A remote sensing estimation method for soil organic carbon storage in plateau swamp wetlands

Abstract

The present invention discloses a remote sensing estimation method for soil organic carbon reserves in plateau swamp wetlands, which belongs to the field of organic carbon measurement technology, including obtaining soil organic carbon density and spectral reflectance data of plateau swamp wetlands, obtaining optical, radar images and digital elevation models of plateau swamp wetlands as well as rainfall and temperature data, preprocessing the data, calculating features related to vegetation, water bodies and soil; extracting spectral features highly correlated with soil organic carbon density; predicting spectral features and screening, integrating remote sensing features with spectral features; optimizing features and quantities; establishing a random forest model based on the optimal features, making an organic carbon density distribution map, and realizing organic carbon reserve estimation. The present invention adopts the above method, which not only retains the high accuracy of predicting soil organic carbon density by measured spectra, but also realizes the prediction of large-scale soil organic carbon density and organic carbon reserves in combination with remote sensing technology, thereby improving work accuracy and work efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of organic carbon measurement, and in particular to a remote sensing estimation method for soil organic carbon reserves in plateau swamp wetlands. Background Art

[0002] As one of the three major ecosystems on Earth, wetlands play an important role in the global carbon cycle and carbon balance, and are an important component of the terrestrial carbon pool in the global carbon cycle. Swamp wetlands are one of the most important wetland types, with more prominent carbon sink capacity and carbon storage. Due to the large amount of surface water in swamp wetlands, the swamp wetland soil is in an oxygen-deficient environment for a long time, and the soil organic matter usually cannot be completely decomposed, so a large amount of organic carbon is fixed in the swamp wetland soil. Plateau swamp wetland plots are fragmented, with rich vegetation coverage, and the spatial heterogeneity of soil organic carbon distribution is stronger.

[0003] Traditional soil organic carbon density monitoring methods are mainly estimated through field surveys, sampling and laboratory measurements. Although this traditional method is accurate, it is costly and time-consuming, making it difficult to use on a large scale. Spatial autocorrelation-based interpolation mapping methods (such as kriging and inverse distance weighted methods) have also been used to monitor soil organic carbon density over large areas in the past, but due to the non-zonal distribution characteristics of wetlands, it is difficult to guarantee the reliability of carbon storage prediction. For large-scale marsh wetlands with inconvenient sampling, traditional large-scale data collection technology is unrealistic, and the accuracy of spatial autocorrelation interpolation mapping methods is difficult to guarantee.

[0004] Near-end spectral technology can establish a direct relationship between soil spectral data and soil component content within a specific spectral wavelength range (350-2500nm), and can achieve high-precision soil organic carbon density estimation. However, the high-precision soil organic carbon density prediction of near-end spectral technology is still reflected in the predicted point space, and it is impossible to achieve large-scale prediction. Remote sensing technology can achieve large-scale synchronous observation. In the past, most studies on wetland carbon storage were concentrated in areas with continuous wetland plots and coastal tidal flats with low vegetation coverage. In plateau areas with complex terrain and fragmented plots, the spatial heterogeneity of soil organic carbon density distribution is stronger, and the prediction effect of soil organic carbon storage will be greatly reduced.

[0005] Therefore, based on the existing soil organic carbon density estimation technology, how to retain the high accuracy of measured spectral prediction of soil organic carbon density and realize large-scale soil organic carbon density prediction with remote sensing technology has become an urgent problem to be solved by technicians in this field. Summary of the invention

[0006] The purpose of the present invention is to provide a remote sensing estimation method for plateau swamp wetland soil organic carbon storage, which realizes the estimation of plateau swamp wetland soil organic carbon density and soil organic carbon storage, and effectively improves work efficiency and estimation accuracy.

[0007] To achieve the above object, the present invention provides a remote sensing estimation method for soil organic carbon storage in plateau swamp wetlands, the steps comprising:

[0008] S1. Obtain the soil organic carbon density and spectral reflectance data of plateau swamp wetland soil;

[0009] S2, obtain optical and radar images and digital elevation model (DEM) of plateau marsh wetlands as well as rainfall and temperature data, pre-process the data, and calculate remote sensing features related to vegetation, water bodies, and soil;

[0010] S3, processing the spectral reflectance data in step S1, and extracting spectral features related to soil organic carbon density using Pearson correlation analysis;

[0011] S4, build an XGBoost model, use remote sensing features to predict spectral features, screen the spectral features, and fuse the screened spectral features with remote sensing features;

[0012] S5. Construct a recursive feature elimination random forest regression model to determine the optimal features and quantity, and quantitatively evaluate the effect of each feature on soil organic carbon prediction;

[0013] S6. Establish a random forest model based on the optimal features obtained in step S5, make an organic carbon density distribution map, and estimate the organic carbon reserves.

[0014] Preferably, in step S1, the soil organic carbon density data of the plateau swamp wetland is obtained based on ground measuring points of the plateau wetland, and the soil spectral reflectance data of the plateau swamp wetland is measured in a dark room.

[0015] Preferably, in step S2, Google Earth Engine is used to preprocess the data and calculate remote sensing features related to vegetation, water bodies and soil, and the remote sensing features include feature normalized vegetation index, ratio vegetation index, enhanced vegetation index, grayscale symbiosis matrix index, plant senescence reflectance index, anthocyanin reflectance index, red edge inflection point, inverted red edge chlorophyll index, normalized water index, differentiated water index, bare soil index, soil conditioned vegetation index, surface water index, slope, terrain relief, terrain humidity index, radar feature VV polarization and VH polarization.

[0016] Preferably, step S3 includes: first filtering, smoothing and denoising the spectral reflectance data, and then performing mathematical transformation, including reciprocal transformation 1 / R, logarithmic transformation Log R, derivative transformation R' and reciprocal post-logarithmic transformation Log(1 / R), and using Pearson correlation analysis to determine spectral features with a correlation greater than 0.65 with soil organic carbon density.

[0017] Preferably, the step S4 includes:

[0018] S41, constructing an XGBoost model with remote sensing features as input variables and spectral features as output variables, and calculating spectral features using the trained XGBoost algorithm;

[0019] S42, use determination coefficient R 2 Screening as a validation indicator, extracting the predicted R 2 Spectral features above 0.8;

[0020] S43. Fusion the screened spectral features with the remote sensing features.

[0021] Preferably, in step S5, eighty percent of the organic carbon density data of plateau swamp soil samples are used for modeling, and twenty percent of the data are used for ten-fold cross-validation to test the model effect.

[0022] Preferably, the formula for estimating the organic carbon reserves in step S6 is:

[0023] C i =SOCD i *A i

[0024]

[0025] Where, SOCD i is the carbon density value of the i-th pixel, A i is the area of ​​the i-th pixel, C i is the carbon storage of the ith pixel, n is the total number of pixels in the study area, C total is the total carbon stock.

[0026] Therefore, the present invention adopts the above-mentioned remote sensing estimation method of soil organic carbon storage in plateau swamp wetlands, which has the following beneficial effects:

[0027] (1) It can quickly select the features related to the soil organic carbon density of marsh wetlands, which not only retains the high accuracy of the measured spectrum in predicting the soil organic carbon density, but also combines remote sensing technology to realize the prediction of large-scale soil organic carbon density and organic carbon storage, thus improving the work accuracy and efficiency;

[0028] (2) When measuring plateau swamp wetlands, only a small number of point measurements are needed to quickly obtain the optimal model for estimating soil organic carbon stocks in plateau swamp wetlands, thereby achieving large-scale, periodic observations of plateau wetlands, saving material and financial resources, and having important significance for the intelligent monitoring and management of carbon sinks in plateau wetlands.

[0029] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention is a method flow chart of an embodiment of a method for remote sensing estimation of soil organic carbon storage in plateau swamp wetlands. DETAILED DESCRIPTION

[0031] Example

[0032] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0033] Reference Figure 1 , a remote sensing estimation method for soil organic carbon storage in plateau marsh wetlands, including:

[0034] S1. Obtain the soil organic carbon density and spectral reflectance data of plateau swamp wetland soil. The soil organic carbon density data of plateau swamp wetland soil is obtained based on the ground measurement points of plateau wetland, and the spectral reflectance data of plateau swamp wetland soil is measured in a dark room.

[0035] Specifically, for the plateau marsh wetland soil samples obtained at the ground measuring points, the soil samples were first air-dried naturally. Then the bulk density and gravel content of the soil were determined, and all non-soil materials (such as plant roots and stones) were removed. After that, the soil samples were ground and passed through a 0.25 mm coarse sieve and a 0.098 mm fine sieve in turn. Four parallel measurements were performed on each sample using a total organic carbon analyzer to determine the soil organic carbon content, and the average value was taken as the soil organic carbon content result, and the soil organic carbon density was calculated. The formula for calculating the soil organic carbon density is:

[0036] SOCD=c*d*r*(1-D)

[0037] Where SOCD represents soil organic carbon density (kg C / m 2 ), c represents soil organic carbon content (%), d represents soil layer thickness (m), and r represents soil bulk density (kg / m 3 ), D is the content of gravel with a diameter greater than 2 mm.

[0038] When measuring the spectral reflectance of plateau swamp wetland soil, a portable ground spectrometer was used in a dark room to measure the spectral reflectance of plateau swamp wetland soil samples in the wavelength range of 300 to 2500 nanometers.

[0039] S2. Obtain optical and radar images and digital elevation models (DEMs) of plateau marsh wetlands as well as rainfall and temperature data, preprocess the data to improve image quality, and calculate remote sensing features related to vegetation, water bodies, and soil; calculate remote sensing features related to vegetation, water bodies, and soil.

[0040] Specifically, we use the Google Earth Engine platform to obtain open Sentinel-1 (plateau wetland radar) data, perform radiation correction, speckle noise filtering and terrain correction on the data, then obtain Sentinel-2 (plateau wetland optical) data, use the s2cloudless algorithm to remove clouds, then perform atmospheric correction and geometric correction, obtain NASADEM data, and import local monthly average precipitation and temperature data.

[0041] Secondly, Google Earth Engine was used to calculate remote sensing features related to vegetation, water and soil. The remote sensing features include the normalized vegetation index NDVI, the ratio vegetation index RVI, the enhanced vegetation index EVI, the NDVI grayscale co-occurrence matrix index (texture features: contrast, dissimilarity, entropy, correlation, angular second moment, quality), the plant senescence reflectance index PSRI, the anthocyanin reflectance index ARI, the red edge inflection point REIP, the inverted red edge chlorophyll index IRECI, the normalized water index NDWI, the differentiated water index MNDWI, the bare soil index BSI, the soil conditioned vegetation index SAVI, the surface water index LSWI, the slope SLOPE, the terrain relief RELIEF, the terrain wetness index TWI, the radar features VV polarization and VH polarization. The remote sensing features and their descriptions are shown in Table 1.

[0042] Table 1 Remote sensing features and their descriptions

[0043]

[0044] S3. Process the spectral reflectance data in step S1 and extract spectral features related to soil organic carbon density using Pearson correlation analysis.

[0045] Specifically, the spectral reflectance data were first smoothed and denoised using Savitzky-Golay filtering, and then mathematical transformations were performed, including reciprocal transformation 1 / R, logarithmic transformation Log R, derivative transformation R', and reciprocal post-logarithmic transformation Log(1 / R). Pearson correlation analysis was used to determine the spectral features with a correlation greater than 0.65 with soil organic carbon density.

[0046] The list of spectral features is:

[0047] G1:608.867,G2:609.444,G3:610.02,G4:610.597,G5:611.174,G6:611.751,G7:612.327,G8:612.904,G9:613.481,G10:614.057,G11:614.634,G12 :615.21,G13:615.787,G14:616.363,G15:616.939,G16:617.516,G17:61 8.092,G18:618.668,G19:619.244,G20:619.821,G21:620.397,G22:620.9 73,G23:621.549,G24:622.125,G25:622.701,G26:623.277,G27:623.853 ,G28:624.429,G29:625.005,G30:625.58,G31:626.156,G32:626.732,G33 :627.307,G34:627.883,G35:628.459,G36:629.034,G37:629.61,G38:63 0.185,G39:630.761,G40:631.336,G41:631.911,G42:632.487,G43:633.0 62,G44:633.637,G45:634.213,G46:634.788,G47:635.363,G48:635.938 ,G49:636.513,G50:637.088,G51:637.663,G52:638.238,G53:638.813,G5 4:639.388,G55:639.963,G56:640.537,G57:641.112,G58:641.687,G59: 642.262,G60:642.836,G61:643.411,G62:643.985,G63:644.56,G64:645. 134,G65:645.709,G66:646.283,G67:646.858,G68:647.432,G69:648.006,G70:648.58,G71:649.155,G72:649.729,G73:650.303,G74:650.877,G7 5:651.451,G76:652.025,G77:652.599,G78:653.173,G79:653.747,G80:6 54.321,G81:654.895,G82:655.468,G83:656.042,G84:656.616,G85:657.189,G86:657.763,G87:658.336,G88:658.91,G89:659.484,G90:660.057,G91:660.63,G92:661.204,G93:661.777,G94:662. 351,G95:662.924,G96:663.497,G97:664.07,G98:664.643,G99:665.216,G100:665.789,G101:666.362,G102:666.935,G103: 667.508,G104:668.081,G105:668.654,G106:669.227,G107:669.8,G108:670.373,G109:670.945,G110:671.518,G111:672. 091,G112:672.663,G113:673.236,G114:673.808,G115:674.381,G116:674.953,G117:675.526,G118:676.098,G119:676.67, G120:677.242,G121:677.815,G122:678.387,G123:678.959,G124:679.531,G125:680.103,G126:680.675,G127:681.247,G1 28:681.819,G129:682.391,G130:682.963,G131:683.535,G132:684.107,G133:684.678,G134:685.25,G135:685.822,G136:6 86.393,G137:686.965,G138:687.536,G139:688.108,G140:688.68,G141:689.251,G142:689.822,G143:690.394,G144:690. 965,G145:691.536,G146:692.108,G147:692.679,G148:693.25,G149:693.821,G150:694.392,G151:694.963,G152:695.534. .

[0048] S4. Build an XGBoost model, use remote sensing features to predict spectral features, screen spectral features, and fuse the screened spectral features with remote sensing features. Specifically, the steps include:

[0049] S41, constructing an XGBoost model with remote sensing features as input variables and spectral features as output variables, and calculating spectral features using the trained XGBoost algorithm;

[0050] S42, use determination coefficient R 2 Screening as a validation indicator, extracting the predicted R 2 Spectral features above 0.8.

[0051] The list of spectral features above 0.8 is G3:610.02,G7:612.327,G11:614.634,G14:616.363,G17:618.092,G20:619.821,G23:621.549,G26:623.277,G29:625.005,G32:626.732,G35:628.459,G38:630.185,G41:631.911,G44:633.637,G47:635.363,G50:637.088,G53:638.813,G56:640 .537,G59:642.262,G62:643.985,G65:645.709,G68:647.432,G71:649.155,G74:650.877,G77:652.599,G80:654.321,G83:656.04 2,G86:657.763,G89:659.484,G92:661.204,G95:662.924,G98:664.643,G101:666.362,G104:668.081,G107:669.8,G110:671.518

[0052] S43. The screened spectral features are integrated with the remote sensing features to form a new feature space.

[0053] S5. Construct a recursive feature elimination random forest regression model to determine the optimal features and quantity, and quantitatively evaluate the effect of each feature on soil organic carbon prediction.

[0054] Specifically, eighty percent of the organic carbon density data of plateau swamp soil samples were used to build the model, and twenty percent of the data were used to test the model effect with ten-fold cross-validation. After determining the optimal features and quantity, the feature importance was used to quantitatively evaluate the effect of each feature on soil organic carbon prediction.

[0055] In the RFE process, one feature is removed at each iteration and the model is trained. Ten-fold cross validation is used to evaluate the performance of the model after each iteration. When the performance no longer improves (R 2The RFE process is stopped when the maximum value is reached and the RMSE is minimized, and the optimal number of features is achieved. RFE modeling is performed using Python language and scikit-learn library to select the best model parameters.

[0056] S6. Establish a random forest model based on the optimal features obtained in step S5, make an organic carbon density distribution map, and estimate the organic carbon reserves.

[0057] Specifically, the optimal feature was used as the input variable and the soil organic carbon density was used as the output variable to establish a random forest model. The trained random forest model was used to predict the soil organic carbon density in the study area. The organic carbon storage was estimated using the formula:

[0058] C i =SOCD i *A i

[0059]

[0060] Where, SOCD i is the carbon density value of the i-th pixel, A i is the area of ​​the i-th pixel, C i is the carbon storage of the ith pixel, n is the total number of pixels in the study area, C total is the total carbon stock.

[0061] Therefore, the present invention adopts the above-mentioned remote sensing estimation method for soil organic carbon reserves in plateau swamp wetlands, which not only retains the high accuracy of predicting soil organic carbon density by measured spectra, but also combines remote sensing technology to realize the prediction of large-scale soil organic carbon density and organic carbon reserves, thereby improving work accuracy and work efficiency.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A remote sensing estimation method for soil organic carbon storage in plateau marsh wetlands, characterized in that the steps include: S1. Obtain the soil organic carbon density and spectral reflectance data of plateau swamp wetland soil; S2, obtain optical and radar images and digital elevation model (DEM) of plateau marsh wetlands as well as rainfall and temperature data, pre-process the data, and calculate remote sensing features related to vegetation, water bodies, and soil; S3, processing the spectral reflectance data in step S1, and extracting spectral features related to soil organic carbon density using Pearson correlation analysis, specifically including: First, the spectral reflectance data were filtered, smoothed and denoised, and then mathematically transformed, including reciprocal transformation 1 / R, logarithmic transformation LogR, derivative transformation R' and reciprocal post-logarithmic transformation Log(1 / R). Pearson correlation analysis was used to determine the spectral features with a correlation greater than 0.65 with soil organic carbon density, where R represents soil spectral reflectance. S4. Construct an XGBoost model, use remote sensing features to predict spectral features, screen spectral features, and fuse the screened spectral features with remote sensing features, including: S41, constructing an XGBoost model with remote sensing features as input variables and spectral features as output variables, and calculating spectral features using the trained XGBoost algorithm; S42, use determination coefficient R 2 Screening as a validation indicator, extracting the predicted R 2 Spectral features above 0.8; S43, fusing the screened spectral features with the remote sensing features; S5. Construct a recursive feature elimination random forest regression model to determine the optimal features and quantity, and quantitatively evaluate the effect of each feature on soil organic carbon prediction; Eighty percent of the organic carbon density data of plateau swamp soil samples were used to build the model, and 20 percent of the data were used to test the model effect with ten-fold cross validation. After determining the optimal features and quantity, the feature importance was used to quantitatively evaluate the effect of each feature on soil organic carbon prediction. S6. Establishing a random forest model based on the optimal features obtained in step S5, making an organic carbon density distribution map, and estimating the organic carbon reserves; The formula for estimating organic carbon storage is: C i =SOCD i *WHOSE i In the formula, SOCD i is the carbon density value of the i-th pixel, A i is the area of ​​the i-th pixel, C i is the carbon storage of the ith pixel, n is the total number of pixels in the study area, C total is the total carbon stock.

2. The remote sensing estimation method for soil organic carbon storage in plateau marsh wetlands according to claim 1 is characterized by: In the step S1, the soil organic carbon density data of the plateau swamp wetland is obtained based on the ground measurement points of the plateau wetland, and the soil spectral reflectance data of the plateau swamp wetland is measured in a dark room.

3. The remote sensing estimation method for soil organic carbon storage in plateau marsh wetlands according to claim 2 is characterized by: In the step S2, Google Earth Engine is used to preprocess the data and calculate remote sensing features related to vegetation, water bodies and soil, wherein the remote sensing features include feature normalized vegetation index, ratio vegetation index, enhanced vegetation index, grayscale symbiosis matrix index, plant senescence reflectance index, anthocyanin reflectance index, red edge inflection point, inverted red edge chlorophyll index, normalized water index, differentiated water index, bare soil index, soil conditioned vegetation index, surface water index, slope, terrain relief, terrain humidity index, radar feature VV polarization and VH polarization.

Images