A polarization remote sensing multi-layer cloud identification method based on radiation simulation data

Abstract

The application provides a polarization remote sensing multi-layer cloud identification method based on radiation simulation data. The process comprises the following steps: obtaining accurate atmospheric top radiation simulation data as a training sample of machine learning based on a vector radiation transfer model; building a multi-layer cloud identification algorithm based on a K-nearest neighbor model, training polarization remote sensing simulation data by using the K-nearest neighbor model based on multi-dimensional information such as intensity radiation, polarization radiation and multi-angle, and realizing identification of multi-layer clouds. The application can be used for multi-layer cloud identification of polarization remote sensing cloud pixels, and provides support for data application of polarization remote sensing.

Description

(I) Technical Field

[0001] This invention relates to a polarization remote sensing multi-layer cloud identification method based on radiation simulation data, which can be used for multi-layer cloud identification of polarization remote sensing cloud pixels, providing support for the application of polarization remote sensing data. (II) Background Technology

[0002] The atmospheric radiation effects of single-layer and multi-layer clouds differ significantly, and accurately identifying single-layer and multi-layer cloud information is crucial for understanding the role of clouds in the radiation balance of the land-atmosphere system. Because the intuitive spectral characteristics of single-layer and multi-layer clouds are not clearly distinguishable, the accuracy of passive remote sensing multi-layer cloud identification is low. Polarization remote sensing can acquire not only intensity radiation information but also polarization radiation information, thus obtaining more information that can characterize cloud and surface properties. Machine learning has powerful learning capabilities, capable of fully exploring the potential patterns and features in data, providing a basis for data classification decisions. Using simulated data as training data, a machine learning model is used for polarization remote sensing multi-layer cloud identification to address the problems existing in multi-layer cloud identification.

[0003] Multi-layered clouds have a significant impact on radiative efficiency and cloud characteristic parameter inversion. Scientists have developed corresponding multi-layered cloud identification methods based on the channel design characteristics of various passive remote sensing instruments. The MODIS multi-layered cloud algorithm uses near-infrared water vapor channels (0.94 μm) and CO2 channels (13.3, 13.6, 13.8, and 14.2 μm) to identify multi-layered clouds by inverting water vapor on independent clouds. Compared with the active remote sensing instrument CALIOP, this method often misclassifies thin multi-layered clouds as single-layered clouds and thick single-layered clouds as multi-layered clouds. Pavolon and Heidinger designed a multi-channel multi-layered cloud identification algorithm using reflectivity information from the 0.65, 1.38, and 1.6 μm channels and brightness temperature information from the 11 and 12 μm channels. This algorithm has been used for cloud detection in AVHRR and VIIRS. MODIS has also adopted this algorithm and used it in the production of multi-layered cloud products in the MODIS C6. Based on the characteristics of the next-generation imager VIIRS, Wang et al. designed a novel multi-layer cloud identification algorithm using three short-wave infrared (1.38, 1.6, and 2.25 μm) and two long-wave infrared (8.5 and 11 μm) channels. Ferlay et al. used cloud vertical distribution data from the active detector millimeter-wave radar (CPR) and lidar (CALIOP) on the A-Train satellite train, along with POLDER3 L2-level cloud product data, as training datasets, and employed a decision tree algorithm to identify multi-layer clouds in POLDER3.

[0004] This invention discloses a polarization remote sensing multi-layer cloud identification method based on radiation simulation data. This method trains polarization remote sensing simulation data with a machine learning model to mine more deep features that can be used for multi-layer cloud identification, achieving a more ideal multi-layer cloud identification effect. (III) Summary of the Invention

[0005] The purpose of this invention is to provide a polarization remote sensing multi-layer cloud identification method based on radiation simulation data that achieves a relatively ideal multi-layer cloud identification effect and has low training data acquisition cost.

[0006] The objective of this invention is achieved through the following technical means:

[0007] A method for identifying multilayer clouds using polarization remote sensing based on radiation simulation data includes:

[0008] Step 1: Use the vector radiative transfer model to obtain simulation data of the radiation characteristics of the top of the atmosphere under atmospheric conditions including single-layer clouds and multi-layer clouds;

[0009] Step 2: Preprocess the simulation data of the top-of-atmosphere radiation characteristics to construct multi-layer cloud recognition training samples;

[0010] Step 3: Build a polarization remote sensing multilayer cloud identification algorithm based on the K-nearest neighbor model;

[0011] Step 4: Train the model from Step 3. Based on the training results, remove information that negatively affects multi-layer cloud recognition and determine the best multi-layer cloud training samples.

[0012] Step 5: Optimize the multi-layer cloud recognition algorithm and train the model based on the best multi-layer cloud training samples from Step 4.

[0013] Step 6: Based on the model trained in Step 5, perform multi-layer cloud identification on the measured data, and compare the multi-layer cloud identification results with the cloud vertical distribution data to verify the effectiveness of the algorithm.

[0014] Furthermore, the flowchart for obtaining the simulation data of the top-of-atmosphere radiation characteristics in step 1 is as follows: Figure 2 As shown, the parameters of the vector radiative transfer model are set, and the top atmospheric radiation characteristics are simulated using vector radiative transfer software to obtain simulation data of the top atmospheric radiation characteristics.

[0015] The beneficial effects of this invention are as follows: This invention is a polarization remote sensing multi-layer cloud identification method based on radiation simulation data. By using multi-dimensional information such as intensity radiation, polarization radiation, and multiple angles, machine learning methods are used to mine more deep features that can be used for multi-layer cloud identification, thereby achieving the identification of multi-layer clouds and providing more accurate cloud parameter data for the reference of polarization remote sensing data. (iv) Description of the attached drawings

[0016] Figure 1 This is a flowchart of a polarization remote sensing multilayer cloud identification method based on radiation simulation data.

[0017] Figure 2 This is a flowchart for obtaining simulation data of the top-of-atmosphere radiation characteristics. (V) Detailed Implementation

[0018] The present invention will be further illustrated below with reference to specific embodiments.

[0019] like Figure 1 As shown, this invention proposes a polarization remote sensing multi-layer cloud identification method based on radiation simulation data, comprising:

[0020] Step 1: Based on the application requirements of multi-layer cloud identification, accurately set the input parameters and use vector radiative transfer software to simulate and calculate the radiation characteristics of the top of the atmosphere, so as to provide reliable training data for multi-layer cloud identification based on machine learning.

[0021] Step 2: Preprocess the radiation simulation data of the atmospheric top under atmospheric conditions including single-layer cloud and multi-layer cloud obtained in Step 1 to construct multi-layer cloud training and recognition samples.

[0022] Step 3: Build a polarization remote sensing multilayer cloud identification algorithm based on the K-nearest neighbor model;

[0023] Step 4: Train the model from Step 3 based on the training samples from Step 2. According to the training results, remove information that has a negative impact on multi-layer cloud recognition, determine the best multi-layer cloud training data, and improve the accuracy of multi-layer cloud recognition.

[0024] Step 5: Based on the optimal multi-layer cloud training data determined in Step 4, optimize the multi-layer cloud recognition algorithm and train the model;

[0025] Step 6: Based on the test data of the model trained in Step 5, identify the multi-layered clouds, compare the identification results with the cloud vertical distribution data, and verify the effectiveness of the algorithm.

[0026] In this example, the use of radiation simulation data as training data samples for multi-layer cloud recognition is more cost-effective than measured data. The number of training samples can be arbitrarily large, and it covers all possible data.

[0027] Input parameters for the vector radiative transfer model:

[0028] Atmospheric molecular parameters: Determine the spectral response function of the instrument's spectral range and set the atmospheric mode according to application requirements.

[0029] Cloud parameters: The radiation contribution of clouds to the top of the atmosphere depends on the scattering characteristics of the clouds. Water clouds are generally equivalent to spherical particles, and their scattering characteristics are calculated using Mie scattering theory. Ice clouds are composed of ice crystal particles of different shapes and sizes, and their scattering characteristics are mainly calculated using methods such as the T-matrix method, the finite difference time-domain method, the discrete dipole approximation method, the geometric optics method, and the improved geometric optics method. The ice cloud scattering characteristic data in this invention comes from the publicly available ice cloud scattering database of Professor Yang Ping's team at Texas A&M University. The scattering characteristics of clouds are mainly determined by factors such as the optical thickness (COD) of the cloud, the effective particle radius of the cloud, and the shape of the cloud particles. In this invention, multi-layered clouds are defined as upper ice clouds and lower water clouds.

[0030] Aerosol parameters: The influence of aerosols on the top-of-atmosphere radiation characteristics is mainly determined by their scattering properties. Aerosol particles include both spherical and non-spherical particles. Mie scattering theory is generally used to calculate the scattering characteristics of spherical particles, while the T-matrix is ​​used to calculate the scattering characteristics of non-spherical particles. The scattering characteristics of aerosols depend on factors such as the complex refractive index, particle size distribution, and aerosol optical depth (AOD).

[0031] K-Nearest Neighbor Model:

[0032] The core of the K-Nearest Neighbors model is to use distance metrics to obtain the k nearest points to the target point, and then determine the classification of the target point according to the classification decision rules.

[0033] Distance metric definition:

[0034]

[0035] In the formula, x i and y i R is an n-dimensional real vector space n Two points on the same surface. When p=2, it is a common Euclidean distance.

[0036] Choosing the k value: Select a smaller k value and choose the optimal k value through cross-validation.

[0037] The classification decision rule uses majority voting, where the class of the input instance is determined by the majority class of its K nearest training instances.

[0038] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to specific embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention. Technical, shape, and structural parts not described in detail in this invention are all well-known technologies.

Claims

1. A method for identifying multi-layer clouds using polarization remote sensing based on radiation simulation data, characterized in that, The specific process of this method includes the following steps: Step 1: Simulate the radiation characteristics of the top of the atmosphere based on the vector radiative transfer model to obtain radiation simulation data of the top of the atmosphere under atmospheric conditions including single-layer clouds and multi-layer clouds. Step 2: Preprocess the data from Step 1 to construct a multi-layer cloud training dataset that meets the input requirements of machine learning networks; Step 3: Build a polarization remote sensing multilayer cloud identification algorithm based on the K-nearest neighbor model; Step 4: Train the model from Step 3 based on the sample set from Step 2. According to the training results, remove information that has a negative impact on multi-layer cloud recognition, determine the best multi-layer cloud training data, and improve the accuracy of multi-layer cloud recognition. Step 5: Based on the optimal multi-layer cloud training data determined in Step 4, optimize the multi-layer cloud recognition algorithm and train the model to determine the optimal parameters. Step 6: Based on the model test data from Step 5, identify the multi-layered clouds from polarization remote sensing, and compare the identification results with the cloud vertical distribution data to verify the effectiveness of the algorithm. The input parameters of the vector radiative transfer model are atmospheric molecular parameters, cloud parameters, and aerosol parameters, respectively. The core of the K-Nearest Neighbors model is to use distance metrics to obtain the k nearest points to the target point, and then determine the classification of the target point according to the classification decision rules. Distance metric definition: In the formula, x i and y i R is an n-dimensional real vector space n Two points on the same surface; when p=2, it is the common Euclidean distance; Selection of k value: Choose a smaller k value and select the optimal k value through cross-validation; The classification decision rule uses majority voting, where the class of the input instance is determined by the majority class of its K nearest training instances.

2. The polarization remote sensing multi-layer cloud identification method based on radiation simulation data according to claim 1, characterized in that, The multi-layered cloud is defined as an upper layer of ice cloud and a lower layer of water cloud.

Images