Arid region cultivated land soil organic matter inversion method based on wavelet transform fused spectral data
Through wavelet transformation, the complexity and uncertainty of data fusion in soil organic matter inversion is solved, high-precision monitoring of soil organic matter, and the soil organic matter in arid areas is improved. Inversion accuracy and stability.
Patent Information
- Application Number
- CN202410894943.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2024-11-15
AI Technical Summary
The existing technology is difficult to effectively combine hyperspectral and multispectral data in soil organic matter inversion, resulting in data fusion complexity, spatial consistency problems and computational complexity. The uncertainty and error caused by a single data source are relatively large, making it difficult to Achieve high-precision monitoring in large areas.
The wavelet transformation method is used to fuse hyperspectral and multispectral data, and combine environmental variables, and use integrated learning algorithms to establish a soil organic matter prediction model. The data consistency and resolution are improved through resampling and standardized processing, and the critical environmental data are selected to reduce uncertainty. .
It improves the accuracy and stability of soil organic matter inversion, enhances the overall quality and resolution of data, and significantly improves the reliability and accuracy of soil organic matter monitoring in arid areas.
Smart Images

Figure SMS_1 
Figure SMS_2 
Figure SMS_3
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of remote sensing data processing, in particular to an inversion method for soil organic matter in cultivated land in arid areas based on wavelet transform and fusion spectral data. Background Art
[0002] Soil salinization is the main factor affecting crop yields in arid areas. The assessment and monitoring of organic matter in salinized soil is of great significance for evaluating the fertility and health of cultivated land in the region. Remote sensing technology has been successfully applied to soil organic matter monitoring. The use of field hyperspectral data is a reliable method for monitoring soil salinity, but this method is not suitable for large-scale monitoring. Multispectral remote sensing data, characterized by its spectral and spatial information, largely makes up for the shortcomings of inversion using hyperspectral data. Existing technologies usually use these two data sources alone for inversion, which makes it difficult to give full play to their respective advantages. The soil formation process is complex, and it is difficult to establish the relationship between field-measured organic matter and noise-free remote sensing data when inverting soil organic matter. It is relatively rare to use a combination of multispectral data and hyperspectral data to invert soil organic matter at a regional scale.
[0003] Remote sensing data fusion technology has the advantage of being able to integrate multi-source remote sensing data, but there are still the following issues that need to be considered when applying remote sensing data fusion technology: (1) Complexity of spectral characteristics: The reflectance spectrum of soil organic matter at different wavelengths has complex characteristics, and its spectral characteristics may be confused by other soil components (such as minerals, water content, salt, etc.), making it difficult to accurately separate organic matter signals in actual measurements. The spectral resolution and number of bands of multispectral data and hyperspectral data are significantly different. Multispectral data usually has fewer bands but high spatial resolution, while hyperspectral data has rich bands but relatively low spatial resolution. This difference in resolution and number of bands increases the complexity of data fusion. (2) Spatial consistency: Multispectral data usually has high spatial resolution, while hyperspectral data may have low spatial resolution. This inconsistency in resolution makes it difficult to directly fuse data. (3) Computational complexity and resource requirements: The high-dimensional characteristics and multi-band information of hyperspectral data increase the computational complexity of data processing and analysis.
[0004] The soil formation process is complex, and there is a great deal of uncertainty in predicting soil properties based solely on spectral data. Therefore, environmental data (climate, topography, texture, etc.) are often used together with remote sensing data to invert soil organic matter, and the selection of which environmental data to use together requires screening. Moreover, the resolution of environmental data from different sources is inconsistent. Effectively integrating multiple data sources into a unified analysis framework requires complex data processing techniques to ensure data heterogeneity and comparability. Furthermore, establishing a model that considers the effects of multiple components requires comprehensive consideration of various environmental factors and parameters, which requires complex model design and parameter optimization.
[0005] Therefore, it is necessary to propose a method that combines field-measured hyperspectral and multispectral images and combines them with environmental variables to jointly invert soil organic matter, so as to effectively solve the problems of spatial coverage limitations, resolution differences and environmental complexity brought by a single data source, thereby achieving high-precision inversion of soil organic matter over a large area. Summary of the invention
[0006] The purpose of the present invention is to provide a method for inverting soil organic matter in cultivated land in arid areas based on wavelet transform fusion spectral data to improve the comprehensive utilization efficiency of data. The method can improve the accuracy of soil organic matter monitoring in cultivated land in arid areas.
[0007] The technical solution of the present invention is:
[0008] A method for inverting soil organic matter in arid cultivated land based on wavelet transform fusion spectral data, specifically comprising the following steps:
[0009] Step (1), obtaining soil samples, hyperspectral data and multispectral data of cultivated land in arid areas, wherein the multispectral data is Sentinel 2 image data;
[0010] Step (2), preprocessing the hyperspectral data, first smoothing and denoising the hyperspectral data, and then resampling the hyperspectral data to obtain resampled hyperspectral data consistent with the band range of Sentinel 2 image data;
[0011] Step (3), fusing the resampled hyperspectral data and the corresponding multispectral data based on wavelet transform to obtain fused spectral data, the specific steps include: reading the resampled hyperspectral data and Sentinel 2 image data from an excel file, and performing standardization on the read data; performing wavelet decomposition on the standardized hyperspectral data and Sentinel 2 image data to obtain their respective low-frequency coefficients and high-frequency coefficients; fusing the low-frequency coefficients and the high-frequency coefficients to obtain fused wavelet coefficients; performing inverse wavelet transform on the fused wavelet coefficients to obtain fused spectral data, and performing de-standardization on the fused wavelet coefficients; and saving the de-standardized data in an excel file;
[0012] Step (4), based on the resampled hyperspectral data, Sentinel 2 image data and the fused spectral data of the two, with bands and bands + environmental variables as independent variables and soil organic matter as dependent variable, an ensemble learning algorithm is used to establish a soil organic matter prediction model for predicting soil organic matter in arid areas;
[0013] Step (5), evaluating the soil organic matter prediction model established in step (4).
[0014] Wherein, obtaining soil samples from cultivated land in arid areas in step (1) includes: collecting soil samples from cultivated land in Yinchuan Plain, Ningxia Hui Autonomous Region, with a total of 231 effective sampling points and a sampling depth of 0 to 20 cm, and the collected soil samples are dried and sieved before being used for soil organic matter determination;
[0015] Acquiring the hyperspectral data in step (1) includes: using ASD's FieldSpec 4 ground feature spectrometer to collect field measured hyperspectral data of the sampling point, performing whiteboard correction before collection, repeating the spectrum of each sampling point for 5 times, and taking the average as the hyperspectral data of the sampling point;
[0016] Acquiring multispectral data in step (1) includes: acquiring Sentinel 2 multispectral image data in the same or similar time period as the hyperspectral sampling time, performing atmospheric correction on the image data, and resampling and file format conversion in combination with SNAP software, and using ArcGIS to extract the reflectance of each band corresponding to the sampling point.
[0017] The preprocessing of the hyperspectral data in step (2) includes: removing the 350-399 nm band and the 2401-2500 nm band with excessive noise; performing smoothing and denoising preprocessing on the data in the 400-2400 nm band using the Savitaky-Golay method; performing scale conversion on the preprocessed data using the resampling method, and resampling the spectrum of the 400-2400 nm band to a range consistent with the Sentinel 2 image band data through the image sensor spectral response function.
[0018] Wherein, in step (3), the resampled hyperspectral band data corresponding to the Sentinel 2 image data band is fused with the Sentinel 2 image band data based on wavelet transform, wherein the Sentinel 2 image band data are B2, B3, B4, B8, B11 and B12; db1 is used when the standardized hyperspectral data and Sentinel 2 image data are respectively decomposed by wavelet, and the number of decomposition layers is 1; the low-frequency coefficients and high-frequency coefficients are respectively fused using the maximum coefficient method.
[0019] Wherein, in step (4), the environmental variables include soil moisture content (SMC) at the sampling point, average annual temperature (MAT), average rainfall (MAP), silt (Silt (%)), clay (Clay (%)), sand (Sand (%)), soil depth to bedrock (Soil depth to bedrock), elevation (DEM), aspect (Aspect), slope (Slope), plane curvature (Plane curvature), profile curvature (Profile curvature) and topographic wetness index (Topographic wetness index, TWI).
[0020] The evaluation parameters of the soil organic matter prediction model include: determination coefficient (R 2 ), root mean square error (RMSE) and mean square error (MSE).
[0021] Among them, the ensemble learning algorithms include: Random forest (RF) model, Extremely randomized trees (ERT) model, Adaptive Boosting (Adaboost) model and Extreme gradient boosting (XGBoost) model. After evaluation in step (5), the RF model based on fused spectral data has the highest accuracy.
[0022] The beneficial effects of the present invention are as follows:
[0023] The present invention converts the resampled hyperspectral data and Sentinel2 image data into the wavelet domain by combining the wavelet transform method, which can help identify important features and patterns in the data. The wavelet transform decomposes the data into low-frequency and high-frequency components, making these components easier to analyze and fuse in the new representation space. The method realizes the efficient fusion of hyperspectral data and Sentinel 2 image data, which not only retains the rich spectral information of hyperspectral data, but also improves the spatial resolution through fusion, thereby improving the overall quality and resolution of the data.
[0024] The method of the present invention resamples the hyperspectral data based on the image spectral response function to make its band range consistent with the image, effectively solving the spatial consistency problem of hyperspectral data and multispectral data.
[0025] The method of the present invention selects soil moisture, which has an important influence on soil organic matter and the corresponding spectral reflectance, from a variety of environmental data, and simultaneously considers the influence of soil texture and soil-to-bedrock depth on soil organic matter. By integrating these multi-source data, the uncertainty and error caused by dependence on a single data source are reduced, and the stability and reliability of soil organic matter inversion in cultivated land in arid areas are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 It is a wavelet fusion flow chart of a method for inverting soil organic matter in cultivated land in arid areas based on wavelet transform fusion of spectral data according to an embodiment of the present invention.
[0027] Figure 2 It is a flow chart of model establishment of a method for inverting soil organic matter in arid cultivated land based on wavelet transform fusion of spectral data according to an embodiment of the present invention. DETAILED DESCRIPTION
[0029] The present invention is further described below in conjunction with specific embodiments:
[0030] like Figure 1 and Figure 2 As shown, the present invention proposes a method for inverting soil organic matter in arid cultivated land by fusion spectral data based on wavelet transform, comprising the following steps:
[0031] Step (1), obtaining soil samples, hyperspectral data and multispectral data of cultivated land in arid areas, wherein the multispectral data is Sentinel 2 image data.
[0032] Soil sample collection: Soil samples were collected from cultivated land in Yinchuan Plain, Ningxia Hui Autonomous Region, with a total of 231 effective sampling points and a sampling depth of 0 to 20 cm. The collected soil samples were dried and sieved before being used for soil organic matter determination. The determination method was potassium dichromate oxidation external heating method.
[0033] Hyperspectral data collection: FieldSpec 4 ground feature spectrometer from ASD was used to collect field measured hyperspectral data of the sampling points, i.e., hyperspectral reflectance. Whiteboard calibration was performed before collection, and the spectrum of each sampling point was measured 5 times, and the average value was taken as the spectral reflectance of the sampling point.
[0034] Multispectral data acquisition: Acquire Sentinel2 multispectral images at the same or similar time as the hyperspectral acquisition time. Sentinel 2, also known as Sentinel 2 satellite, is a high-resolution multispectral imaging satellite. The revisit period of one satellite is 10 days, and the revisit period of two complementary satellites (2A and 2B) is 5 days. The Sentinel2 satellite carries a multispectral instrument (MSI) that can cover 13 spectral bands with ground resolutions of 10m, 20m and 60m respectively. Among them, the blue (B2), green (B3), red (B4) and near-infrared (B8) bands have a resolution of 10m; the ground sampling distance of short-wave infrared SWIR (B11 and B12) is 20m. Sentinel 2 mainly monitors the earth's surface and provides related telemetry services. The multispectral imager (MSI) it carries can provide detailed images of vegetation, soil and water cover, inland waterways and coastal areas. All Sentinel 2 images were declouded using quality control bands. The data had undergone geometric and orthorectification. In order to convert the atmospheric apparent reflectance into surface reflectance, the ESA plug-in Sen2cor was used to process the Sentinel-2B data and perform atmospheric correction. The SNAP software was used for resampling and file format conversion, and the reflectance corresponding to the sampling points was extracted in ArcGIS.
[0035] Step (2) preprocesses the hyperspectral data by first smoothing and denoising the hyperspectral data, and then resampling the hyperspectral data to obtain resampled hyperspectral data that is consistent with the Sentinel 2 band range.
[0036] Specifically, first, the hyperspectral data collected in step (1) was exported using ViewSpec Pro software to remove the edge bands with excessive noise, including the 350-399nm band and the 2401-2500nm band; the data in the 400-2400nm band was smoothed and denoised using the Savitaky-Golay method. Then, the measured hyperspectral data was scaled using the resampling method: the spectral resampling tool in ENVI software was used to establish a spectral library with the measured hyperspectral data as the benchmark, and the spectrum of the spectrometer 400-2400nm band was resampled to a range consistent with the Sentinel 2 image band using the image sensor spectral response function.
[0037] Step (3), fusing the resampled hyperspectral data and the corresponding multispectral data based on wavelet transform to obtain fused spectral data.
[0038] Specifically, step (3) includes the steps of: reading resampled hyperspectral data and Sentinel2 image data from an excel file, standardizing the read data to eliminate the scale differences between different data sources and ensure that the data are compared and analyzed at the same scale; performing wavelet decomposition on the standardized hyperspectral data and Sentinel 2 image data to obtain their respective low-frequency coefficients and high-frequency coefficients; fusing the low-frequency coefficients and high-frequency coefficients to obtain fused wavelet coefficients; performing inverse wavelet transform using the fused wavelet coefficients to obtain fused spectral data, and performing de-standardization on the fused data; and saving the fused data as an excel file. The resampled hyperspectral band data corresponding to the Sentinel 2 band and the Sentinel 2 image band data (B2, B3, B4, B8, B11 and B12) are fused based on wavelet transform, wherein the wavelet decomposition uses db1, the decomposition layer is 1, and the fusion is performed using the maximum coefficient method.
[0039] The maximum coefficient method formula is: C fused (i,j)=max(C1(i,j),C2(i,j));
[0040] In the formula, C fused (i,j) represents the element in the i-th row and j-th column of the fused coefficient matrix; C1(i,j) represents the element in the i-th row and j-th column of the first set of wavelet decomposition coefficient matrices; C2(i,j) represents the element in the i-th row and j-th column of the second set of wavelet decomposition coefficient matrices; max(a,b) represents the larger value of a and b.
[0041] The following is a code example for step (3): #Data Standardization scaler = StandardScaler() hyperspectral_scaled=scaler.fit_transform(hyperspectral_values) Sentinel2_scaled=scaler.fit_transform(Sentinel2_values) #wavelet decomposition def wavelet_decompose(data,wavelet='db1',level=1): coeffs=pywt.wavedec2(data,wavelet,level=level) return coeffs hyperspectral_coeffs=wavelet_decompose(hyperspectral_scaled) Sentinel2_coeffs=wavelet_decompose(Sentinel2_scaled) #Fusion based on the maximum coefficient method def max_selection_fusion(coeffs1,coeffs2): fused_coeffs = [] for c1,c2 in zip(coeffs1,coeffs2): if isinstance(c1,tuple): fused_tuple=tuple(np.maximum(c1_i,c2_i)for c1_i,c2_iin zip(c1,c2)) fused_coeffs.append(fused_tuple) else: fused_coeffs.append(np.maximum(c1,c2)) return fused_coeffs fused_coeffs=max_selection_fusion(hyperspectral_coeffs,Sentinel2_coeffs) # Wavelet reconstruction and denormalization def wavelet_reconstruct(coeffs,wavelet='db1'): return pywt.waverec2(coeffs,wavelet) fused_data=wavelet_reconstruct(fused_coeffs) fused_data_unscaled=scaler.inverse_transform(fused_data) #Save the fused data as an excel file fused_data_df=pd.DataFrame(fused_data_unscaled,columns=hyperspectral_data.columns) fused_file_path='fused_data_wavelet.xlsx' fused_data_df.to_excel(fused_file_path,index=False) fused_file_path
[0042] Step (4) is to establish a soil organic matter prediction model based on the measured resampled hyperspectral data, Sentinel 2 image data and the fused spectral data of the two, with bands and bands + environmental variables as independent variables and soil organic matter as the dependent variable, using an ensemble learning algorithm (Ensemble Learning) to predict soil organic matter in arid areas.
[0043] Ensemble learning is a machine learning technique that improves prediction performance by combining multiple basic models. The ensemble learning algorithm used in the present invention integrates bagging (guided aggregation algorithm) and boosting (boosting method), and the specific algorithm models include random forest (RF) model, extremely randomized trees (ERT) model, adaptive boosting (Adaboost) model and extreme gradient boosting (XGBoost) model.
[0044] The environmental variables include soil moisture content (SMC), mean annual temperature (MAT), average rainfall (MAP), sand (%), silt (%), clay (%), soil depth to bedrock, elevation (DEM), aspect, slope, plane curvature, profile curvature and topographic wetness index (TWI).
[0045] SMC is the measured soil moisture content at the sampling point; MAT and MAP are from the "China Surface Climate Data Daily Dataset (V3.0)" of the National Meteorological Science Data Center (http: / / data.cma.cn / data / cdcdetail / dataCode / SURF_CLI_CHN_MUL_html). The terrain data (Digital Elevation Model, DEM) was downloaded from http: / / search.asf.alaska.edu / # / . The data set is ALOS PALSAR, and its spatial resolution is 12.5m. In ArcGIS, slope, aspect, plan curvature, profile curvature, and terrain moisture index were calculated based on DEM; soil texture data (including sand, silt, and clay) were obtained from the Resource and Environmental Science Data Center of the Chinese Academy of Sciences (http: / / www.resdc.cn) with a resolution of 1 km; soil to bedrock depth data were obtained from the literature: Yan, FP, et al., Depth-to-bedrock map of China at a spatial resolution of 100 meters, Earth System Science Data Discussions (2020), https: / / doi.org / 10.5194 / essd-2018-103. All the above environmental data were processed to a resolution of 10 m for model establishment.
[0046] Step (5), evaluating the soil organic matter prediction model established in step (4).
[0047] The model building process is run using Python programs, and the grid search method is used for hyperparameter tuning. Only the main parameters are grid searched, and the rest of the parameters use the default values of the Scikit-learn toolkit.
[0048] In the embodiment of the present invention, the selection ranges of each model parameter are: RF:'n_estimators':range(2,50,1),'max_features':(2,4,6,8),'max_depth':range(2,15,2); ERT:'n_estimators':range(1,30,1),'max_depth':range(2,15,1); Adaboost:'n_estimators':range(1,30,1),'learning_rate':np.linspace(0.01,2,20); XGBoost:'n_estimators':range(2,20,1),'learning_rate':np.linspace(0.01,2,20),'max_depth':range(2,15,1),'min_child_weight':range(2,15,1).
[0049] The model was evaluated by 10-fold cross validation, and the determination coefficient (R 2 ), root mean square error (RMSE) and mean square error (MSE).
[0050]
[0051]
[0052]
[0053] In the above formula: x and are the measured value and the average of the measured values, y and are the predicted value and the average of the predicted values respectively. i is the model prediction value; is the model measured value; is the average value predicted by the model; n is the number of samples.
[0054] The experimental results show that after adding environmental variables, the accuracy of all models has been improved to varying degrees. As can be seen from Table 1, when the band is used as the independent variable, the hyperspectral data performs best among all models, indicating that its high resolution and spectral information richness have advantages in modeling. The effect of Sentinel 2 data is generally not as good as that of hyperspectral data, but the accuracy of the model can be improved by fusion. As can be seen from Table 2, when the band + environmental variables are used as independent variables, the RF model of fused data performs best (R2 = 0.96, RMSE = 0.80, MSE = 0.64), indicating that the performance of the model can be significantly improved by combining band data and environmental variables. After the introduction of environmental variables, the performance of Sentinel 2 data is also greatly improved (R2 = 0.95, RMSE = 0.86, MSE = 0.75), showing the importance of environmental variables in improving model accuracy.
[0055] Considering that fused spectral data can provide better overall performance, especially when combined with environmental variables, the RF model performs best on fused spectral data. Therefore, in the fusion of measured resampled hyperspectral data and Sentinel-2 image data, the use of random forest (RF) model and fused spectral data is the best choice. This method can effectively combine the advantages of the two types of data and provide higher prediction accuracy and stability.
[0056] Table 1 Soil organic matter inversion model with band as independent variable
[0057] Table 2 Soil organic matter inversion model with band + environmental variables as independent variables
[0058] The present invention provides a method for estimating soil organic matter by fusing measured hyperspectral data and Sentinel 2 image data. Through the above method, the advantages of hyperspectral data and Sentinel 2 image data can be fully utilized to improve the comprehensive utilization efficiency of data and the prediction accuracy of the model, providing an effective solution for remote sensing data processing and providing effective technical support for soil organic matter inversion.
Claims
1. A method for inverting soil organic matter in arid areas based on wavelet transform fusion spectral data, characterized in that: The method specifically comprises the following steps: Step (1), obtaining soil samples, hyperspectral data and multispectral data of cultivated land in arid areas, wherein the multispectral data is Sentinel 2 image data; Step (2), preprocessing the hyperspectral data, first smoothing and denoising the hyperspectral data, and then resampling the hyperspectral data to obtain resampled hyperspectral data consistent with the band range of Sentinel 2 image data; Step (3), fusing the resampled hyperspectral data and the corresponding multispectral data based on wavelet transform to obtain fused spectral data, the specific steps include: reading the resampled hyperspectral data and Sentinel 2 image data from an excel file, and performing standardization on the read data; performing wavelet decomposition on the standardized hyperspectral data and Sentinel 2 image data to obtain their respective low-frequency coefficients and high-frequency coefficients; fusing the low-frequency coefficients and the high-frequency coefficients to obtain fused wavelet coefficients; performing inverse wavelet transform on the fused wavelet coefficients to obtain fused spectral data, and performing de-standardization on the fused wavelet coefficients; and saving the de-standardized data in an excel file; Step (4), based on the resampled hyperspectral data, Sentinel 2 image data and the fused spectral data of the two, with bands and bands + environmental variables as independent variables and soil organic matter as dependent variable, an ensemble learning algorithm is used to establish a soil organic matter prediction model for predicting soil organic matter in arid areas; Step (5), evaluating the soil organic matter prediction model established in step (4).
2. The method according to claim 1, characterized in that In step (1), the step of obtaining soil samples from cultivated land in arid areas comprises: collecting soil samples from cultivated land in Yinchuan Plain, Ningxia Hui Autonomous Region, with a total of 231 effective sampling points and a sampling depth of 0 to 20 cm. The collected soil samples are dried and sieved before use for soil organic matter determination; the step of obtaining hyperspectral data comprises: using ASD's FieldSpec 4 ground feature spectrometer to collect field measured hyperspectral data of the sampling points, performing whiteboard correction before collection, repeating the spectrum of each sampling point for 5 times, and taking the average as the hyperspectral data of the sampling point; the step of obtaining multispectral data comprises: obtaining Sentinel 2 multispectral image data within the same or similar time period as the hyperspectral sampling time, performing atmospheric correction on the image data, and resampling and file format conversion in combination with SNAP software, and extracting the reflectance of each band corresponding to the sampling point using ArcGIS.
3. The method according to claim 1, characterized in that In step (2), the preprocessing of the hyperspectral data includes: removing the 350-399nm band and the 2401-2500nm band with excessive noise; using the Savitaky-Golay method to perform smoothing and denoising preprocessing on the data in the 400-2400nm band; using the resampling method to scale the preprocessed data, and resampling the spectrum of the 400-2400nm band to a range consistent with the Sentinel 2 image band data through the image sensor spectral response function.
4. The method according to claim 1, characterized in that: In step (3), the resampled hyperspectral band data corresponding to the Sentinel 2 image data band is fused with the Sentinel 2 image band data based on wavelet transform, wherein the Sentinel 2 image band data are B2, B3, B4, B8, B11 and B12; db1 is used when the standardized hyperspectral data and Sentinel 2 image data are respectively decomposed by wavelet, and the number of decomposition layers is 1; the low-frequency coefficients and high-frequency coefficients are respectively fused using the maximum coefficient method.
5. The method according to claim 1, characterized in that The environmental variables described in step (4) include soil moisture content (SMC), mean annual temperature (MAT), mean rainfall (MAP), silt (%), clay (%), sand (%), soil depth to bedrock, elevation (DEM), aspect, slope, plane curvature, profile curvature and topographic wetness index (TWI) at the sampling point.
6. The method according to claim 1, characterized in that The evaluation parameters of the soil organic matter prediction model include: determination coefficient (R 2 ), root mean square error (RMSE) and mean square error (MSE).
7. The method according to claim 1, characterized in that The ensemble learning algorithm includes: Random forest (RF) model, Extremely randomized trees (ERT) model, Adaptive Boosting (Adaboost) model and Extreme gradient boosting (XGBoost) model. After evaluation in step (5), the RF model based on fused spectral data has the highest accuracy.
Citation Information
Cited By
Soil organic carbon on-line detection system based on spectral analysis
CN119779986A
Data analysis method for monitoring soil pollution spatial distribution based on image fusion
CN120847375A
