A remote sensing method for large-area lake particulate organic carbon

Abstract

The present application relates to a kind of remote sensing methods of large area lake particle organic carbon, comprising: obtaining satellite remote sensing data and calculating its equivalent reflectivity;Based on the equivalent reflectivity, the reflection peak height PH1 of green light band of multiple lakes is calculated;Several lake samples with the highest and lowest PH1 value are taken respectively as turbid water sample and clear water sample;And the correlation model of remote sensing reflectivity and POC concentration is established for turbid water sample and clear water sample;With preset step, the PH1 value threshold range of the multiple lakes is traversed, the multiple segmentation thresholds are used to redivide lake and estimate POC concentration, calculate remote sensing accuracy, with the PH1 value corresponding to the highest remote sensing accuracy as the final water classification threshold;Using the water classification threshold, multiple to-be-measured lakes are divided into turbid water sample and clear water sample, and the parameters of the correlation model are recalibrated, to obtain the final POC concentration estimation model.The method of the present application can realize the synchronous remote sensing of POC of different types of lakes in large area.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology, specifically relating to a remote sensing method for particulate organic carbon in large-area lakes. Background Technology

[0002] Lake particulate organic carbon (POC) is a direct product of photosynthesis by aquatic algae, an important nutrient source for plankton and benthic organisms, and a key component of the lake carbon cycle, making it of significant research importance for regulating the global carbon cycle and regional aquatic environment. However, due to the combined influence of natural and anthropogenic factors, POC concentrations between and within lakes exhibit spatiotemporal variability. Therefore, it is essential to utilize large-area coverage and periodically revisited satellite data for dynamic remote sensing monitoring of lake POC.

[0003] Lake pollutant eutrophic organic compounds (POCs) originate from algae and sediment resuspension. However, POCs themselves are not photosensitive, and the POC content and optical properties vary among different algae and suspended matter, posing significant challenges to simultaneous remote sensing of POCs in different types of lakes over large areas. Currently reported lake POC remote sensing methods include: ① optical remote sensing, which infers POC concentration by analyzing reflectance, absorption coefficient, and scattering coefficient; ② remote sensing modeling, which uses statistical analysis or machine learning methods to construct POC data based on remote sensing and measured data; ③ remote sensing indexing, which estimates POC concentration using color index, difference index, and ratio index; and ④ data fusion, which integrates remote sensing data with water quality data and geographic information data to estimate POC concentration. However, previous satellite remote sensing studies on lake POCs have focused only on single or a few shallow eutrophic lakes, with most studies concentrating on lakes in the middle and lower reaches of the Yangtze River. Unlike shallow eutrophic lakes, 67% of lakes globally exhibit characteristics of great depth (≥6.0 meters), low algal productivity, and high transparency.

[0004] Therefore, to better understand the global lake carbon cycle, it is urgent to develop a POC remote sensing algorithm applicable to shallow / deep, clear / turbid, and oligotrophic / eutrophic lakes. Due to the significant differences in the bio-optical characteristics of different water types, previous studies have shown that "classification first, then algorithm construction" is an effective approach for large-scale lake chlorophyll (Chl-a) or total suspended matter (TSM) remote sensing monitoring. This approach has also been applied to synchronous remote sensing monitoring of POC concentrations in turbid eutrophic lakes and marginal sea areas. However, in practical applications, several technical challenges remain to be addressed, including:

[0005] (1) Differences in lake water environment lead to different POC compositions, and it is necessary to cluster water bodies with similar POC compositions based on water environment characteristic parameters.

[0006] (2) Water color remote sensing satellite data can only provide remote sensing reflectance of fixed bands. It is necessary to determine the remote sensing index and segmentation threshold to distinguish clear / turbid water bodies in order to achieve high-precision remote sensing of POC concentration.

[0007] (3) The optical properties of POCs in different types of water bodies vary significantly, and it is necessary to combine satellite data and POC optical properties to construct corresponding remote sensing algorithms. Summary of the Invention

[0008] To achieve simultaneous remote sensing monitoring of POC concentration in different types of lakes, this invention first constructs a remote sensing method for particulate organic carbon in large-area lakes based on bio-optical synchronous survey data.

[0009] To achieve the above-mentioned technical objectives, the present invention adopts the following solution:

[0010] A remote sensing method for particulate organic carbon in large-area lakes includes:

[0011] Acquire satellite remote sensing data and calculate its equivalent reflectance using measured water reflectance;

[0012] The reflection peak height PH1 of the green light band of multiple lakes was calculated based on the equivalent reflectance.

[0013] Several lake samples with the highest pH1 value and several lake samples with the lowest pH1 value were selected as turbid water samples and clear water samples, respectively; and a correlation model between remote sensing reflectance and POC concentration was established for turbid water samples and clear water samples.

[0014] The pH1 value threshold range of the multiple lakes is traversed with a preset step size to determine candidate pH1 values. The lakes are reclassified into turbid water samples and clear water samples using the candidate pH1 values ​​as the segmentation threshold. The remote sensing accuracy of the turbid water samples and clear water samples is calculated based on the relevant model. The pH1 value corresponding to the highest remote sensing accuracy is used as the final water classification threshold.

[0015] Using the water body classification threshold, multiple lakes to be tested are divided into turbid water samples and clear water samples, and the parameters of the relevant models are recalibrated to obtain the final POC concentration estimation model.

[0016] In a preferred embodiment, the satellite remote sensing data is OLCI / Sentinel-3A data.

[0017] As a preferred embodiment, the reflection peak height PH1 of the green light band is calculated with the reflectance of the 490nm and 754nm bands as a baseline, which can reduce the influence of atmospheric correction error to a certain extent.

[0018] Furthermore, the reflection peak height PH1 of the green light band is calculated based on the following formula:

[0019] PH1 = R rs (560)-(R rs (490)+0.27×(R rs (754)-R rs (490)))

[0020] In the formula: R rs (λ) represents the remote sensing reflectance at wavelength λ.

[0021] As a preferred implementation method, the 10% of lake samples with the highest pH value and the 10% of lake samples with the lowest pH value are taken as turbid water samples and clear water samples, respectively.

[0022] As a preferred implementation method, for turbid water samples and clear water samples, correlation models with POC concentration in different mathematical forms are established based on single-band remote sensing reflectance and remote sensing reflectance band combinations, respectively. The model with the highest correlation is selected as the POC estimation model for the corresponding water sample.

[0023] As a preferred implementation method, a univariate quadratic correlation model between POC concentration and the height of the Chl-a fluorescence peak in the red light band is established for turbid water samples;

[0024] For clear water samples, a linear correlation model was established between logarithmic POC concentration and the near-infrared-blue-green light band ratio index.

[0025] As a preferred implementation, the Chl-a fluorescence peak height in the red band is calculated using the reflectance of the 681nm and 754nm bands as a baseline, which can mitigate the impact of atmospheric correction errors to some extent.

[0026] As a preferred implementation, the near-infrared-blue-green band ratio index (Index) and the red band Chl-a fluorescence peak height (PH2) are calculated based on the following formulas:

[0027] Index = R rs (754) / R rs (490)-R rs (754) / R rs (560)

[0028] PH2=R rs (708)-(R rs (681)+0.37×(R rs (754)-R rs (681)))

[0029] In the formula, R rs (λ) represents the remote sensing reflectance at wavelength λ.

[0030] As a preferred implementation, remote sensing accuracy is evaluated based on mean absolute percentage error, root mean square error, and / or relative deviation indicators.

[0031] As a preferred implementation, the method further includes: estimating the lake POC on different days using remote sensing data and the final POC concentration estimation model, and then calculating the monthly average and lake average POC concentrations to obtain a remote sensing distribution map of POC concentration in a large area of ​​lakes.

[0032] This invention, through comprehensive analysis of measured data, reveals that water bodies with high POC concentrations are generally more turbid. Based on this, the invention proposes first classifying lake water bodies into clear and turbid types, and then constructing POC remote sensing algorithms for each type, thus solving the water body classification problem before POC algorithm construction. Addressing the issues of water body type classification indicators and segmentation thresholds, this invention proposes a green light band reflectance peak height index, and uses this index to classify clear / turbid lake water bodies, determining the optimal segmentation threshold through a stepwise search method. Regarding the construction of POC remote sensing algorithms for different water body types, this invention constructs a band ratio index to invert the POC concentration in clear water by comprehensively analyzing the spectral characteristics of the two types of water bodies, and uses the Chl-a fluorescence peak height in the 708nm band to invert the POC concentration in turbid water.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] ①This invention first classifies water body types based on the bio-optical characteristics of POCs and then constructs remote sensing algorithms for each type, which can be applied to synchronous remote sensing of POCs in different types of lakes in a large area;

[0035] ② The constructed algorithm uses band ratio and reflectance peak height to classify water body types and estimate POC concentration, which can reduce the impact of atmospheric correction error to a certain extent;

[0036] ③ The algorithm's accuracy is higher than other published remote sensing algorithms. Applying the constructed algorithm to OLCI / Sentinel-3A satellite data, it achieved, for the first time, nationwide accuracy for lakes (≥20m). 2 Remote sensing mapping of POC concentration. Attached Figure Description

[0037] Figure 1 These are remote sensing indicators of POC (Positive Occurrence) for different types of lakes.

[0038] Figure 2 This is a comparison of the remote sensing accuracy of POC concentration in lakes. Detailed Implementation

[0039] Taking different regions and types of lakes in China as examples, this invention details a specific implementation method for constructing a remote sensing algorithm for POC concentration applicable to large-area lakes, as follows:

[0040] (1) Atmospheric correction of satellite remote sensing data. OLCI data (OLCI / Sentinel-3A) from the cloudless Sentinel-3A satellite covering the entire country in July 2017 were obtained from the European Copernicus Open Data Center (https: / / scihub.copernicus.eu / ). Atmospheric correction was then performed using the C2RCC processor of the SNAP (Sentinel Application Platform) software platform to obtain water reflectance, and geometric clipping was performed to obtain data showing a water area greater than 20 km² across the country. 2 Lake water reflectance data.

[0041] (2) Calculation of equivalent reflectance of satellite. To construct a POC remote sensing algorithm suitable for the new advanced OLCI / Sentinel-3A satellite sensor, this invention calculates the equivalent reflectance of each band of OLCI / Sentinel-3A based on the spectral response function of each band of OLCI / Sentinel-3A using hyperspectral data of water reflectance observed in the field. The calculation formula is shown in equation (1):

[0042]

[0043] In the formula, f(x) is the spectral response function of each band of OLCI / Sentinel-3A, and L(x) is the solar irradiance at the average distance between the Earth and the Sun.

[0044] (3) Determination of water body classification indicators. Based on the survey data of 1269 stations in 49 lakes in China, the bio-optical characteristics of POC at different stations and their correlation with water environment characteristics were comprehensively analyzed. The POC concentration and water reflectance of lake water under different salinity, transparency and nutrient status were compared. It was found that the POC concentration of clear water is usually lower, while the POC concentration of turbid water is usually higher, and turbid water usually shows a significant reflectance peak in the green light band. Therefore, this invention designs a method for calculating the reflectance peak height (PH1) in the green light band based on the equivalent reflectance of OLCI / Sentinel-3A. The reflectance of the 490nm and 754nm bands is used as a baseline for calculation, which can reduce the influence of atmospheric correction error to a certain extent. PH1 is used to distinguish between clear water and turbid water. The formula for calculating PH1 is as shown in equation (2):

[0045] PH1 = R rs (560)-(R rs (490)+0.27×(R rs (754)-Rrs (490))) (2)

[0046] In the formula, R rs (490), R rs (560) and R rs (754) are the equivalent reflectances of the OLCI / Sentinel-3A bands with center wavelengths of 490 nm, 560 nm and 754 nm, respectively, calculated from the measured water body spectra.

[0047] (4) Selection of POC Inversion Indicators. The 10% of samples with the lowest pH1 values ​​were selected as clear water samples, while the 10% of samples with the highest pH1 values ​​were selected as turbid water samples. Then, the reflectance spectral characteristics of POC in clear and turbid water samples were analyzed. It was found that the logarithmic POC concentration in clear water samples showed a significant linear correlation with the near-infrared-blue-green band ratio index, while the logarithmic POC concentration in turbid water samples showed a quadratic correlation with the Chl-a fluorescence peak height near 708 nm (characterized using pH2, calculated with reflectance at 681 nm and 754 nm as a baseline, which can weaken the influence of atmospheric correction error to some extent). Figure 1 Therefore, this invention uses Index and PH2 to estimate the POC concentration in clear and turbid water bodies, respectively, and the calculation formula is shown in equation (3):

[0048]

[0049] In the formula, R rs (490), R rs (560), R rs (681), R rs (708) and R rs (754) are the equivalent reflectances of the OLCI / Sentinel-3A bands with center wavelengths of 490 nm, 560 nm, 681 nm, 708 nm and 754 nm, respectively, calculated from the measured water body spectra.

[0050] (5) Determination of water body classification threshold. The pH1 threshold range is calculated from the measured OLCI / Sentinel-3A band equivalent reflectance to -0.011 to 0.036. Therefore, this invention uses 0.001 as the interval to obtain the POC remote sensing accuracy of clear and turbid water bodies when traversing all pH1 segmentation thresholds. The pH1 value at which the POC achieves the highest remote sensing accuracy is the water body classification threshold (pH1 = 0.0125). The accuracy evaluation indexes are shown in equation (4):

[0051]

[0052] In the formula, MAPD is the mean absolute percentage error, RMSE is the root mean square error, bias is the relative deviation, and POC is the mean absolute percentage error. i mod and POC i field These are the remote sensing and measured values ​​of POC concentration, respectively.

[0053] (6) POC Algorithm Parameter Calibration. Using the determined PH1 segmentation threshold, all samples were divided into clear and turbid water bodies. Then, based on the equivalent reflectance of the OLCI / Sentinel-3A band, Index and PH2 were calculated using equation (3). Finally, Index was linearly fitted to the logarithmic value of POC concentration in clear water, and PH2 was fitted to the logarithmic value of POC concentration in turbid water using a quadratic equation, thus calibrating the POC algorithm parameters. The POC remote sensing algorithm after parameter calibration is shown in equation (5):

[0054]

[0055] In the formula, ln(POC) is the logarithmic value of POC concentration, and the index Index and PH1 are defined as shown in formula (3).

[0056] (7) POC remote sensing accuracy assessment. Based on the equivalent reflectance of the OLCI / Sentinel-3A band, the POC concentration is estimated by simulation using Equation (5), and the MAPD, RMSE, and bias error values ​​are calculated by comparing them with the measured POC concentration. At the same time, the parameters of the commonly used NDCI (Normalized Difference Carbon Index) method for POC remote sensing of turbid water bodies are recalibrated based on all measured data, and the POC concentration is estimated using the recalibrated formula (6), and the accuracy is compared with the POC remote sensing model (Equation (5)) constructed in this invention.

[0057]

[0058] In the formula, ln(POC) is the logarithmic value of POC concentration, and R rs (413), R rs (443), R rs (490) and R rs (560) represents the equivalent reflectance of the OLCI / Sentinel-3A bands with center wavelengths of 413 nm, 443 nm, 490 nm, and 560 nm, calculated from measured water body spectra. The results show that the MAPD accuracy of the model and the NDCI index in this invention is 35.93% and 66.92%, respectively, indicating that the POC model constructed in this invention has higher POC remote sensing accuracy. Figure 2 ).

[0059] (8) Remote sensing mapping of POC concentration in lakes across the country. The calibrated POC remote sensing model (5) was applied to estimate the POC concentration on different days using OLCI / Sentinel-3A satellite remote sensing reflectance. Then, the monthly average and lake average POC concentrations were calculated to obtain a remote sensing distribution map of POC concentration in lakes across the country.

Claims

1. A remote sensing method for particulate organic carbon in a large-area lake, characterized in that, include: Acquire satellite remote sensing data and calculate its equivalent reflectance using measured water reflectance; The reflection peak height PH1 of the green light band of multiple lakes was calculated based on equivalent reflectance: ; In the formula: R rs (λ) represents the remote sensing reflectance at wavelength λ; Several lake samples with the highest pH1 value and several lake samples with the lowest pH1 value were selected as turbid water samples and clear water samples, respectively; and a correlation model between remote sensing reflectance and POC concentration was established for turbid water samples and clear water samples. The pH1 value threshold range of the multiple lakes is traversed with a preset step size to determine candidate pH1 values. The lakes are reclassified into turbid water samples and clear water samples using the candidate pH1 values ​​as the segmentation threshold. The remote sensing accuracy of the turbid water samples and clear water samples is calculated based on the relevant model. The pH1 value corresponding to the highest remote sensing accuracy is used as the final water classification threshold. Using the aforementioned water body classification threshold, multiple lakes to be tested were divided into turbid water samples and clear water samples. The parameters of the correlation model were recalibrated to obtain the final POC concentration estimation model. For turbid water samples, a univariate quadratic correlation model was established between POC concentration and the height of the Chl-a fluorescence peak in the red light band. For clear water samples, a linear correlation model was established between the logarithmic POC concentration and the near-infrared-blue-green light band ratio index.

2. The method according to claim 1, characterized in that, The satellite remote sensing data is OLCI / Sentinel-3A data.

3. The method according to claim 1, characterized in that, The top 10% of lake samples with the highest pH value and the bottom 10% of lake samples with the lowest pH value were selected as turbid water samples and clear water samples, respectively.

4. The method according to claim 1, characterized in that, For turbid and clear water samples, correlation models with POC concentration were established based on single-band remote sensing reflectance and remote sensing reflectance band combinations, respectively. The model with the highest correlation was selected as the POC estimation model for the corresponding water sample.

5. The method according to claim 1, characterized in that, The peak height of Chl-a fluorescence in the red band was calculated using the reflectance at the 681 nm and 754 nm bands as a baseline.

6. The method according to claim 5, characterized in that, The near-infrared-blue-green band ratio index (Index) and the red band Chl-a fluorescence peak height (PH2) are calculated based on the following formulas: ; In the formula, R rs (λ) represents the remote sensing reflectance at wavelength λ.

7. The method according to claim 1, characterized in that, Also includes: The POC concentration of lakes on different days was estimated using remote sensing data and the final POC concentration estimation model. Then, the monthly average and lake average POC concentrations were calculated to obtain a remote sensing distribution map of POC concentration in lakes over a large area.

Images