Desert reclamation area soil respiration rate inversion method based on random forest algorithm

Abstract

The invention discloses a desert reclamation area soil respiration rate inversion method based on a random forest algorithm, and the method comprises the steps: measuring the soil of a desert reclamation area, and obtaining actually-measured soil respiration data, UAV hyperspectrum, thermal infrared data and ASD ground canopy hyperspectral radiance data; a characteristic wave band is screened out through the UAV hyperspectrum and actually measured soil respiration data, and TVDI is calculated through the UAV hyperspectrum and used for representing environment factors; three methods of FLD, 3FLD and SFM are used, and three unmanned aerial vehicle SIF values are calculated by using UAV hyperspectral data; using ASD ground canopy hyperspectral radiance data to verify the unmanned aerial vehicle SIF values obtained by the three methods, and selecting unmanned aerial vehicle SIF data with the highest precision as input data of soil respiration inversion; and inverting the soil respiration rate of the desert reclamation area based on a random forest algorithm. According to the method, the soil respiration rate can be rapidly and quantitatively monitored in a large-area and real-time manner.

Description

Technical Field

[0001] The invention belongs to the technical field of soil respiration inversion methods, and in particular relates to a soil respiration rate inversion method in a desert reclamation area based on a random forest algorithm. Background Art

[0002] Soil respiration is the second largest flux in the ecosystem carbon cycle. Accurate inversion of soil respiration in desert reclamation areas is the basis for understanding the carbon cycle in desert reclamation areas, and is of great significance for assessing carbon sequestration in desert ecosystems and improving regional carbon cycle.

[0003] At present, the traditional soil respiration measurement method is the in-situ monitoring method, that is, using instruments on site to collect soil carbon flux data. This method is time-consuming, labor-intensive, costly, and has a small coverage area. Compared with traditional monitoring methods, unmanned aerial vehicle (UAV) remote sensing monitoring of soil respiration has the advantages of strong real-time performance, wide range, and low cost. Compared with satellite remote sensing, UAV can provide raw data with high spatial resolution and high spectral resolution. At the same time, combined with computer models and field measured data, remote sensing inversion results that are highly consistent with the actual situation can also be obtained.

[0004] According to existing data, soil respiration generally includes three organic biological processes (i.e., soil microbial heterotrophic respiration, soil animal respiration, and plant root autotrophic respiration) and one inorganic chemical oxidation process. In actual calculations, it is believed that soil respiration is approximately composed of plant root autotrophic respiration and soil microbial heterotrophic respiration. In desert areas, soil conditions are poor, and heterotrophic respiration is not strongly affected by soil temperature and humidity. Therefore, plant root autotrophic respiration becomes the main contribution to soil respiration in desert areas. In recent years, sunlight-induced chlorophyll fluorescence (SIF), as a remote sensing indicator directly related to plant photosynthetic activity, has been widely used in vegetation pest and disease monitoring, drought monitoring, and terrestrial carbon cycle. Compared with traditional vegetation spectral indices such as NDVI, RVI, EVI, etc., SIF, as a new remote sensing vegetation indicator, can quantitatively represent photosynthetic activity and can be used to quantitatively reflect various phenomena that affect vegetation life activities and the impact of vegetation photosynthetic activity on the environment. Summary of the invention

[0005] The purpose of the present invention is to provide a method for inverting soil respiration rate in desert reclamation areas based on a random forest algorithm, which can quickly and quantitatively monitor soil respiration rate over a large area and in real time.

[0006] The technical solution adopted by the present invention is a method for inverting soil respiration rate in desert reclamation areas based on a random forest algorithm, which is specifically implemented in the following steps: S1. Measure the soil in the desert reclamation area to obtain measured soil respiration data, UAV hyperspectral, thermal infrared data and ASD ground canopy hyperspectral radiance data; S2, using UAV hyperspectral and measured soil respiration data to screen out characteristic bands, and using UAV hyperspectral to calculate TVDI to characterize environmental factors; S3, using the three methods of FLD, 3FLD and SFM, the UAV hyperspectral data were used to calculate the three UAV SIF values; S4, using ASD ground canopy hyperspectral radiance data to verify the UAV SIF values ​​obtained by the three methods in step S3, and selecting the UAV SIF data with the highest accuracy as the input data for soil respiration inversion; S5. Inversion of soil respiration rate in desert reclamation areas based on random forest algorithm.

[0007] The present invention is also characterized in that: Step S2 is specifically as follows: S21, extracting the UAV hyperspectral curve at each RTK coordinate of the measured soil respiration sample point, using the Person correlation coefficient method to calculate the correlation between different bands and the measured soil respiration data, and selecting the four bands with the highest correlation coefficients as the characteristic bands for soil respiration inversion; S22. Calculate TVDI based on UAV hyperspectral reflectance images to characterize environmental factors.

[0008] In step S21, if the selected characteristic band is a continuous band, the characteristic bands falling in different color light intervals are averaged according to the color light interval of the visible spectrum as the final characteristic band for independent variable input of subsequent model construction.

[0009] In step S3, the FLD method calculates the UAV SIF using the following formula:

[0010] In the formula, l in It represents the wavelength of the band within the Fraunhofer line, and its value is 690.4nm; l out It represents the wavelength of the band within the Fraunhofer line, and its value is 686.1nm; L ( l ) indicates the wavelength l The apparent radiance of plants at E ( l ) indicates the wavelength l The solar irradiance at Indicated in wavelength l in The transmitted SIF at .

[0011] In step S3, the 3FLD method calculates the UAV SIF using the following formula:

[0012] In the formula, lin It represents the wavelength of the band within the Fraunhofer line, and its value is 690.4nm; left , right Represents the wavelengths of the bands on both sides of the Fraunhofer line, which are 686.1nm and 694.7nm respectively; ; ; L ( l ) indicates the wavelength l The apparent radiance of plants at E ( l ) indicates the wavelength l The solar irradiance at Indicated in wavelength lin The transmitted SIF at .

[0013] In step S3, the SFM method calculates fluorescence using the following formula:

[0014] In the formula, L ( l ) indicates the wavelength l The apparent radiance of plants at = , E ( l ) indicates the wavelength l Solar irradiance at a is the true reflectivity; b represents the calculated fluorescence; the inversion bands used in the calculation process are 686.1nm, 690.4nm and 694.7nm.

[0015] Step S4 is specifically as follows: Using the FLD, 3FLD and SFM methods, three ground SIF values ​​were calculated using the ASD ground canopy hyperspectral radiance data as ground verification data for the three UAV SIF values ​​obtained in step S3. Before verification, the UAV SIF calculated by the FLD and 3FLD methods was normalized, and the UAV SIF data with the highest fitting accuracy with the ground SIF was selected as the input data for soil respiration inversion.

[0016] Step S5 is specifically as follows: S51, constructing a random forest model, using SIF as an independent variable representing vegetation information during the modeling process, and then inputting the measured soil respiration data, thermal infrared images, characteristic bands and TVDI images obtained in the above steps into the constructed model to estimate the soil respiration rate and obtain a soil respiration spatial distribution map; S52, extracting the values ​​of the measured sample points on the soil respiration estimation image, comparing them with the in-situ measured values ​​of the soil respiration, and evaluating the accuracy of the model estimation.

[0017] The beneficial effects of the present invention are: (1) The method of the present invention uses UAV hyperspectral data to invert soil respiration rate, which not only reduces the cost of data collection, but also greatly increases the distribution range and real-time performance of measured data, providing a new method for ecological restoration in desert reclamation areas; (2) The method of the present invention introduces SIF as the vegetation representation factor in the remote sensing inversion of soil respiration, uses hyperspectral UAV image data to calculate SIF and other related indices in the study area, and combines it with the random forest algorithm to improve the estimation accuracy of soil respiration, thereby achieving the purpose of large-scale, real-time and rapid quantitative monitoring of soil respiration. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 The spatial distribution diagram of the ground surface temperature obtained in Example 1 of the present invention; Figure 3 The temperature value obtained in Example 1 of the present invention is a spatial distribution map of drought index; Figure 4 This is a spatial distribution diagram of the UVA inversion SIF results obtained using the FLD method in Example 1 of the present invention; Figure 5 This is a spatial distribution diagram of the UVA inversion SIF results obtained using the 3FLD method in Example 1 of the present invention; Figure 6 This is a spatial distribution diagram of the UVA inversion SIF results obtained using the SFM method in Example 1 of the present invention; Figure 7 This is a fitting accuracy diagram of the UAV SIF and the ground SIF obtained by using the FLD method in Example 1 of the present invention; Figure 8 This is a fitting accuracy diagram of the UAV SIF and the ground SIF obtained by using the 3FLD method in Example 1 of the present invention; Figure 9 This is a fitting accuracy diagram of the UAV SIF and the ground SIF obtained by using the SFM method in Example 1 of the present invention; Figure 10 The soil respiration space distribution map obtained in Example 1 of the present invention; Figure 11 This is a precision comparison chart for verifying the inversion results in Example 1 of the present invention. DETAILED DESCRIPTION

[0019] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] The present invention provides a method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm. Figure 1 As shown, the specific implementation steps are as follows: S1, data preparation and processing: The prepared data include: measured soil respiration data, obtained through the EGM-5 soil carbon flux measurement system; UAV hyperspectral data and thermal infrared data, obtained through preprocessing with Mega Cube and ENVI 5.3 software; ASD ground canopy hyperspectral radiance data, collected through the ground ASD ground feature spectrometer. The specific methods for obtaining each data are as follows: (1) Measured soil respiration data: Field measured soil respiration data, using EGM-5 portable infrared gas analyzer to measure the soil respiration rate of the surface soil in the study area, the measurement time is 10:00-15:00, the soil respiration ring is buried in the corresponding point 1-2 days in advance to reduce the error caused by the destruction of the soil structure caused by the buried respiration ring, and the position of the respiration ring will not change after it is buried. The instrument sampling time is 100s, and the measurement is repeated 3 times for each sampling point, and the average value is taken as the soil respiration rate of the sample point. The soil carbon flux measurement system composed of soil temperature sensor (STP-2), soil respiration chamber (SRC-2) and EGM-5 soil carbon flux host is used to collect soil respiration rate data, and the change of soil temperature at that time is measured. The soil moisture content at the sample point is measured using the British PR2 soil profile moisture rapid tester. The precise coordinates of the sampling point are obtained using GPS-RTK.

[0021] (2) UAV hyperspectral and thermal infrared data: Field UAV hyperspectral data collection was carried out under clear, cloudless or lightly clouded, windless weather conditions, and the collection time was 11:00-15:00. Before the flight of the drone, the hyperspectral sensor was calibrated for radiation using the matching calibration gray cloth, the imaging spectrometer field of view (FOV) was set to 25°, and the corresponding flight altitude, heading, and lateral overlap were set. Thermal infrared images were obtained by carrying a Zenmuse XT2 thermal infrared (7.5~13.5μm) sensor on the drone, and the resolution of the thermal infrared camera was 0.05℃.

[0022] The main steps of indoor drone data processing are: 1. Route segmentation; 2. Radiation calibration; 3. Image strip stitching and geo-registration; 4. Reconstruct spectral information and output results.

[0023] The output image contains 150 spectral bands, covering the spectral range of 400-1000nm, with a pixel size of 0.05×0.05m. A total of two images of radiance and reflectance and a thermal infrared image of the study area are output for subsequent data analysis.

[0024] (3) ASD ground canopy hyperspectral radiance data: The acquisition time of ASD ground canopy hyperspectral radiance data should not differ from that of UAV hyperspectral acquisition time by more than 30 minutes. During acquisition, the probe should be suspended about 50 cm above the canopy to ensure that there is no shadow in the acquisition area. Ten spectral curves are collected for each canopy, and the arithmetic mean is taken as the pure spectral curve of the canopy after jump point correction. The spectral coverage range of this type of spectrometer is from 350nm to 2500nm, and the fluorescence spectrum (670~850nm) is selected for analysis.

[0025] S2, characteristic band screening and spectral index calculation: S21, characteristic band screening: First, the RTK coordinates of the measured soil respiration sample points are imported into the computer, and the UAV hyperspectral curve at each coordinate is extracted. Then, the Person correlation coefficient method is used to calculate the correlation between different bands and the measured soil respiration data, and the four bands with the highest correlation coefficients are selected as the characteristic bands for soil respiration inversion. If the selected characteristic bands are continuous bands, the average value of the characteristic bands falling in different color light intervals is calculated according to the color light interval of the visible spectrum as the final characteristic band for the independent variable input of the subsequent model construction.

[0026] The Person correlation coefficient can reflect the closeness of the linear correlation between two variables. The random variables X and Y are composed of sets of n samples, X is , Y is … , the correlation coefficient calculation formula between X and Y is: (1) In the formula, and are the sample means of random variables X and Y. The closer the absolute value of the correlation coefficient between two variables is to 1, the stronger the linear correlation between the two variables is, and vice versa.

[0027] S22, spectral index is calculated based on UAV hyperspectral reflectance image to characterize environmental factors: In desert ecosystems, the soil is chronically short of water, and soil respiration will become more sensitive to soil moisture when the temperature is relatively high. Therefore, the method of the present invention uses the hyperspectral reflectance image of the UAV to calculate the normalized difference vegetation index (NDVI), and then uses the NDVI and thermal infrared remote sensing images to calculate the temperature value drought index (TVDI). TVDI is used to characterize environmental factors. Research has found that this spectral index can better represent the local soil moisture conditions.

[0028] S3, Calculation of UAV-SIF images The image SIF is calculated using UAV hyperspectral radiance images. The O2 absorption band (686nm~688nm) is selected as the inversion window of UAV-SIF, and the UAV SIF value is calculated using three methods: Fraunhofer dark line method (FLD), three-band Fraunhofer dark line method (3FLD) and spectrum fitting method (SFM).

[0029] S31, Fraunhofer dark line method (FLD) The FLD algorithm uses a band within the Fraunhofer lines ( l in ) and a band outside the Fraunhofer lines ( l out ) and put it into the fluorescence emission equation to obtain equation (2): (2) Assumptions , , it can be deduced from formula (2): (3) In the formula, L ( l ) indicates the wavelength l The apparent radiance of plants at r ( l ) indicates the wavelength l The true reflectivity at E ( l ) indicates the wavelength l The solar irradiance at Indicated in wavelength lin The transmitted SIF at .

[0030] In the method of the present invention, the band within the Fraunhofer line ( l in ) The wavelength is selected as 690.4nm, which is outside the Fraunhofer line ( l out ) The wavelength is selected as 686.1nm.

[0031] S32, three-band Fraunhofer dark line method (3FLD) 3FLD is an improvement on FLD. 3FLD needs to use a band within the Fraunhofer line and two bands on both sides of the Fraunhofer line, and introduces two parameters related to the wavelengths on both sides. and . By derivation, we can get: (4) (5) Substituting (4) and (5) into (3) yields: (6) In the formula, L ( l ) indicates the wavelength l The apparent radiance of plants at r ( l ) indicates the wavelength l The true reflectivity at E ( l ) indicates the wavelength l The solar irradiance at Indicated in wavelength l in The transmitted SIF at .

[0032] In the method of the present invention, the band within the Fraunhofer line ( l in ) The wavelength is selected as 690.4nm, which is outside the Fraunhofer line ( l left and l right ) The wavelengths are selected as 686.1nm and 694.7nm respectively.

[0033] S33, spectral fitting method (SFM) This method replaces the fluorescence emission equation with r ( l )and ( l ) is recorded as a constant a and b , rewrite the formula into the form of formula (7), and calculate the fluorescence by fitting the unknown coefficients: (7) In the formula, = , a is the true reflectivity, b represents the calculated fluorescence.

[0034] The inversion bands used when calculating the UAV SIF using the SFM method are 686.1nm, 690.4nm, and 694.7nm.

[0035] S4, evaluating the accuracy of the UAV SIF images obtained by the three methods in step S3, and selecting the data with the highest fitting accuracy with the ground SIF as input data.

[0036] Using ASD ground canopy hyperspectral radiance data, the three methods mentioned above (FLD, 3FLD, SFM) were used to calculate three ground SIFs respectively as ground verification data for the three UAV SIFs obtained in step S3. During verification, the calculation methods of ground SIF and UAV SIF were kept consistent. Using 686-690nm as the ground SIF retrieval window, the UAV SIF calculated by the FLD and 3FLD methods needs to be normalized to reduce the errors caused by the retrieval method and the original data.

[0037] When using ASD ground canopy hyperspectral radiance data to retrieve SIF, the FLD method uses the bands within the Fraunhofer lines ( l in ) has a wavelength of 686nm, in the band outside the Fraunhofer line ( l out ) wavelength is 688nm; the 3FLD method uses a band within the Fraunhofer line ( l in ) has a wavelength of 686nm, in the band outside the Fraunhofer line ( l left and l right ) wavelengths are 688nm and 691nm respectively; the wavelength used in the SFM method is 684nm-688nm.

[0038] S5, Inversion of soil respiration rate in desert reclamation area based on random forest.

[0039] Principle of random forest algorithm: Random forest (RF) is essentially an improvement on the decision tree algorithm. It combines multiple decision trees together. Each tree in the forest has the same distribution and is independent of each other. The classification error depends on the classification ability of each tree and the correlation between them. A random method is used to split each node for feature selection, and then the errors generated in different situations are compared. The number of selected features is determined by the intrinsic estimation error, classification ability, and correlation that can be detected. The classification ability of a single tree may be small, but after randomly generating a large number of decision trees, a test sample can select the most likely classification through the classification results of each tree after statistics.

[0040] After obtaining the trained random forest model, new samples are input into the model, and each decision tree in the "forest" makes a judgment separately. By synthesizing the judgments of all decision trees, it is determined which class the input sample belongs to, thus predicting the input sample as that class.

[0041] Algorithm process: (1) If the total number of training samples is N, then each single decision tree randomly draws n samples from the N training sets with replacement (sampling with replacement) as the training samples for this single tree; (2) Let the number of input features of the training examples be M. Then, when splitting at each node of each decision tree, randomly select m (m << M) input features from the M input features, and then select the best one from these m input features for splitting. m does not change during the process of constructing the decision tree; (3) Each tree continues to split like this until all the training examples at this node belong to the same class.

[0042] S51. Construct a random forest model. During the modeling process, use SIF as the independent variable representing vegetation information, and then input the measured soil respiration data, thermal infrared images, characteristic bands, and TVDI images obtained in the above steps into the constructed model to estimate the soil respiration rate and obtain the soil respiration spatial distribution map.

[0043] S52. Use a computer to extract the values of the measured sample points on the soil respiration estimation image and compare them with the in-situ measured values of soil respiration to evaluate the accuracy of the model estimation.

[0044] Example 1 In this example, the reclamation area of Hongshaquan Coal Mine is selected as the research object, and the soil respiration rate of the desert reclamation area is inverted based on SIF and combined with the random forest algorithm. The main results are as follows: (1) Characteristic band screening The measured soil respiration data in the study area is negatively correlated with the reflectance of each band of the UAV. In the band range of 400nm - 700nm, the correlation between the soil respiration data and the reflectance of each band of the UAV passes the 0.01 significance test (|correlation coefficient| > 0.4) and shows a "continuous aggregation" distribution. To reduce the number of characteristic bands, taking the difference in color light as the secondary screening basis, within the spectral range of the already screened characteristic bands, combined with the magnitude of the correlation coefficient value and the band selection scheme of the Landsat series satellite multispectral sensors, four color lights (violet, indigo, green, red) are selected, and the average value of four characteristic bands is obtained within the wavelength range corresponding to each color light as the final band for modeling. A total of 4 characteristic bands are obtained through the above method, as shown in Table 1 for subsequent remote sensing model construction.

[0045] Table 1 Results of characteristic band screening

[0046] (2) Calculation of relevant spectral indices In order to represent the soil temperature and humidity in the study area, thermal infrared and TVDI images of the study area were obtained, and the results are as follows: Figure 2 , Figure 3 shown.

[0047] (3) UAV-SIF image computing The spatial distribution of UAV SIF inversion results is as follows: Figure 4-Figure 6 As shown, the calculation results of the FLD and 3FLD methods are normalized.

[0048] (4) Verification of UAV-SIF estimation results using ASD ground canopy hyperspectral radiance data The fitting results are as follows Figure 7-Figure 9 As shown in the figure, according to the fitting results, when using the spectral fitting method (SFM), the SIF fitting accuracy calculated by UAV-SIF and ground ASD data is the highest (R 2 =0.722), so the UAV-SIF calculated by the SFM method was used as the input data for constructing the soil respiration inversion model.

[0049] (5) The soil respiration inversion model is constructed using the characteristic band image values, thermal infrared, TVDI, UAV-SIF values ​​and measured soil respiration data at different measured sample points. After the model is obtained, the result image obtained in the above steps is input as the model independent variable, and finally the spatial distribution map of soil respiration in the study area is obtained, as shown in Figure 2. Figure 10 As shown in the figure, it can be seen that soil respiration in the study area is stronger in the vegetation coverage area, and the soil respiration rate is 0.22±0.14μmol·m -2 ·s -1 .

[0050] In this example, a total of 45 measured soil respiration data were collected, of which 40 were used for model construction and correlation calculation of characteristic bands, and 5 sample data were reserved for verification of the inversion results and fitting R 2 =0.76, and passed the 95% significance test (P<0.05). Figure 11 shown.

[0051] Example 2 The soil respiration rate inversion method in desert reclamation areas based on random forest algorithm is implemented in the following steps: S1. Measure the soil in the desert reclamation area to obtain measured soil respiration data, UAV hyperspectral, thermal infrared data and ASD ground canopy hyperspectral radiance data; S2, using UAV hyperspectral and measured soil respiration data to screen out characteristic bands, and using UAV hyperspectral to calculate TVDI to characterize environmental factors; S3, using the three methods of FLD, 3FLD and SFM, the UAV hyperspectral data were used to calculate the three UAV SIF values; S4, using ASD ground canopy hyperspectral radiance data to verify the UAV SIF values ​​obtained by the three methods in step S3, and selecting the UAV SIF data with the highest accuracy as the input data for soil respiration inversion; S5. Inversion of soil respiration rate in desert reclamation areas based on random forest algorithm.

[0052] Example 3 Based on Example 2, step S2 is specifically as follows: S21. Extract the UAV hyperspectral curve at each RTK coordinate of the measured soil respiration sample point, use the Person correlation coefficient method to calculate the correlation between different bands and the measured soil respiration data, and select the four bands with the highest correlation coefficients as the characteristic bands for soil respiration inversion; if the selected characteristic band is a continuous band, then according to the color light interval of the visible spectrum, the characteristic bands falling in different color light intervals are averaged as the final characteristic band for the independent variable input of the subsequent model construction; S22. Calculate TVDI based on UAV hyperspectral reflectance images to characterize environmental factors.

Claims

1. A method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm, characterized in that: Follow the steps below to implement it: S1. Measure the soil in the desert reclamation area to obtain measured soil respiration data, UAV hyperspectral, thermal infrared data and ASD ground canopy hyperspectral radiance data; S2, using UAV hyperspectral and measured soil respiration data to screen out characteristic bands, and using UAV hyperspectral to calculate TVDI to characterize environmental factors; S3, using the three methods of FLD, 3FLD and SFM, the UAV hyperspectral data were used to calculate the three UAV SIF values; S4, using ASD ground canopy hyperspectral radiance data to verify the UAV SIF values ​​obtained by the three methods in step S3, and selecting the UAV SIF data with the highest accuracy as the input data for soil respiration inversion; S5. Inversion of soil respiration rate in desert reclamation areas based on random forest algorithm.

2. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 1 is characterized in that: Step S2 is specifically as follows: S21, extracting the UAV hyperspectral curve at each RTK coordinate of the measured soil respiration sample point, using the Person correlation coefficient method to calculate the correlation between different bands and the measured soil respiration data, and selecting the four bands with the highest correlation coefficients as the characteristic bands for soil respiration inversion; S22. Calculate TVDI based on UAV hyperspectral reflectance images to characterize environmental factors.

3. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 2 is characterized in that: In step S21, if the selected characteristic band is a continuous band, the characteristic bands falling in different color light intervals are averaged according to the color light interval of the visible spectrum as the final characteristic band for independent variable input of subsequent model construction.

4. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 1, characterized in that: In step S3, the FLD method calculates the UAV SIF using the following formula: In the formula, λ in It represents the wavelength of the band within the Fraunhofer line, and its value is 690.4nm; λ out It represents the wavelength of the band within the Fraunhofer line, and its value is 686.1nm; L ( λ ) indicates the wavelength λ The apparent radiance of plants at E ( λ ) indicates the wavelength λ The solar irradiance at Indicated in wavelength λ in The transmitted SIF at .

5. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 1, characterized in that: In step S3, the 3FLD method calculates the UAV SIF using the following formula: In the formula, λin It represents the wavelength of the band within the Fraunhofer line, and its value is 690.4nm; λleft , λright Represents the wavelengths of the bands on both sides of the Fraunhofer line, which are 686.1nm and 694.7nm respectively; ; ; L ( λ ) indicates the wavelength λ The apparent radiance of plants at E ( λ ) indicates the wavelength λ The solar irradiance at Indicated in wavelength λin The transmitted SIF at .

6. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 1, characterized in that: In step S3, the SFM method calculates fluorescence using the following formula: In the formula, L ( λ ) indicates the wavelength λ The apparent radiance of plants at = , E ( λ ) indicates the wavelength λ Solar irradiance at a is the true reflectivity; b represents the calculated fluorescence; the inversion bands used in the calculation process are 686.1nm, 690.4nm and 694.7nm.

7. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 1, characterized in that: Step S4 is specifically as follows: Using the FLD, 3FLD and SFM methods, three ground SIF values ​​were calculated using the ASD ground canopy hyperspectral radiance data as ground verification data for the three UAV SIF values ​​obtained in step S3. Before verification, the UAV SIF calculated by the FLD and 3FLD methods was normalized, and the UAV SIF data with the highest fitting accuracy with the ground SIF was selected as the input data for soil respiration inversion.

8. The method for inverting soil respiration rate in desert reclamation areas based on random forest algorithm according to claim 1, characterized in that: Step S5 is specifically as follows: S51, constructing a random forest model, using SIF as an independent variable representing vegetation information during the modeling process, and then inputting the measured soil respiration data, thermal infrared images, characteristic bands and TVDI images obtained in the above steps into the constructed model to estimate the soil respiration rate and obtain a soil respiration spatial distribution map; S52, extracting the values ​​of the measured sample points on the soil respiration estimation image, comparing them with the in-situ measured values ​​of the soil respiration, and evaluating the accuracy of the model estimation.

Images