Remote sensing estimation and mapping method and device for grassland above-ground biomass

Abstract

The invention belongs to the technical field of remote sensing, and provides a remote sensing estimation and mapping method and device for grassland aboveground biomass. The method comprises the following steps: S1, acquiring grassland biomass measured data, meteorological data, optical remote sensing data, Sentinel-1SAR data and topographic data of a sample in a research area; making a grassland mask pattern; s2.1, extracting feature variables related to grassland biomass, including meteorological data feature variables, optical data feature variables, radar data feature variables and topographic data feature variables, and constructing a model data set; s2.2, acquiring raster data of the research area, and constructing a cartographic data set in combination with the grassland mask map; s3, constructing a stacked integrated learning model based on the constructed model data set; and S4, estimating and drawing the grassland aboveground biomass according to the drawing data set and the stacked ensemble learning model. According to the method, the grassland biomass estimation precision is improved, the data acquisition speed is increased, and the method is more suitable for large-scale application.

Description

Technical Field

[0001] The invention relates to the field of remote sensing technology, and in particular to a method and a device for remote sensing estimation and mapping of grassland aboveground biomass. Background Art

[0002] As one of the largest ecosystems in the world, grassland ecosystems occupy a pivotal position in terrestrial ecosystems. They not only play an irreplaceable role in maintaining ecological balance, producing forage crops and providing carbon sinks, but their above-ground biomass (AGB) is also an important indicator for measuring the productivity and carbon storage of grassland ecosystems. However, accurate assessment of the total biomass of grasslands and its changes are crucial for a deep understanding of the biogeochemical processes of terrestrial ecosystems and ensuring the long-term sustainable use of grassland resources.

[0003] Although traditional field survey methods are known for their accuracy in obtaining vegetation data, their time-consuming, high-cost and destructive characteristics have seriously hindered their widespread application in large-scale regional monitoring. To overcome this limitation, remote sensing monitoring technology has become the first choice for large-scale monitoring due to its macroscopic vision and comprehensive performance. Compared with traditional field measurements, remote sensing technology provides a new, efficient and non-destructive method for estimating grassland biomass (especially biomass at the regional scale).

[0004] Remote sensing optical imaging technology plays an important role in large-scale vegetation monitoring, but it also has disadvantages such as susceptibility to weather influences, conflicts between imaging period and resolution, limitations on surface coverage and accuracy, and complexity and high cost of data processing. These disadvantages limit the application of remote sensing optical imaging under certain specific conditions.

[0005] Although remote sensing monitoring technology has made significant progress, how to apply it to the estimation of grassland AGB to obtain more accurate, efficient and reproducible grassland aboveground biomass inversion mapping methods remains an urgent problem to be solved. Summary of the invention

[0006] In view of this, an embodiment of the present invention provides a method and device for remote sensing estimation and mapping of aboveground biomass of grassland, so as to solve or partially solve the above-mentioned problems.

[0007] In a first aspect, an embodiment of the present invention provides a method for remote sensing estimation and mapping of aboveground biomass of grassland, comprising the following steps:

[0008] S1: Obtain the measured data of grassland biomass, meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data in the study area; obtain the land cover classification of the study area and make a grassland mask map;

[0009] S2: Construct a data set based on the data obtained in step S1, including:

[0010] S2.1: extracting characteristic variables related to grassland biomass from the meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data, including meteorological data characteristic variables, optical data characteristic variables, radar data characteristic variables, and terrain data characteristic variables; constructing a model data set containing a plurality of the characteristic variables according to the sample point coordinates of the sample grassland biomass measured data;

[0011] S2.2: Obtain raster data of the study area, including optical raster data, radar raster data, terrain raster data and meteorological raster data, and construct a mapping data set in combination with the grassland mask map obtained in step S1;

[0012] S3: A stacked ensemble learning model constructed based on the constructed model dataset;

[0013] S4: Estimating and mapping grassland aboveground biomass based on the mapping dataset and the stacked ensemble learning model.

[0014] According to a specific implementation of the embodiment of the present invention, in the step S1, the meteorological data needs to be subjected to meteorological interpolation, data splicing and clipping processing.

[0015] According to a specific implementation method of an embodiment of the present invention, in step S1, the optical remote sensing data is generated by merging Landsat8 and Landsat9 images after cloud removal, band information is extracted from the optical remote sensing data, and the vegetation index is constructed using the band information.

[0016] According to a specific implementation manner of the embodiment of the present invention, in step S1, radar backscatter data is extracted from the Sentinel-1SAR data, and a radar vegetation index is constructed according to the radar backscatter coefficient.

[0017] According to a specific implementation of an embodiment of the present invention, the preparation of the grassland mask map includes resampling the resolution of the land cover classification result map to match the resolution of Landsat 8 and Landsat 9 images.

[0018] According to a specific implementation of the embodiment of the present invention, step S3 is specifically:

[0019] S3.1: Divide the model data set into test data and training data;

[0020] S3.2: Build multiple machine learning models;

[0021] S3.3: Construct the stacked ensemble learning model based on multiple machine learning models.

[0022] According to a specific implementation method of an embodiment of the present invention, the stacked ensemble learning model is a hierarchical model integration framework; the first layer is composed of multiple different base learners, the second layer is a prediction model and a linear regression algorithm is selected as a meta-learner.

[0023] In a second aspect, an embodiment of the present invention provides a remote sensing estimation and mapping device for grassland aboveground biomass, comprising:

[0024] Data acquisition module, used to obtain the measured data of grassland biomass, meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data in the study area; obtain the land cover classification of the study area and make a grassland mask map;

[0025] The data set construction module is used to construct the data set, including: extracting characteristic variables related to grassland biomass from the meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data, including meteorological data characteristic variables, optical data characteristic variables, radar data characteristic variables, and terrain data characteristic variables; constructing a model data set containing multiple characteristic variables according to the sample point coordinates of the sample grassland biomass measured data; obtaining the raster data of the study area, including optical raster data, radar raster data, terrain raster data, and meteorological raster data, and constructing a mapping data set in combination with the grassland mask map;

[0026] A model building module, used to build a stacked ensemble learning model based on the constructed model data set;

[0027] A mapping module is used to estimate and map the aboveground biomass of grassland based on the mapping data set and the stacked ensemble learning model.

[0028] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the remote sensing estimation and mapping method in the aforementioned first aspect or any implementation method of the first aspect are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0030] Figure 1A flow chart showing a method for remote sensing estimation and mapping of grassland aboveground biomass provided by an embodiment of the present invention is shown;

[0031] Figure 2 A schematic diagram of a stacking ensemble learning model training and verification process in an embodiment of the present invention is shown;

[0032] Figure 3 The grassland coverage map of West Ujimqin Banner in the embodiment of the present invention is shown;

[0033] Figure 4 The biomass map of grassland in West Ujimqin Banner in the embodiment of the present invention is shown;

[0034] Figure 5 A structural block diagram of a remote sensing estimation and mapping device for grassland aboveground biomass provided by an embodiment of the present invention is shown;

[0035] Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0036] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0037] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0038] Figure 1 A method for remote sensing estimation and mapping of grassland aboveground biomass provided by an embodiment of the present invention, a flowchart of the steps, see Figure 1 , the method comprises the following steps:

[0039] S1: Data preparation: Obtain the measured data of grassland biomass, meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data in the study area; obtain the land cover classification of the study area and make a grassland mask map; specifically:

[0040] S1.1: Obtain the measured data of grassland biomass in the study area;

[0041] In order to obtain the real grassland biomass data in the study area, the following detailed field survey steps are required:

[0042] S1.1.1: Plot selection and sampling:

[0043] First, with the sampling point location as the core, typical plots with uniform species distribution, flat terrain and sufficient area were selected. Then, key information such as latitude and longitude, altitude and main vegetation types of each plot were recorded in detail to ensure the representativeness of the plot and data traceability. Three 1m×1m plots were evenly arranged in the selected plots, and the above-ground vegetation in the plots was mowed to remove impurities such as soil and stones to ensure the purity of the biomass, and the fresh weight was recorded by weighing.

[0044] S1.1.2: Biomass drying weighing and data recording;

[0045] The collected vegetation biomass is placed in an oven and undergoes a drying process until it reaches a constant weight. This step is intended to eliminate the influence of moisture and thus obtain dry weight data of the biomass.

[0046] The average value of the aboveground biomass (AGB) of the grassland in each plot was obtained by taking the arithmetic mean of the dry weight of three or five plots in each plot. This average value effectively reflects the overall status of the grassland biomass in the plot.

[0047] S1.2: Obtain meteorological data for the study area;

[0048] In order to obtain the meteorological data (covering temperature and precipitation) of the study area, preprocessing work was carried out, including meteorological interpolation, data splicing and clipping steps, and finally the meteorological data were integrated into a unified format for each month of the year.

[0049] S1.3: Obtain optical remote sensing data of the study area;

[0050] The optical images of large study areas are mainly affected by weather (clouds and rain), resulting in missing images in some areas after cloud removal. In order to obtain optical remote sensing data of the study area within the effective time, this embodiment uses the Google EarthEngine platform to collect and simply pre-process (atmospheric correction and radiation calibration) multi-band data (including visible and near-infrared, short-wave infrared, and thermal infrared bands) of Landsat8 and Landsat9 satellites. Through time, regional screening and cloud mask filtering of the quality assessment band (QA), the median image is obtained to enhance data stability, and finally merged to generate high-quality remote sensing image data, effectively overcoming the problem of missing data caused by long cycles and weather influences. The details are as follows:

[0051] The band information is extracted from the acquired image data to calculate various vegetation indices.

[0052] The vegetation index (as shown in Table 2 below) is constructed using the band information (as shown in Table 1 below).

[0053]

[0054] Further explanation is as follows: NIR: near infrared band; SWIR1: shortwave infrared 1band; SWIR2: shortwave infrared 2band; ST_10: Band 10surface temperature.

[0055]

[0056]

[0057] S1.4: Obtain Sentinel-1SAR data for the study area;

[0058] Sentinel-1SAR data obtained through preprocessing (geometric correction, radiation correction, despeckling, etc.) can more accurately reflect the surface characteristics.

[0059] The radar backscatter data is extracted based on the acquired image data and the radar vegetation index is calculated.

[0060] The vegetation index (as shown in Table 4 below) is constructed using the backscatter coefficient (as shown in Table 3 below).

[0061]

[0062]

[0063]

[0064] S1.5: Obtain topographic data of the study area;

[0065] Terrain data includes (elevation, slope, aspect). Elevation data comes from COPERNICUS / DEM / GLO30, with a spatial resolution of 30m, provided by the cloud computing platform GEE. Slope and aspect data are calculated and generated from elevation data.

[0066] S1.6: Obtain real-time land cover classification of the study area and produce grassland mask map;

[0067] This embodiment uses the Dynamic World (DW) dataset, a 10-meter resolution global land use and land cover (LULC) dataset generated based on Sentinel-2 images and deep learning technology, with the help of Google Earth Engine and the cloud AI platform. In order to generate a near-real-time land cover classification result map for the study area, a time range of half a month before and after the field survey was determined, and GEE was used to perform pattern synthesis on the highest probability labels in the DW dataset to create a monthly classification product. Subsequently, the resolution of the classification result map was resampled from 10 meters to 30 meters to match the resolution of Landsat8 and Landsat 9 images, and the monthly land cover classification result map of the study area was produced accordingly. Finally, the grassland mask was extracted based on the classification results, providing key information for subsequent grassland analysis and remote sensing data interpretation.

[0068] S2: Build the dataset:

[0069] Construct a data set based on the data obtained in step S1, including:

[0070] S2.1: Extract characteristic variables related to grassland biomass from meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data, including meteorological data characteristic variables, optical data characteristic variables, radar data characteristic variables, and terrain data characteristic variables; construct a model data set containing multiple characteristic variables based on the sample point coordinates of the sample grassland biomass measured data;

[0071] Extract characteristic variables related to grassland biomass, including optical data characteristic variables (such as vegetation index and band), radar data characteristic variables (such as backscatter coefficient and constructed index), terrain data characteristic variables (such as slope and aspect), and meteorological data (such as temperature and rainfall). Ensure that all characteristic variables are consistent with the actual values ​​of grassland biomass in time and space.

[0072] S2.2: Obtain raster data of the study area, including optical raster data, radar raster data, terrain raster data and meteorological raster data, and construct a mapping dataset in combination with the grassland mask map obtained in step S1.

[0073] S3: Based on the constructed model dataset, a stacked ensemble learning model is constructed, specifically:

[0074] S3.1: Divide the model dataset into test and training data;

[0075] Generally, the training data is 70% of the total data, and the test data is 30% of the total data.

[0076] S3.2: Build multiple machine learning models;

[0077] In this embodiment, multiple machine learning models such as extreme random trees (ERT), gradient boosted decision trees (GBDT), random forest regression (RF), and extreme gradient boosting (XGBoost) are constructed.

[0078] A machine learning model was built using Python language, and grid search and five-fold cross validation were used to optimize the hyperparameters of the algorithm used.

[0079] S3.3: Construct a stacked ensemble learning model based on multiple machine learning models;

[0080] For high-precision models, a new stacking ensemble learning model is constructed; Stacking is a hierarchical model integration framework, the first layer is composed of multiple different base learners, in this embodiment, ERT, RF, GBDT and XGBoost are selected, and in the actual construction process, models with good experimental results can also be selected for construction. Under the premise that the effects of each model are relatively good, they are integrated for prediction to reduce the deviation of the model and obtain more ideal results; in the second-layer prediction model, in order to avoid overfitting, the linear regression (LR) algorithm with simple structure and strong generalization ability is selected as the meta-learner. Figure 2 As shown, the specific steps are as follows:

[0081] S3.3.1: Training set division: Based on the principle of cross-validation, the original training set was randomly divided into five equal and non-overlapping sub-datasets.

[0082] S3.3.2: Base learner training: For each base learner, four of the five sub-datasets are randomly selected as the training set for this round, and the remaining one is used as the test set. This step follows the five-fold cross-validation method, that is, each base learner will be trained and tested five times, each time using a different test set. All test results are averaged on the original data test set as the stacking transformation of the training data for this base learner. Repeat this process to complete the first-layer model training of all base learners.

[0083] S3.3.3: Meta-learner training: Use the AGB prediction values ​​generated by the first-layer base learner to construct a new training set. This new training set is used to train the prediction model of the second layer, namely the meta-learner. Through this process, the final biomass prediction value can be obtained.

[0084] S3.4: Model optimization;

[0085] Variables can significantly affect the efficiency and accuracy of the model. To improve computational efficiency, the most informative AGB estimation features were selected for analysis by analyzing the importance of features, and low-importance features were removed to reduce the dimensionality. Among them, the AGB estimation features are variable features constructed for estimating biomass.

[0086] S4: Estimation and mapping of grassland aboveground biomass based on mapping datasets and stacked ensemble learning models.

[0087] The embodiments of the present invention have the following technical effects:

[0088] First, remote sensing optical imaging technology is of great value in large-scale vegetation monitoring, but it faces weather interference and limitations in imaging period and resolution. Combining Landsat8 and Landsat9 images can alleviate these problems, and the Dynamic World dataset on the Google EarthEngine platform provides near-real-time land cover classification, which helps to accurately draw grassland cover maps. In addition, the use of ensemble learning methods can more effectively integrate the advantages of each base learner and improve model performance. Finally, the data and the constructed model are used to map the aboveground biomass of the grassland.

[0089] Second, the present invention proposes a more accurate, efficient and reproducible method for inversion and mapping of grassland aboveground biomass. This method not only improves the accuracy of grassland biomass estimation through high-resolution remote sensing data, but also speeds up data acquisition, making it more suitable for large-scale applications. At the same time, the estimation model is more sophisticated, using a variety of data to comprehensively consider a variety of factors, providing an important reference for accurate assessment of grassland yield, carbon storage and other research.

[0090] The following takes the estimation of aboveground biomass of the grassland in West Ujimqin Banner as an example to further explain the remote sensing estimation and mapping methods of aboveground biomass of this grassland:

[0091] S1: Data preparation:

[0092] S1.1: Obtain the real grassland biomass data of the sample area in Xiwuzhumuqin Banner (abbreviated as Xiwu Banner);

[0093] S1.2: Obtain meteorological data of West Ujimqin Banner;

[0094] Obtain the meteorological station data of each township in Xiwuqi through the Xiwuzhumuqin Banner Meteorological Station (the data is usually in csv or other text format). The meteorological station data usually includes the station location (latitude and longitude), observation time and observation value (monthly average temperature, monthly average precipitation data).

[0095] Use the python library to read the data and ensure that the data format is correct and there are no missing values. Use ArcGIS software to import the preprocessed data for interpolation (common interpolation methods include inverse distance weighted interpolation (IDW), Kriging interpolation (Kriging), radial basis function interpolation (RBF), etc.) and draw it into a TIF format image (when exporting TIF files, ensure that the correct projection and coordinate system information is set). Convert discrete station data into spatially continuous data to generate spatially continuous meteorological data for the study area.

[0096] S1.3: Obtain optical remote sensing data of West Ujimqin Banner;

[0097] Follow the method in section 1.3 of the method steps to obtain the multi-band merged image of Xiwuqi.

[0098] The band information is extracted from the acquired image data to calculate various vegetation indices.

[0099] The vegetation index is calculated using the vegetation index formula involved in the method steps.

[0100] You can use professional remote sensing software (such as ArcGIS, ENVI, etc.) to calculate the vegetation index, such as using the BandMath tool in ENVI to calculate the vegetation index. The steps to calculate the ratio vegetation index (RVI) are as follows: First, select the "Band Math" tool under "Basic Tools" from the main menu. Then, enter the band operation formula in the "Enter an expression" text box. For example, if the near-infrared band is b4 and the red band is b3, the RVI formula is "float(b4) / float(b3)". Since most remote sensing image data is byte type, the float() function is required to convert the data type to ensure the accuracy of the calculation. Then, match the band variable in the formula with the actual data in the dialog box. Finally, click the "OK" button, ENVI will perform the operation and generate an output image of the RVI value.

[0101] You can write band calculation code in GEE to directly calculate the vegetation index.

[0102] The vegetation index can be calculated using Python and other program scripts to generate vegetation index images.

[0103] S1.4: Acquire Sentinel-1SAR data of West Ujimqin Banner;

[0104] The method step S1.4 can be used to obtain the same radar vegetation index, wherein the same radar vegetation index is calculated as above.

[0105] S1.5: Obtain the topographic data of West Ujimchin Banner;

[0106] The terrain data includes (elevation, slope, and aspect). The elevation data comes from COPERNICUS / DEM / GLO30, with a spatial resolution of 30m, and is provided by the cloud computing platform GEE.

[0107] Slope and aspect are usually calculated using geographic information system (GIS) software or professional remote sensing image processing software, such as ArcGIS, ENVI, etc., combined with elevation data (digital elevation model DEM). By calculating the ratio of the elevation change to the horizontal distance of each pixel point and using mathematical methods such as inverse trigonometric functions, the values ​​of slope and aspect can be accurately obtained.

[0108] S1.6: Obtain real-time land cover classification of the study area and produce grassland mask map;

[0109] The grassland mask can be obtained by using the method step S1.6. Figure 3 shown.

[0110] S2: Build the dataset:

[0111] S2.1: Build model dataset:

[0112] According to the coordinates of the sample points where grassland biomass was measured, a data set containing a variety of relevant characteristic variables was constructed.

[0113] S2.2: Constructing mapping dataset:

[0114] Get the raster data of the area, including optical raster data, radar raster data, terrain raster data and meteorological raster data.

[0115] S3: Build the model:

[0116] S3.1: Divide the data set constructed based on S2.1 into training and test data (generally the training data is 70% of the total data, and the test data is generally 30% of the total data).

[0117] S3.2: Build a machine learning model to build a new stacking ensemble learning model for the high-precision model;

[0118] Use python code to build four high-precision models: Extreme Random Tree (ERT), Gradient Boosting Decision Tree (GBDT), Random Forest Regression (RF) and Extreme Gradient Boosting (XGBoost) to establish a stacking ensemble learning model.

[0119] A machine learning model was built using Python language, and grid search and five-fold cross validation were used to optimize the hyperparameters of the algorithm used.

[0120] S3.3: Grassland biomass model optimization;

[0121] Variables can significantly affect the efficiency and accuracy of the model. Recursive Feature Elimination (RFE) is an efficient feature selection method that aims to screen out the feature subset that is most relevant to model performance through an iterative process. This method is based on feature sorting technology and achieves reverse sequence reduction by gradually eliminating the features that are considered the least relevant in each step until the optimal feature subset is obtained.

[0122] To improve computational efficiency, the most informative AGB estimation features were selected for analysis by analyzing the importance of features, and low-importance features were removed to reduce the dimensionality.

[0123] The root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R-square, R 2 ) as the prediction index. Among them, RMSE and MAE measure the global and local absolute errors of the prediction results. The smaller the values ​​of the two, the higher the prediction accuracy of the model; R 2 It measures the overall prediction effect. The larger the value, the better the model fitting effect. The specific calculation is shown in formulas (1) to (3).

[0124]

[0125] Where:y i is the measured grassland AGB of the i-th sampling point; is the estimated grassland AGB of the i-th sample point; is the average value of the measured grassland AGB; n is the number of sample points.

[0126] According to the above feature selection method, the model can be optimized by comprehensively considering the characteristics of each type of variable and the evaluation index. In the feature selection method, it is also possible to screen according to the correlation analysis of the specific data corresponding to the biomass. Correlation analysis: By calculating the correlation coefficient between each feature and the biomass (such as the Pearson correlation coefficient), the direct correlation between each feature and the biomass can be evaluated.

[0127] S4: Mapping:

[0128] The mapping dataset and constructed model were used to estimate and map grassland aboveground biomass.

[0129] Ensure that the data resolution of each variable (raster format) selected by the feature is consistent, and use Python code to input it into the optimized model to estimate the biomass of the grassland, and finally generate a complete grassland biomass map of Xiwu Banner. Import the generated grassland biomass map into ArcGIS software, and also import the generated grassland mask. Use the mask extraction in the software to extract only the grassland part of the grassland biomass map of the entire Xiwu Banner, and finally complete the mapping, as shown in the figure below: Figure 4 shown.

[0130] The embodiments of the present invention have the following technical effects:

[0131] 1. Data preparation and preprocessing techniques:

[0132] Acquire and integrate real grassland biomass data, meteorological data, optical remote sensing data, Sentinel-1SAR data and terrain data of Xiwuzhumuqin Banner.

[0133] Use Python libraries to read and process the data, ensuring that the data is in the correct format and has no missing values.

[0134] ArcGIS or similar software is used to perform data interpolation to convert discrete site data into spatially continuous data.

[0135] Extract band information from remote sensing images and calculate various vegetation indices, such as the Ratio Vegetation Index (RVI).

[0136] 2. Feature selection and data set construction:

[0137] According to the coordinates of the sample points where grassland biomass was measured, a data set containing a variety of relevant characteristic variables was constructed.

[0138] Use geographic information system (GIS) or remote sensing image processing software to calculate terrain data (such as elevation, slope, and aspect).

[0139] Create a grass mask map for extracting grass parts in the subsequent mapping process.

[0140] 3. Machine learning model construction and optimization:

[0141] Divide the data set into training and testing, usually the training data accounts for 70% of the total data and the testing data accounts for 30%.

[0142] Build stacking ensemble learning models using high-precision models such as Extreme Random Trees (ERT), Gradient Boosted Decision Trees (GBDT), Random Forest Regression (RF), and Extreme Gradient Boosting (XGBoost).

[0143] Grid search and five-fold cross validation are used to optimize hyperparameters and improve model performance.

[0144] Recursive feature elimination (RFE) is used for feature selection to remove low-importance features and reduce the model dimension.

[0145] The model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R 2 ) as a predictive indicator.

[0146] 4. Grassland biomass mapping technology:

[0147] Mapping of grassland aboveground biomass was performed using mapping data and the optimized model.

[0148] Make sure all input data is of consistent resolution and feed it into the model using Python code.

[0149] Estimate grassland biomass and generate a complete grassland biomass map of West Ujimqin Banner.

[0150] ArcGIS software was used to perform mask extraction and only the biomass map of the grassland part was extracted.

[0151] Figure 5 A structural block diagram of a remote sensing estimation and mapping device for grassland aboveground biomass provided by an embodiment of the present invention, the device comprising:

[0152] Data acquisition module, used to obtain the measured data of grassland biomass, meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data in the study area; obtain the land cover classification of the study area and make a grassland mask map;

[0153] The data set construction module is used to construct the data set, including: extracting characteristic variables related to grassland biomass from meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data, including meteorological data characteristic variables, optical data characteristic variables, radar data characteristic variables, and terrain data characteristic variables; constructing a model data set containing multiple characteristic variables based on the sample point coordinates of the sample grassland biomass measured data; obtaining the raster data of the study area, including optical raster data, radar raster data, terrain raster data, and meteorological raster data, and combining the grassland mask map to construct a mapping data set;

[0154] A model building module is used to build a stacked ensemble learning model based on the built model dataset;

[0155] A mapping module for estimating and mapping grassland aboveground biomass based on mapping datasets and stacked ensemble learning models.

[0156] Figure 5The functions of each module in the embodiment correspond to the contents in the corresponding method embodiment and will not be repeated here.

[0157] Figure 6 The schematic diagram of the structure of the electronic device 60 provided in the embodiment of the present invention is shown, and the electronic device 60 includes at least one processor 601 (such as a CPU), at least one input and output interface 604, a memory 602, and at least one communication bus 603, which are used to realize the connection and communication between these components. At least one processor 601 is used to execute the computer instructions stored in the memory 602, so that the at least one processor 601 can execute any of the embodiments of the aforementioned remote sensing estimation and mapping methods. The memory 602 is a non-transitory memory, which may include a volatile memory, such as a high-speed random access memory (RAM: Random Access Memory), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection with at least one other device or unit is realized through at least one input and output interface 604 (which may be a wired or wireless communication interface).

[0158] In some implementations, the memory 602 stores a program 6021, and the processor 601 executes the program 6021 to execute the content of any of the aforementioned table partitioning method embodiments.

[0159] The electronic device may exist in a variety of forms, including but not limited to:

[0160] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0161] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0162] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0163] (4) Specific server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0164] (5) Other electronic devices with data interaction functions.

[0165] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for remote sensing estimation and mapping of grassland aboveground biomass, characterized in that: The following steps are involved: S1: Obtain the measured data of grassland biomass, meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data in the study area; Obtain land cover classification of the study area and prepare grassland mask map; S2: Construct a data set based on the data obtained in step S1, including: S2.1: extracting characteristic variables related to grassland biomass from the meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data, including meteorological data characteristic variables, optical data characteristic variables, radar data characteristic variables, and terrain data characteristic variables; constructing a model data set containing a plurality of the characteristic variables according to the sample point coordinates of the sample grassland biomass measured data; S2.2: Obtain raster data of the study area, including optical raster data, radar raster data, terrain raster data and meteorological raster data, and construct a mapping data set in combination with the grassland mask map obtained in step S1; S3: A stacked ensemble learning model constructed based on the constructed model dataset; S4: Estimating and mapping grassland aboveground biomass based on the mapping dataset and the stacked ensemble learning model.

2. The remote sensing estimation and mapping method according to claim 1, characterized in that: In the step S1, the meteorological data needs to be processed by meteorological interpolation, data splicing and clipping.

3. The remote sensing estimation and mapping method according to claim 1, characterized in that: In the step S1, the optical remote sensing data is generated by merging Landsat8 and Landsat9 images after cloud removal, band information is extracted from the optical remote sensing data, and the vegetation index is constructed using the band information.

4. The remote sensing estimation and mapping method according to claim 3, characterized in that: The band information is shown in Table 1 below: The vegetation index is shown in Table 2 below: 。 5. The remote sensing estimation and mapping method according to claim 1, characterized in that: In the step S1, radar backscatter data is extracted from the Sentinel-1SAR data, and a radar vegetation index is constructed according to the radar backscatter coefficient.

6. The remote sensing estimation and mapping method according to claim 5, characterized in that: The radar backscatter coefficients are shown in Table 3 below: The radar vegetation index is shown in Table 4 below:

7. The remote sensing estimation and mapping method according to claim 1, characterized in that: The step S3 is specifically as follows: S3.1: Divide the model data set into test data and training data; S3.2: Build multiple machine learning models; S3.3: Construct the stacked ensemble learning model based on multiple machine learning models.

8. The remote sensing estimation and mapping method according to claim 7, characterized in that: The stacked ensemble learning model is a hierarchical model integration framework; the first layer is composed of multiple different base learners, the second layer is a prediction model and a linear regression algorithm is selected as a meta-learner.

9. A remote sensing estimation and mapping device for grassland aboveground biomass, characterized in that: The remote sensing estimation and mapping method according to any one of claims 1 to 8 comprises: Data acquisition module, used to obtain the measured data of grassland biomass, meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data in the study area; obtain the land cover classification of the study area and make a grassland mask map; The data set construction module is used to construct the data set, including: extracting characteristic variables related to grassland biomass from the meteorological data, optical remote sensing data, Sentinel-1SAR data, and terrain data, including meteorological data characteristic variables, optical data characteristic variables, radar data characteristic variables, and terrain data characteristic variables; constructing a model data set containing multiple characteristic variables according to the sample point coordinates of the sample grassland biomass measured data; obtaining the raster data of the study area, including optical raster data, radar raster data, terrain raster data, and meteorological raster data, and constructing a mapping data set in combination with the grassland mask map; A model building module, used to build a stacked ensemble learning model based on the constructed model data set; A mapping module is used to estimate and map the aboveground biomass of grassland based on the mapping data set and the stacked ensemble learning model.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the remote sensing estimation and mapping method according to any one of claims 1 to 8 are implemented.

Images