Offshore area nutritive salt concentration remote sensing inversion method based on satellite fusion

Abstract

The invention discloses an offshore area nutritive salt concentration remote sensing inversion method based on satellite fusion, and the method comprises the following steps: S1, carrying out the Rayleigh correction of the original L1C data of the optical images of a high-spatial-resolution satellite and a low-spatial-resolution satellite, and obtaining the remote sensing reflectance after the Rayleigh correction; s2, performing mask processing on cloud and land pixels in the remote sensing reflectance ratio data after Rayleigh correction; s3, constructing a nutritive salt training data set based on low-space satellite data; s4, constructing a cross-satellite fusion training data set; s5, an AutoGluon-DIN machine learning model and an AutoGluon-DIP machine learning model which are based on the low-resolution satellite are established, and model training is carried out on the AutoGluon-DIN machine learning model and the AutoGluon-DIP machine learning model; s6, an AutoGluon-transfer machine learning model fusing the high-resolution satellite and the low-resolution satellite is established, and model training is carried out; and S7, sequentially applying the trained AutoGluon-transfer machine learning, the AutoGluon-DIN machine learning model and the AutoGluon-DIP machine learning model to a high-resolution satellite, and obtaining a high-spatial-resolution remote sensing inversion product of the nutritive salt concentration of the offshore area.

Description

Technical Field

[0001] The present invention relates to the field of ocean observation technology, and in particular to a remote sensing inversion method for nutrient concentration in nearshore waters based on satellite fusion. Background Art

[0002] Nutrients such as nitrogen and phosphorus promote the growth of phytoplankton, while driving primary production and maintaining biodiversity. However, human activities from agriculture, urban runoff and industrial emissions often emit excessive nutrients, leading to eutrophication of nearshore waters. This process not only degrades water quality, but also destroys the ecological functions of water bodies and causes harmful algal blooms. Algal blooms have a negative impact on water resources, marine life, and local economies that rely on tourism and fisheries, so effective water quality monitoring and assessment is crucial. The concentrations of dissolved inorganic nitrogen (DIN) and dissolved inorganic phosphorus (DIP) are often used to characterize nutrient levels and are key parameters for assessing water quality.

[0003] Since the physical mechanism relationship between the current optical satellite signals and the concentration of water nutrients (including dissolved inorganic nitrogen and dissolved inorganic phosphorus) is not yet clear, most optical satellite inversion algorithms for nutrient concentration inversion are still based on empirical algorithms and neural network algorithms, and the input parameters include satellite products such as remote sensing reflectance (Rrs), sea surface temperature (SST), chlorophyll concentration (Chl) and sea surface salinity (SSS). However, for coastal waters, the resolution of environmental factor products such as SST, Chl and SSS is low, making it difficult to serve as an effective input for the algorithm; in addition, Rrs must be atmospherically corrected, and the atmospheric correction algorithm has large uncertainties in coastal shallow waters. If the atmospheric correction fails, it is easy to cause invalid or erroneous Rrs data, which in turn affects the inversion accuracy of DIN and DIP, and it is impossible to obtain a reliable remote sensing nutrient distribution. In particular, the water bodies in coastal waters have large temporal and spatial heterogeneity, and operational nutrient monitoring requires higher temporal and spatial resolution. In addition, especially for small estuaries and bays like Xiamen Bay, nutrient inversion algorithms require high spatial resolution satellite data input. However, due to the long revisit period of high spatial resolution satellites, the amount of measured data that can be matched is small, and model development is therefore limited by the amount of data. Summary of the invention

[0004] To solve the above problems, the present invention proposes a remote sensing inversion method for nutrient concentration in coastal waters based on satellite fusion, which can accurately estimate the distribution of nutrient DIN and DIP concentrations in coastal waters through the automatic machine learning model AutoGluon.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion, comprising the following steps:

[0007] S1. Rayleigh correction is performed on the original L1C data of the optical images of the high spatial resolution satellite and the low spatial resolution satellite respectively to obtain the remote sensing reflectance after Rayleigh correction. The calculation formula is:

[0008]

[0009] In the formula, ρ rc is the remote sensing reflectance after Rayleigh correction; λ is the wavelength; ρ t is the reflectivity observed by the satellite sensor at the top of the atmosphere; t g is the gas transmittance in the atmosphere; ρ r is the Rayleigh reflectivity produced by multiple molecular scattering under aerosol-free conditions; T is the direct transmittance; ρ g It is the scattered signal caused by solar flare;

[0010] S2, masking the cloud and land pixels in the Rayleigh-corrected remote sensing reflectance data;

[0011] S3, the ρ of the low spatial resolution satellite selected in step S2 rc The water body pixels are matched with the measured nutrient concentrations of the corresponding pixels to construct a nutrient training data set;

[0012] S4, the ρ of the low spatial resolution satellite selected in step 2 rc The ρ of water body pixels and the high spatial resolution satellite passing through the study area on the same day rc Water body pixels are matched to construct satellite fusion training data set;

[0013] S5. Establish an AutoGluon-DIN machine learning model for estimating dissolved inorganic nitrogen concentration and an AutoGluon-DIP machine learning model for estimating dissolved inorganic phosphorus concentration, and use the nutrient salt training dataset to train the AutoGluon-DIN machine learning model and the AutoGluon-DIP machine learning model respectively;

[0014] S6. Establish 8 AutoGluon-transfer machine learning models to map the band signals of high spatial resolution satellites to the bands within the common wavelength range of low spatial resolution satellites; use the satellite fusion training data set to train the 8 AutoGluon-transfer machine learning models respectively;

[0015] S7. Apply the trained AutoGluon-DIN machine learning model, AutoGluon-DIP machine learning model and AutoGluon-transfer machine learning model to the satellite-fused remote sensing inversion of nutrient concentration in coastal waters.

[0016] Preferably, the specific process of step S2 is: a threshold is set according to the characteristics of the satellite sensor, and pixels above the threshold are determined to be cloud or land pixels and are directly eliminated, and only bands within the wavelength range shared by high spatial resolution satellites and low spatial resolution satellites are retained.

[0017] Preferably, the matching process in step S3 covers nutrient concentration data measured in different seasons and under different atmospheric conditions.

[0018] Preferably, when performing model training in step S5, the input of the AutoGluon-DIN machine learning model is the ρ of the low spatial resolution satellite rc (λ) and related satellite metadata, and the output is the concentration of dissolved inorganic nitrogen; the input of the AutoGluon-DIP machine learning model is the ρ of low spatial resolution satellite rc (λ) and related satellite metadata, the output is dissolved inorganic phosphorus concentration.

[0019] Preferably, when performing model training in step S6, the inputs of the eight AutoGluon-transfer machine learning models are ρ of the high spatial resolution satellite, rc (λ), the output is ρ of a low spatial resolution satellite rc (λ).

[0020] Preferably, the specific process of step S7 is: applying the trained AutoGluon-DIN machine learning model and AutoGluon-DIP machine learning model to the low spatial resolution satellite optical images in the satellite fusion verification data set in step S4 to generate a low spatial resolution nutrient concentration distribution map of the nearshore waters on the same day; applying the trained AutoGluon-DIN machine learning model, AutoGluon-DIP machine learning model and AutoGluon-transfer machine learning model to the high spatial resolution satellite optical images in the satellite fusion verification data set in step S4 to generate a high spatial resolution nutrient concentration distribution map of the nearshore waters on the same day; resampling the high spatial resolution nutrient concentration distribution to low spatial resolution, and comparing it with the low spatial resolution nutrient concentration distribution map to draw a density scatter comparison map.

[0021] After adopting the above technical solution, the present invention has the following beneficial effects:

[0022] 1. The present invention adopts the Rayleigh-corrected remote sensing reflectance ρ rc As the input of the algorithm, it does not require atmospheric correction, thus avoiding the invalidity and error problems of Rrs data. On this basis, it integrates multiple neural network models to significantly improve the accuracy of the model's nutrient concentration inversion in coastal waters such as Xiamen Bay.

[0023] 2. The present invention first constructs a nutrient inversion algorithm based on low spatial resolution satellite matching data, and then develops a fusion algorithm for satellite data of different resolutions. The constructed algorithm can then be directly applied to high spatial resolution satellite data, effectively solving the problem of insufficient training data when matching with measured data due to the long revisit time of high spatial resolution satellites.

[0024] 3. The present invention uses an open source automatic machine learning algorithm and integrates multiple neural network models. Compared with traditional empirical algorithms and single neural network models, it greatly improves the accuracy of the nutrient concentration inversion model in the Xiamen Bay area and can realize high temporal and spatial resolution remote sensing observation of nutrients in small bays. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a structural diagram of an AutoGluon neural network for nutrient estimation in an embodiment of the present invention;

[0026] Figure 2 is the verification result of the nutrient concentration algorithm in the embodiment of the present invention;

[0027] Figure 3 1 is a structural diagram of an AutoGluon neural network for band migration of Sentinel-2 satellites and Sentinel-3 satellites in an embodiment of the present invention;

[0028] Figure 4 The low spatial resolution and high spatial resolution nutrient concentration distribution maps of the Xiamen Bay area on the same day are obtained by inverting the Sentinel-3 and Sentinel-2 satellite image data in the embodiment of the present invention;

[0029] Figure 5 It is a scatter plot comparison of the normalized density of nutrient salt concentration at low spatial resolution and high spatial resolution in the Xiamen Bay area on the same day in an embodiment of the present invention;

[0030] Figure 6 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0032] like Figures 1 to 6 As shown, a method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion includes the following steps:

[0033] S1. Rayleigh correction is performed on the original L1C data of the optical images of the high spatial resolution satellite and the low spatial resolution satellite respectively to obtain the remote sensing reflectance after Rayleigh correction. The calculation formula is:

[0034]

[0035] In the formula, ρ rc is the remote sensing reflectance after Rayleigh correction; λ is the wavelength; ρ t is the reflectivity observed by the satellite sensor at the top of the atmosphere; t g is the gas transmittance in the atmosphere; ρ r is the Rayleigh reflectivity produced by multiple molecular scattering under aerosol-free conditions; T is the direct transmittance; ρ g It is the scattered signal caused by solar flare;

[0036] S2, masking the cloud and land pixels in the Rayleigh-corrected remote sensing reflectance data;

[0037] The specific process of step S2 is as follows: a threshold is set according to the characteristics of the satellite sensor, and pixels above the threshold are determined to be cloud or land pixels and are directly eliminated, and only the bands within the common wavelength range of high spatial resolution satellites and low spatial resolution satellites are retained;

[0038] S3, the ρ of the low spatial resolution satellite selected in step S2 rc The water body pixels are matched with the measured nutrient concentrations of the corresponding pixels to construct a nutrient training data set;

[0039] The matching process in step S3 covers the nutrient concentration data measured in different seasons and under different atmospheric conditions;

[0040] S4, the ρ of the low spatial resolution satellite selected in step 2 rc The ρ of water body pixels and the high spatial resolution satellite passing through the study area on the same day rc Water body pixels are matched to construct satellite fusion training data set;

[0041] S5. Establish an AutoGluon-DIN machine learning model for estimating dissolved inorganic nitrogen concentration and an AutoGluon-DIP machine learning model for estimating dissolved inorganic phosphorus concentration, and use the nutrient salt training dataset to train the AutoGluon-DIN machine learning model and the AutoGluon-DIP machine learning model respectively;

[0042] When the model is trained in step S5, the input of the AutoGluon-DIN machine learning model is the ρ of the low spatial resolution satellite. rc(λ) and related satellite metadata, and the output is the concentration of dissolved inorganic nitrogen; the input of the AutoGluon-DIP machine learning model is the ρ of low spatial resolution satellite rc (λ) and related satellite metadata, the output is dissolved inorganic phosphorus concentration;

[0043] S6. Establish 8 AutoGluon-transfer machine learning models to map the band signals of high spatial resolution satellites to the bands within the common wavelength range of low spatial resolution satellites; use the satellite fusion training data set to train the 8 AutoGluon-transfer machine learning models respectively;

[0044] When the model is trained in step S5, the input of the AutoGluon-DIN machine learning model is the ρ of the low spatial resolution satellite. rc (λ) and related satellite metadata, and the output is the concentration of dissolved inorganic nitrogen; the input of the AutoGluon-DIP machine learning model is the ρ of low spatial resolution satellite rc (λ) and related satellite metadata, the output is dissolved inorganic phosphorus concentration;

[0045] S7. Apply the trained AutoGluon-DIN machine learning model, AutoGluon-DIP machine learning model and AutoGluon-transfer machine learning model to the satellite-fused remote sensing inversion of nutrient concentration in coastal waters;

[0046] The specific process of step S7 is: applying the trained AutoGluon-DIN machine learning model and AutoGluon-DIP machine learning model to the low spatial resolution satellite optical images in the satellite fusion verification data set in step S4 to generate a low spatial resolution nutrient concentration distribution map of the nearshore waters on the same day; applying the trained AutoGluon-DIN machine learning model, AutoGluon-DIP machine learning model and AutoGluon-transfer machine learning model to the high spatial resolution satellite optical images in the satellite fusion verification data set in step S4 to generate a high spatial resolution nutrient concentration distribution map of the nearshore waters on the same day; resampling the high spatial resolution nutrient concentration distribution to low spatial resolution, and comparing it with the low spatial resolution nutrient concentration distribution map to draw a density scatter comparison map.

[0047] An example of applying the technical solution of the present invention to invert the nutrient salt concentration in the Xiamen Bay area:

[0048] Download the optical satellite image of Xiamen Bay. This example uses Sentinel-2 as a high-spatial-resolution optical satellite and Sentinel-3 as a low-spatial-resolution satellite. The open-source software Acolite is used to perform Rayleigh correction on the original L1C satellite image to calculate the ρ of each pixel. rc , the calculation formula is:

[0049]

[0050] In the formula, ρ rc is the remote sensing reflectance after Rayleigh correction; λ is the wavelength; ρ t is the reflectivity observed by the satellite sensor at the top of the atmosphere; t g is the gas transmittance in the atmosphere; ρ r is the Rayleigh reflectivity produced by multiple molecular scattering under aerosol-free conditions; T is the direct transmittance; ρ g It is the scattered signal caused by solar flare;

[0051] The pixel ρ rc The (865) value is used as the basis for judging cloud or land pixels. If the value is greater than 0.2, it is marked as a land or cloud pixel and masked; if the value is less than 0.2, it is marked as a water pixel and retained. All bands of Sentinel-3 have a spatial resolution of 300m, while the spatial resolutions of different bands of Sentinel-2 are inconsistent, ranging from 10m to 60m. To facilitate the construction of the nutrient inversion algorithm, linear interpolation is used to resample the different bands of the Sentinel-2 satellite to a unified spatial resolution of 10m. In order to construct the satellite fusion algorithm, only the common or similar bands of Sentinel-2 and Sentinel-3 are retained, and finally 8 bands are selected; the bands of Sentinel-2 are marked as λ S2 , including 443, 490, 560, 620, 709, 754, 779 and 865nm; the band of Sentinel-3 is marked as λ S3 , respectively 443, 492, 560, 665, 704, 740, 783 and 865nm.

[0052] The Sentinel-3 satellite ρ rc (λ S3) The water body pixels are matched with the measured nutrient data to construct a nutrient data set. The measured data used in this example include environmental monitoring station data (Monitoring data), voyage data (Survey data) and four buoy data (Buoy data). The four buoys are located in Baozhu Island (BZY), Yefengzhai (YFZ), Tongan Bay (TAW) and Hulishan (HLS) areas of Xiamen Bay. The study area is the Xiamen Bay area, and in order to improve the representativeness of the data, the coverage of the environmental monitoring station data is expanded to the coast of Fujian. Match the Sentinel-3 satellite data and measured nutrient data in different seasons and atmospheric conditions from 2018 to 2022. The matching data of the Hulishan buoy are selected to construct the nutrient verification data set, with 390 sets of data; the data matched with other measured data are used to construct the nutrient training data set, with a total of 2117 sets of data.

[0053] The Sentinel-3 satellite ρ rc (λ S3 ) Water pixel and its Sentinel-2 satellite image on the same day rc (λ S2 ) water pixels were matched to construct a satellite fusion training dataset. The Sentinel-3 and Sentinel-2 satellite images that passed through Xiamen Bay on the same day throughout 2022 were matched. The data covered different seasons and different atmospheric conditions, and a total of 254,233 sets of data were matched. In addition, the Sentinel-2 and Sentinel-3 images on October 21, 2019 were matched separately to construct a satellite fusion verification dataset.

[0054] Two AutoGluon machine learning models were established for DIN and DIP concentration estimation, respectively labeled as AutoGluon-DIN and AutoGluon-DIP (see Figure 1 ), the model is trained using the nutrient training data set in step 3. The input of the model is the Sentinel-3 satellite p rc (λ),ρ rc (λ) / ρ rc (560) and image date, the output is ln(DIN) or ln(DIP) respectively. After testing, the input variable ρ is increased rc (λ) / ρ rc (560) and image date can effectively improve the accuracy of the model; taking the logarithm of DIN / DIP as output can effectively stabilize the variance and reduce data skew, significantly improving the robustness of the model. The number of stacking layers of AutoGluon is set to 1, and the number of cross-validation folds is set to 5.

[0055] The accuracy of the AutoGluon-DIN / DIP model was verified using the nutrient validation dataset. The time series of the measured DIN / DIP (DINin situ / DINin situ) and the inverted DIN / DIP (DINest / DINest) in 2022 were plotted, and the absolute correlation coefficient (R 2 ), root mean square error (RMSD) and mean absolute correlation coefficient (MAPD) ​​(see Figure 2 ). Overall, R 2 It is close to 0.6 and the MAPD is less than 30%, indicating that the nutrient concentration estimated by the satellite is consistent with the concentration measured by the buoy, proving that the nutrient model of the present invention has high accuracy.

[0056] Eight AutoGluon machine learning models (AutoGluon-transfer) were built to convert the band data of the Sentinel-2 satellite into the corresponding band signals of the Sentinel-3 satellite (see Figure 3 ). The model is trained to convert the Sentinel-2 band ρ rc (λ S2 ) and Sentinel-3 band ρ rc (λ S3 ) to map the relationship between them so as to apply the AutoGluon-DIN / DIP model to the high spatial resolution Sentinel-2 data. In the embodiment, the satellite fusion training data set is used for model training. The inputs of the 8 models are respectively the ρ of the Sentinel-2 satellite rc (λ S2 ), the output is the ρ of the Sentinel-3 satellite rc (λ S3 ).

[0057] The AutoGluon-DIN / DIP model was applied to the Sentinel-3 satellite images in the satellite fusion validation dataset to generate a low spatial resolution (300 m) nutrient concentration distribution product map of Xiamen Bay. At the same time, the AutoGluon-DIN / DIP and AutoGluon-transfer models were applied to the Sentinel-2 satellite images in the satellite fusion validation dataset to generate a high spatial resolution (10 m) nutrient concentration distribution product map of Xiamen Bay on the same day (see Figure 4 ). The high spatial resolution nutrient concentration distribution product is resampled to a low spatial resolution and compared with the low spatial resolution nutrient concentration distribution map to draw a density scatter comparison map (see Figure 5 ). In general, most points are located near the 1:1 line, R 2It is close to 0.6 and the MAPD is less than 20%, indicating that the inversion results at the two resolutions have good consistency.

[0058] The above is only a preferred 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 should be based on the protection scope of the claims.

Claims

1. A remote sensing inversion method for nutrient concentration in coastal waters based on satellite fusion, characterized in that: The following steps are involved: S1. Rayleigh correction is performed on the original L1C data of the optical images of the high spatial resolution satellite and the low spatial resolution satellite respectively to obtain the remote sensing reflectance after Rayleigh correction. The calculation formula is: In the formula, ρ rc is the remote sensing reflectance after Rayleigh correction; λ is the wavelength; ρ t is the reflectivity observed by the satellite sensor at the top of the atmosphere; t g is the gas transmittance in the atmosphere; ρ r is the Rayleigh reflectivity produced by multiple molecular scattering under aerosol-free conditions; T is the direct transmittance; ρ g It is the scattered signal caused by solar flare; S2, masking the cloud and land pixels in the Rayleigh-corrected remote sensing reflectance data; S3, the ρ of the low spatial resolution satellite selected in step S2 rc The water body pixels are matched with the measured nutrient concentrations of the corresponding pixels to construct a nutrient training data set; S4, the ρ of the low spatial resolution satellite selected in step 2 rc The ρ of water body pixels and the high spatial resolution satellite passing through the study area on the same day rc Water body pixels are matched to construct satellite fusion training data set; S5. Establish an AutoGluon-DIN machine learning model for estimating dissolved inorganic nitrogen concentration and an AutoGluon-DIP machine learning model for estimating dissolved inorganic phosphorus concentration, and use the nutrient salt training dataset to train the AutoGluon-DIN machine learning model and the AutoGluon-DIP machine learning model respectively; S6. Establish 8 AutoGluon-transfer machine learning models to map the band signals of high spatial resolution satellites to the bands within the common wavelength range of low spatial resolution satellites; use the satellite fusion training data set to train the 8 AutoGluon-transfer machine learning models respectively; S7. Apply the trained AutoGluon-DIN machine learning model, AutoGluon-DIP machine learning model and AutoGluon-transfer machine learning model to the satellite-fused remote sensing inversion of nutrient concentration in coastal waters.

2. The method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion as claimed in claim 1, characterized in that: The specific process of step S2 is as follows: a threshold is set according to the characteristics of the satellite sensor, and pixels above the threshold are determined to be cloud or land pixels and are directly eliminated, and only bands within the wavelength range shared by high spatial resolution satellites and low spatial resolution satellites are retained.

3. The method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion as claimed in claim 1, characterized in that: The matching process in step S3 covers the nutrient concentration data measured in different seasons and under different atmospheric conditions.

4. The method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion as claimed in claim 1, characterized in that: When the model is trained in step S5, the input of the AutoGluon-DIN machine learning model is the ρ of the low spatial resolution satellite. rc (λ) and related satellite metadata, and the output is the concentration of dissolved inorganic nitrogen; the input of the AutoGluon-DIP machine learning model is the ρ of low spatial resolution satellite rc (λ) and related satellite metadata, the output is dissolved inorganic phosphorus concentration.

5. The method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion as claimed in claim 1, characterized in that: When the model is trained in step S6, the inputs of the eight AutoGluon-transfer machine learning models are respectively ρ of the high spatial resolution satellite rc (λ), the output is ρ of a low spatial resolution satellite rc (λ).

6. The method for remote sensing inversion of nutrient concentration in coastal waters based on satellite fusion as claimed in claim 1, characterized in that: The specific process of step S7 is: applying the trained AutoGluon-DIN machine learning model and AutoGluon-DIP machine learning model to the low spatial resolution satellite optical images in the satellite fusion verification data set in step S4 to generate a low spatial resolution nutrient concentration distribution map of the nearshore waters on the same day; applying the trained AutoGluon-DIN machine learning model, AutoGluon-DIP machine learning model and AutoGluon-transfer machine learning model to the high spatial resolution satellite optical images in the satellite fusion verification data set in step S4 to generate a high spatial resolution nutrient concentration distribution map of the nearshore waters on the same day; resampling the high spatial resolution nutrient concentration distribution to low spatial resolution, and comparing it with the low spatial resolution nutrient concentration distribution map to draw a density scatter comparison map.

Images