixel blocks, that is, use GEDI LAI as output and the corresponding Sentinel-2 multi-feature set of 3×3 pixel blocks as input, and use deep neural network for training. In order to control the uncertainty caused by the difference in GEDI spot quality, the samples are weighted with the sensitivity of GEDI LAI. Sensitivity represents the minimum proportion of all echoes that penetrate the canopy and reach the ground. In theory, the higher the proportion, the higher the accuracy of GEDI LAI. The above training process can be expressed by formula (2).

[0091]

[0092] Among them, Φ is the network parameter; N is the total number of samples in the nearly synchronous observation samples of multiple feature sets; ω i is the weight of the i-th sample; is the output value corresponding to the i-th sample in the deep neural network; is the LAI result corresponding to the i-th sample.

[0093] The network structure of the deep neural network used in deep learning is shown in Figure 2 The network adopts an encoder-decoder architecture. The encoder and decoder use residual networks as basic units to prevent neural network degradation as the depth increases. The encoder part is composed of a series of residuals with 64, 128, and 256 convolution kernels, and the decoder part is composed of a series of residuals with 128 and 64 convolution kernels, which are used to gradually reduce the dimension of the features. Each residual in the encoder and decoder parts is looped 3 times and then Figure 2 As shown in the figure, skip connections are made to increase the connectivity of features between layers. The Sentinel-2 multi-feature set first passes through a two-dimensional convolution (Conv2D) with 64 convolution kernels and a size of 3×3, BatchNorm, and an activation function ReLU, and then is input into the Encoder-Decoder structure to obtain the output.

[0094] Residual structure see Figure 3 , the structure can be expressed as formula (3).

[0095]

[0096] Among them, X l and X l+1They represent the input and output of the lth residual block respectively, and ReLU is used as the activation function. Represents the residual part, which is a series of one-dimensional convolution Conv1D, batch normalization (BatchNorm), activation function ReLU, Conv1D, BatchNorm and attention mechanism (Efficient Channel Attention, ECA), W l Represents the parameters of the residual part. h() is the dimension adjustment function of the skip connection (shortcut), which consists of Conv1D and Batchnormalization and is used to adjust the input X l Reduce or increase the dimension so that its dimension (size()) is the same as The output dimensions remain consistent.

[0097] Among them, the ECA attention mechanism is used to further improve the effectiveness of feature extraction by the network and reduce the collinearity between multiple vegetation indices. Its structure is shown in Figure 4 The ECA attention mechanism can effectively avoid the information loss caused by the channel attention mechanism due to dimensionality reduction, which can be expressed as:

[0098] w=σ(C1D k (y))*y (4)

[0099] Where w is the output of ECA; σ is the Sigmoid function; C1D k () is a one-dimensional convolution with a convolution kernel size of k; y is the input feature of ECA.

[0100] The convolution kernel size k and number C are expressed as:

[0101]

[0102] C=2 γ×k-b (6)

[0103] In the formula, |·| odd represents the nearest odd number, γ and b can be set to 2 and 1 respectively.

[0104] Step 7: Based on the learned mapping relationship That is, the trained deep neural network takes all Sentinel-2 multi-feature sets as input to obtain the LAI estimation result The process can be expressed as:

[0105]

[0106] The estimation accuracy of LAI in forest areas in this application is higher than that of existing SL2P tools. The reason is that through the point-surface fusion of GEDI LAI and Sentinel-2 (related to the idea of ​​fusion of overall GEDI and Sentinel-2, not limited to a certain step in this application), GEDI can overcome the saturation effect commonly found in passive optical remote sensing data, and the accuracy advantage of GEDI LAI products in forest areas compared with existing inversion results based on passive optical remote sensing.

[0107] Compared with the LAI inversion results of forest areas combining measured data and Sentinel, the time series LAI estimation results of this application can more effectively reflect the temporal dynamics of LAI. The reason is that GEDI has a 4-day orbital revisit period. Compared with the limited measured data in time series, GEDI can more completely characterize the temporal dynamics of LAI. The strategy of searching for the corresponding temporally consistent GEDI LAI spot based on each Sentinel-2 effective pixel (step 4) ensures the adequacy of the sample quantity while ensuring the temporal observation consistency of GEDI and Sentinel-2 multi-feature sets.

[0108] The estimation accuracy of LAI in forest areas in this application is higher than similar machine learning methods such as random forest. The reason is that using pixel blocks as input (step 5) can effectively weaken the influence of GEDI position error; in step 6, weighting is performed based on GEDI's sensitivity, which effectively considers the quality differences between different GEDI spots; the deep neural network designed in step 6 has stronger generalization ability than machine learning methods, and can fully mine the effective information of Sentinel-2 multi-feature set.

[0109] Based on the same inventive concept, the embodiment of the present application also provides an estimation device for implementing the above-mentioned method for estimating the time-series vegetation leaf area index of a large area with medium spatial resolution by cooperating with satellite-borne laser radar and multispectral remote sensing. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiments of the device for estimating the time-series vegetation leaf area index of a large area with medium spatial resolution by cooperating with satellite-borne laser radar and multispectral remote sensing provided below can be referred to the limitations of the method for estimating the time-series vegetation leaf area index of a large area with medium spatial resolution by cooperating with satellite-borne laser radar and multispectral remote sensing in the above text, and will not be repeated here.

[0110] In an exemplary embodiment, a large-area medium spatial resolution time series vegetation leaf area index estimation device using satellite-borne laser radar and multispectral remote sensing is provided, comprising:

[0111] The data acquisition module is used to obtain multispectral reflectance data and satellite-borne lidar data of the area to be measured during the period to be measured; the multispectral reflectance data is obtained using Sentinel-2, based on the high spatiotemporal resolution characteristics of a 20-meter spatial resolution and a 5-day revisit period; the satellite-borne lidar data is all GEDI data obtained based on the characteristics of a 4-day orbital revisit period.

[0112] The dataset construction module is used to construct a multi-feature dataset based on multispectral reflectance data.

[0113] The matching module is used to match the multi-feature data set and the space-borne lidar data in space and time to obtain matching data.

[0114] The estimation module is used to input the matching data and the satellite-borne lidar data into the estimation model to obtain the LAI estimation result; the estimation model is trained using a deep neural network; the estimation model is used to determine the mapping relationship between the multi-feature data set matched in space and time and the satellite-borne lidar data, and then determine the LAI estimation result.

[0115] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store large-area medium spatial resolution time series vegetation leaf area index estimation data that cooperates with satellite-borne laser radar and multispectral remote sensing. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a large-area medium spatial resolution time series vegetation leaf area index estimation method that cooperates with satellite-borne laser radar and multispectral remote sensing is implemented.

[0116] Those skilled in the art will understand that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0117] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0118] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0119] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization of the corresponding device owner. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0120] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0121] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0122] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for estimating leaf area index of vegetation in a large area with medium spatial resolution in a time series by using satellite-borne laser radar and multispectral remote sensing, characterized in that: The method comprises: Obtain multispectral reflectance data and satellite-borne lidar data for the area to be measured during the period to be measured; the multispectral reflectance data is obtained using Sentinel-2, based on the high spatiotemporal resolution characteristics of a 20-meter spatial resolution and a 5-day revisit cycle; the satellite-borne lidar data is based on the characteristics of a 4-day orbital revisit cycle, and all GEDI data for the Global Ecosystem Dynamics Survey are obtained; constructing a multi-feature data set based on the multispectral reflectance data; Matching the multi-feature data set with the satellite-borne laser radar data in space and time to obtain matching data; The matching data and the satellite-borne laser radar data are input into an estimation model to obtain an LAI estimation result; the estimation model is trained using a deep neural network; the estimation model is used to determine a mapping relationship between a multi-feature data set matched in space and time and the satellite-borne laser radar data, and then determine the LAI estimation result.

2. The method for estimating leaf area index of vegetation in a large area with medium spatial resolution in a time series by cooperating with spaceborne laser radar and multispectral remote sensing according to claim 1 is characterized in that: Constructing a multi-feature data set based on the multispectral reflectance data specifically includes: Based on the multispectral reflectance data, effective pixel extraction processing is performed to obtain effective pixel data; wherein, cloud shadow scenes are extracted with SCL=3, and cloud scenes are extracted with SCL=10; effective pixel data are extracted with MSK_CLDPRB less than 5%, MSK_SNWPRB less than 5% and non-cloud shadow scenes and non-cloud scenes; SCL is the scene classification band; MSK_CLDPRB is the cloud probability; MSK_SNWPRB is the snow probability; The multi-feature data set is determined according to the effective pixel data.

3. The method for estimating leaf area index of vegetation in a large area with medium spatial resolution in a time series by cooperating with spaceborne laser radar and multispectral remote sensing according to claim 2 is characterized in that: The multi-feature data set includes: reflectivity, multiple vegetation indices, solar zenith angle, and geospatial information; The reflectivity includes visible light, near infrared and short-wave infrared; the geospatial information is used to measure spatial proximity; The geospatial information includes: Among them, Spatial is the geographic spatial information; R is a constant, set to 1; θ is the longitude of the valid pixel data; is the latitude of the valid pixel data.

4. The method for estimating leaf area index of vegetation in a large area with medium spatial resolution in a time series by cooperating with spaceborne laser radar and multispectral remote sensing according to claim 2 is characterized in that: The multi-feature data set and the satellite-borne laser radar data are spatially and temporally matched to obtain matching data, specifically including: Based on the set screening rules, the GEDI data is screened to obtain the screened GEDI LAI spots; For each valid pixel data in the multi-feature data set, the GEDI LAI light spot is searched for by valid pixel data one by one, based on the interval time before and after the observation date of the valid pixel data at the same position within the set time period, so as to time match the multi-feature data set and the GEDI LAI light spot, and obtain the searched GEDI LAI light spot; Centered on the pixel where the searched GEDI LAI spot is located, a search is performed in the multi-feature dataset based on a 3×3 pixel block to spatially match the multi-feature dataset and the GEDI LAI spot, and obtain near-synchronous observation samples of the multi-feature set based on pixel blocks; The searched GEDI LAI spots and multi-feature set near-synchronous observation samples are identified as matching data.

5. The method for estimating leaf area index of vegetation in a large area with medium spatial resolution in a time series by cooperating with spaceborne laser radar and multispectral remote sensing according to claim 2, characterized in that: The method for determining the estimation model specifically includes: Build deep neural networks; Inputting historical matching data, historical satellite-borne lidar data and corresponding LAI results into a deep neural network, training the network parameters of the deep neural network with the goal of minimizing the mean square error between the output of the deep neural network and the LAI result to obtain a trained deep neural network; the LAI result is the leaf area index of the GEDI L2B product in the GEDI data; The trained deep neural network is determined as the estimation model.

6. The method for estimating leaf area index of vegetation in a large area with medium spatial resolution in time series by cooperating with spaceborne laser radar and multispectral remote sensing according to claim 1, characterized in that: The deep neural network adopts an encoding-decoding architecture; the encoding part is composed of a series of Residuals with convolution kernels of 64, 128, and 256 respectively; the decoding part is composed of a series of Residuals with convolution kernels of 128 and 64 respectively.

7. The method for estimating leaf area index of vegetation in a large area with medium spatial resolution in time series by cooperating with spaceborne laser radar and multispectral remote sensing according to claim 1, characterized in that: The mathematical expression for training the network parameters is: Among them, Φ is the network parameter; N is the total number of samples in the nearly synchronous observation samples in the multi-feature dataset; ω i is the weight of the i-th sample; is the output value corresponding to the i-th sample in the deep neural network; is the LAI result corresponding to the i-th sample.

8. A device for estimating leaf area index of vegetation in a large area with medium spatial resolution in a time series by cooperating with space-borne laser radar and multispectral remote sensing, characterized in that: The device comprises: The data acquisition module is used to obtain multispectral reflectance data and satellite-borne laser radar data of the area to be measured during the period to be measured; the multispectral reflectance data is obtained by using Sentinel-2, based on the high spatiotemporal resolution characteristics of 20-meter spatial resolution and 5-day revisit cycle; the satellite-borne laser radar data is all GEDI data of the Global Ecosystem Dynamics Survey obtained based on the characteristics of the 4-day orbital revisit cycle; A data set construction module, used to construct a multi-feature data set based on the multispectral reflectance data; A matching module, used for matching the multi-feature data set with the space-borne laser radar data in space and time to obtain matching data; An estimation module is used to input the matching data and the satellite-borne laser radar data into an estimation model to obtain an LAI estimation result; the estimation model is trained using a deep neural network; the estimation model is used to determine a mapping relationship between a multi-feature data set matched in space and time and the satellite-borne laser radar data, thereby determining the LAI estimation result.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the large-area medium spatial resolution time series vegetation leaf area index estimation method of coordinated satellite-borne laser radar and multispectral remote sensing as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for estimating the time-series vegetation leaf area index in a large area with medium spatial resolution by coordinating satellite-borne laser radar and multispectral remote sensing as described in any one of claims 1 to 6 is implemented.

Images