A space-borne lidar cloud detection method based on signal and background noise fusion

Abstract

A kind of spaceborne laser radar cloud detection method based on signal and background noise fusion, including obtaining laser radar system parameters;Obtain ground profile signal size and background noise rate;Respectively by ground profile signal calculation laser apparent reflectivity R l , background apparent reflectivity is calculated by background noise rate R b , and the normalized cloud difference index representing the relative difference between the two is constructed NDCI ;Whether the region has cloud is identified by preset NDCI Threshold value.The application gives a kind of spaceborne laser radar cloud detection method based on signal and noise apparent reflectivity difference, without relying on prior ground reflectivity data and complete backscattering profile data, directly using laser radar ground profile reflection signal and background noise rate, realizing cloud detection by the difference of apparent reflectivity.Effectively solve the problem that the existing ICESat-2 cloud detection method exists daytime misjudgment, relies on prior information, provide a kind of efficient cloud detection technology path for spaceborne laser radar.

Description

Technical Field

[0001] This invention relates to a cloud detection method for spaceborne lidar, specifically a cloud detection method for spaceborne lidar based on signal and background noise fusion, belonging to the field of lidar technology. Background Technology

[0002] Clouds are widespread in the Earth's atmosphere, with a global average coverage of 67%. In satellite-based Earth remote sensing, the presence of clouds can significantly interfere with satellite payloads, particularly impacting the acquisition of surface information by spaceborne lidar. Clouds with significant optical thickness can severely attenuate laser pulse signals, sometimes preventing them from reaching the surface. While clouds with less optical thickness may allow some laser light to penetrate to the surface, they can reduce the signal-to-noise ratio of the received signal and may also introduce additional modulation to the surface information in the signal. Therefore, in applications that rely on signal radiation characteristics, such as shallow sea topographic mapping, ice surface and lake depth inversion, and ocean optical parameter inversion, data affected by cloud cover must be removed.

[0003] Currently, various on-orbit or decommissioned spaceborne lidar systems have designed cloud detection or signal quality control algorithms tailored to their specific characteristics. For example, the CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) satellite identifies backscattering layers, uses the 5D CAD (Cloud-Aerosol Discrimination) algorithm to distinguish between clouds and aerosols, and then uses depolarization ratio, backscattering intensity, and color ratio to identify cloud phases. Meanwhile, GEDI (Global Ecosystems Dynamics Investigation Lidar) determines data anomalies by comparing the height difference between the highest reflectance layer and the prior DEM. ICESat-2 (Ice, Cloud and land Elevation Satellite-2) developed a cloud detection method based on apparent reflectance. This method uses pre-inputted true surface reflectance as a reference and compares it with the lidar's measured reflectance to determine the presence of clouds; however, this method heavily relies on prior surface reflectance data.

[0004] In most cases, when actively detecting target signals, spaceborne lidar systems treat solar background light scattered from the atmosphere and the Earth's surface as background noise. For example, the ATLAS (Advanced Topographic Laser Altimeter System) uses a highly sensitive multi-anode PMT as a single-photon detector. Although the solar background noise photons it detects are several orders of magnitude lower than the signal, they still maintain a good linear relationship with the solar background light. This allows the ATLAS system to be considered equivalent to a single-wavelength passive radiometer, providing an additional data source for obtaining surface remote sensing information. In recent years, some scholars have attempted to use the background noise of the ATLAS system for remote sensing, such as using the correlation between background noise intensity and surface reflectance characteristics to identify sea ice, land-sea boundaries, snow-covered areas, and land and vegetation-covered areas. Other scholars have derived remote sensing reflectance from background noise data in high seas areas, and some studies have used background noise data to estimate the optical thickness of aerosols over the ocean and invert the optical thickness of clouds in ocean areas. However, these studies all treated the background noise of spaceborne lidar as independent data, without fusing it with active detection signals, thus failing to fully realize the potential of combining the two in cloud detection. Summary of the Invention

[0005] This invention aims to provide a spaceborne lidar cloud detection method based on the fusion of signal and background noise, which solves the problems of existing cloud detection methods based on apparent reflectivity relying on prior surface reflectivity data and failing to effectively utilize background noise information.

[0006] The method includes the following steps:

[0007] Step 1: Obtain the parameters of the lidar system, including laser emission energy, lidar system optical efficiency, receiver system quantum efficiency, filter bandwidth, instrument calibration coefficient, lidar flight altitude, and also obtain information on the sun's position and spectral irradiance, including the sun's altitude angle and the corresponding band of extraterrestrial solar spectral irradiance.

[0008] Step 2: Obtain the surface profile signal magnitude and background noise rate.

[0009] Step 3: Calculate the apparent reflectivity of the laser from the surface profile signals. R l Calculate the apparent reflectance of the background from the background noise rate. R b And construct a normalized cloud difference index to characterize the relative difference between the two. NDCI (Normalized DifferenceCloud Index).

[0010] Step 4, through the preset NDCI Threshold identification area: Is there cloud cover?

[0011] This invention presents a cloud detection method for spaceborne lidar based on the difference in apparent reflectivity between signal and noise. It eliminates the need for prior surface reflectivity data and complete backscattering profile data, directly utilizing the lidar surface profile reflection signal and background noise rate to calculate their apparent reflectivity. The difference in apparent reflectivity is then used to detect clouds. This effectively solves the problems of daytime misjudgment and reliance on prior information in existing ICESat-2 cloud detection methods, providing a highly efficient cloud detection technology path for spaceborne lidar. Attached Figure Description

[0012] Figure 1 This is a schematic diagram illustrating the principle of the invention. When clouds are present, the apparent reflectivity of the laser... R l It will be much smaller than the apparent reflectance of the background. R b And there is no apparent reflectivity of laser in the cloud. R l and background apparent reflectance R b They are basically the same size, and the difference between them can be used to detect clouds.

[0013] Figure 2 These are contour plots of laser apparent reflectance, background apparent reflectance, and normalized cloud difference index (NDI) according to the present invention. (a), (b), and (c) are simulated contour plots of laser apparent reflectance, background apparent reflectance, and NDI under different surface reflectances and different cloud optical thicknesses, assuming a typical marine aerosol optical thickness of 0.1. (d) is a contour plot of the NDI under different solar altitude angles and different cloud optical thicknesses, assuming a surface reflectance of 0.2.

[0014] Figure 3 These are the results of algorithm recognition and cloud map verification. (a) shows the ICESat-2 trajectory near Malaysia and the corresponding Sentinel-2 cloud map. Blue dots represent cloudless areas identified using ATL03 data, and red dots represent... NDCI The area was identified as having clouds, and this is where the cloud cover is used. NDCI The threshold is 0.8. (b) is a magnified view of the orange box area in (a). (c) is the ICESat-2 point cloud image corresponding to the trajectory in (b), where gray point clouds represent cloudless areas, red point clouds represent areas with photons, blue point clouds represent signal photons, and the yellow background represents the official ICESat-2 cloud product. layer_flag The area was detected as having clouds. Detailed Implementation

[0015] like Figure 1As shown, the apparent surface reflectance after atmospheric attenuation can be obtained from laser profile data, which is defined in this paper as laser apparent reflectance. R l Using the sun as the radiation source, the apparent reflectance calculated from background noise is defined in this paper as background apparent reflectance. R b It encompasses the combined effects of the atmosphere and ground features. Under clean atmospheric conditions, R b and R l The values ​​will be relatively close, but as the optical thickness of the cloud increases, due to the higher reflectivity of the cloud top, R b It usually manifests as an increase in size, while R l The difference will be significantly reduced, and the relative difference between the two provides a theoretical possibility for cloud detection.

[0016] Based on the above facts, the core concept of the spaceborne lidar cloud detection method based on signal and background noise fusion in this invention is that, in addition to being used as an active remote sensing device, the noise data of the spaceborne lidar also contains important surface radiation information, and can be regarded as a high-sensitivity narrow-band radiometer with sunlight as the radiation source. During data processing, the solar background noise of the spaceborne lidar itself should not be regarded as interference, but rather as a unique passive detection data, thereby achieving self-fusion of active and passive detection. In this invention, this idea comprehensively utilizes the difference between the apparent reflectivity of laser in active detection and the apparent reflectivity of passive detection to achieve the detection of signals affected by clouds. Therefore, this invention can also be regarded as a spaceborne lidar cloud detection method based on the difference in apparent reflectivity between signal and noise.

[0017] like Figure 2 The figure shows the apparent reflectivity of laser under different surface reflectivities and cloud optical thicknesses. R l and background apparent reflectance R b and normalized cloud difference index NDCI Simulated contour map. From Figure 2 As can be seen in (a), when the optical thickness of the cloud is only 0.1, R l It is approximately half the reflectivity of actual ground features, at which point the two-way atmospheric transmittance is approximately 0.5, and this transmittance increases with increasing cloud optical thickness. R l Gradually decreasing. From Figure 2 As can be seen in (b), when the cloud optical thickness is 0.10, R bCompared to the actual surface reflectivity, which is low, cloud optical thickness increases. R b It gradually increases and eventually reaches over 0.90, essentially matching the reflectivity of snow. From Figure 2 As can be seen in (c), in the high-reflectivity surface region where the cloud optical thickness is 0.10, theoretically, when R l Greater than R b At this time NDCI The value may be negative, as the optical thickness of the cloud increases. NDCI As the cloud optical thickness gradually increases, when it exceeds 1.1, almost the entire surface... NDCI The values ​​are all greater than 0.9, and the calculated values ​​are... R b Greater than R l More than 19 times. From Figure 2 As can be seen in (d), when the solar altitude angle changes from 20° to 80°, NDCI The value remains basically unchanged, indicating NDCI The value is not sensitive to changes in the solar altitude angle. (Comparison) Figure 2 In (a), (b), and (c), it can also be observed that when the cloud optical thickness varies between 0.1 and 1, R l and R b The changes depend not only on the optical thickness of the cloud but also on the actual reflectivity of the Earth's surface, relatively speaking. NDCI The response to clouds is more direct. Theoretical simulations have taken into account areas with low reflectivity. NDCI Choosing a threshold that is too small may lead to false detections; therefore, we will use... NDCI When the threshold is greater than 0.8, it is judged as cloudy. R b Greater than R l Nine times that.

[0018] Taking ICESat-2 data as an example, the apparent reflectivity of laser is calculated as follows:

[0019]

[0020] In the formula, N p The number of signal photons, z The distance from the satellite to the Earth's surface. E To emit pulse energy, D c Dead zone correction factor F This refers to the instrument calibration coefficients for the lidar.h Let be Planck's constant. v The optical frequency corresponding to the laser. η r For the sake of receiving system efficiency, η q For quantum efficiency.

[0021] Background apparent reflectance R b The calculation method can be expressed as:

[0022]

[0023] In the formula, f noise For noise rate, N0 λ This refers to the solar spectral irradiance outside the atmosphere, for example, 1.83 W / (m²) at 532 nm. 2 ·nm),Δ λ For filter bandwidth, θ r It receives half the field of view. A r To receive the telescope area, θ s This is the solar zenith angle.

[0024] The normalized cloud difference index is calculated as follows:

[0025] .

[0026] Instance verification

[0027] Figure 3 This is a comparison of satellite cloud images using ICESat-2 data and Sentinel-2 data, based on the algorithm of this invention. Figure 3 (a) shows the trajectory of the ICESat-2 satellite flying from south to north (in the direction of the yellow arrow) over the vicinity of Malaysia and the corresponding full-color image from Sentinel-2. ICESat-2 flew over this region between 02:58:00 and 02:58:12 UTC on June 12, 2020, while the Sentinel-2 image was captured at 02:57:06 UTC on June 12, 2020, a difference of less than 2 minutes. Red dots represent areas identified as cloudy by the algorithm of this invention, while blue dots represent areas detected as cloudless by the algorithm. It can be seen that the algorithm's identification results are basically consistent with the satellite cloud image. Figure 3 In (b), it is a magnified view of the orange rectangular area in (a). It can be seen that the algorithm effectively identified the signal photon point cloud affected by the cloud, and the boundary of the point cloud is consistent with the cloud boundary presented in the optical image. Figure 3(c) shows the point cloud data corresponding to the trajectory in (b). The gray point cloud represents the original ATL03 photon point cloud, the blue point cloud represents the signal photons, and photons in areas affected by the cloud are marked in red. The area within the yellow rectangle is... layer_ flag Areas marked as cloudy. (Comparison) Figure 3 In the regions corresponding to the green dashed boxes 1 and 2 between (b) and (c), the algorithm in this paper can effectively identify the boundaries of clouds. layer_flag Because its resolution is only 280m, it is difficult to effectively distinguish the boundaries of clouds.

[0028] To investigate the algorithm's performance under different land surface types, this invention selected four representative land surface regions for study. Among them, the Greenland region (62°N-65°N, 49°W-43°W) and the Sahara Desert region (18.5°N-20.9°N, 10.2°W-1.6°W) are areas with relatively stable surface reflectance. Typical marine and nearshore land surface regions were also selected, including eastern China (29°N-36°N, 119°E-123°E) and the surrounding southern region of China (1°N-26°N, 113°E-115°E). The data used in this study are ATL03 and ATL09 data products from the ICESat-2 satellite passing through the above four regions from January 2019 to May 2024. The final effective data volume for each study region was as follows: Greenland region 226 tracks, Sahara Desert region 469 tracks, Eastern China region 304 tracks, and Southern China surrounding region 259 tracks.

[0029] In the analysis of cloud detection results from hundreds of tracks of data collected in four regions—Eastern China, Southern China, Greenland, and the Sahara Desert—the official ICESat-2 cloud product was used. layer_flag When used as the truth value, the cloud recognition results with a fixed threshold of 0.8 are shown in Table 1. Table 1 presents the cloud recognition results of the proposed method in four study regions. NDCI The detection threshold is fixed at 0.80. Comparing the confusion matrices of the detection results for the four study areas in the table, it can be seen that the discrimination results of the algorithm in this paper... F The values ​​are all greater than 0.80, indicating the best cloud recognition performance in the Greenland region. F The value is approximately 0.95, which is the worst performing value in the Sahara Desert region. F The I value is approximately 0.80, indicating good overall recognition results. Using a single fixed threshold may not accurately describe the characteristics of different land cover types. Table 2 shows the cloud recognition results when using the maximum Jordan Index and corresponding to the optimal threshold. It can be seen that after changing the threshold, the evaluation indicators for each region improved compared to Table 1, particularly in the Greenland region. F The value increased to around 0.98 in the Sahara Desert region. FThe value increased to around 0.81.

[0030] Table 1. Cloud recognition results with a fixed threshold of 0.80.

[0031]

[0032] Table 2 Classification results at the optimal threshold

[0033]

[0034] Therefore, without loss of generality, the present invention NDCI The threshold can be between 0.58 and 0.85.

Claims

1. A cloud detection method for spaceborne lidar based on signal and background noise fusion, characterized in that: Includes the following steps: Step 1: Obtain the parameters of the lidar system, including laser emission energy, lidar system optical efficiency, receiving system quantum efficiency, filter bandwidth, instrument calibration coefficient, lidar flight altitude, and also obtain information on the sun's position and spectral irradiance, including the sun's altitude angle and the corresponding band of extra-atmospheric solar irradiance. Step 2: Obtain the surface profile signal magnitude and background noise rate; Step 3: Calculate the apparent reflectivity of the laser from the surface profile signals. R l Calculate the apparent reflectance of the background from the background noise rate. R b And construct a normalized cloud difference index to characterize the relative difference between the two. NDCI ; Step 4, through the preset NDCI Threshold identification area: Is there cloud cover? 2. The method as described in claim 1, characterized in that... Normalized cloud difference index NDCI The calculation method is as follows: 。 3. The method as described in claim 1, characterized in that step 4 is as follows. NDCI The threshold is between 0.58 and 0.

85.

4. The method according to any one of claims 1-3, characterized in that: When using ICESat-2 data, the laser apparent reflectivity R l Background apparent reflectance R b The calculation method is as follows: In the formula, N p The number of signal photons, z The distance from the satellite to the Earth's surface. E To emit pulse energy, D c Dead zone correction factor F This refers to the instrument calibration coefficients for the lidar. h Let be Planck's constant. v The optical frequency corresponding to the laser. η r For the sake of receiving system efficiency, η q For quantum efficiency; In the formula, f noise For noise rate, Δ represents the solar spectral irradiance outside the atmosphere. λ For filter bandwidth, θ r It receives half the field of view. A r To receive the telescope area, θ s This is the solar zenith angle.

Images