Atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data

Abstract

The invention discloses an atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data. The atmospheric temperature and humidity profiler inversion algorithm comprises the following steps: step 1, carrying out space-time matching preprocessing on the reanalysis data; step 2, brightness temperature forward modeling; step 3, completing neural network training; step 4, outputting a profile; and step 5, business packaging is carried out. According to the inversion algorithm disclosed by the invention, a sounding sample is replaced by high-resolution reanalysis data, and cross-resolution feature matching and dynamic deviation correction are combined, so that the technical problem that the temperature and humidity inversion precision is reduced due to the fact that an atmospheric temperature and humidity profiler depends on a historical sounding sample in an area without sounding data is solved; and high-precision atmospheric temperature, humidity and water vapor vertical distribution parameters can be obtained without depending on sounding data.

Description

Technical Field

[0001] The present invention belongs to the field of atmospheric remote sensing technology, and particularly relates to an atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data in this field. The invention is a meteorological parameter inversion method based on reanalysis data instead of sounding data, and is suitable for real-time data processing of microwave radiometers in areas without sounding observations. Background Art

[0002] The inversion algorithm for atmospheric temperature and humidity profiles uses sounding data as prior training samples. A neural network model is established using sounding profiles and microwave brightness temperature data. The high spatial resolution of sounding data is used to improve the accuracy of temperature and humidity inversion. Similarly, the dependence of the inversion results on a single sounding station can be reduced by dynamically adjusting the weight of the brightness temperature channel. This method performs well in areas with dense meteorological observation stations, but its algorithm requires high temporal and spatial continuity of the sounding data.

[0003] Atmospheric temperature and humidity profilers, core instruments in atmospheric remote sensing, rely on sounding data for their traditional inversion algorithms. These instruments rely on sounding data for calibration in plateau regions, requiring ground observations to correct for low-altitude biases when inverting humidity. The temporal and spatial discontinuities of sounding data limit the algorithm's universality. Sounding observation sites are primarily located over land, lacking real-time sounding data support in oceans, plateaus, and remote areas. Sounding data coverage is insufficient, and inversion accuracy significantly decreases in areas without soundings.

[0004] The above analysis shows that the existing atmospheric temperature and humidity profile inversion method has technical bottlenecks in areas without sounding data. It is urgent to develop an inversion method based on reanalysis data to break the reliance on single data and make up for the shortcomings of modeling in areas without sounding data. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an atmospheric temperature and humidity profiler inversion algorithm driven by FNL reanalysis data.

[0006] The present invention adopts the following technical solutions:

[0007] An improved atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data includes the following steps:

[0008] Step 1: Reanalyze data for spatiotemporal matching preprocessing:

[0009] Based on the FNL reanalysis data, combined with the latitude, longitude and altitude information of the target stations, standardized inputs were constructed through spatial interpolation, terrain correction and time synchronization;

[0010] Step 2, brightness temperature forward modeling:

[0011] Based on the temperature, humidity, and pressure information of the reanalysis data, the brightness temperature data for the corresponding period is calculated and the brightness temperature forward modeling is completed;

[0012] Step 3: Complete the neural network training:

[0013] Adopting BP network structure, inputting training data, setting the number of neurons and transfer function of the network, training the neural network, and establishing the inversion model of atmospheric temperature and humidity profiles;

[0014] Step 4, profile output:

[0015] Based on the neural network training and combined with the real-time measurement of atmospheric brightness temperature, temperature, humidity and pressure data, the tropospheric atmospheric parameter profile is output;

[0016] Step 5: Business packaging:

[0017] Output schema file.

[0018] Furthermore, in step 1:

[0019] The following parameters of the FNL reanalysis data were obtained with a temporal resolution of 6 hours, a spatial resolution of 1° × 1°, and 26 vertical layers: temperature, specific humidity, geopotential height, and surface pressure, and then matched with the target station information;

[0020] Get the average terrain height H of the FNL reanalysis data grid FNL The actual altitude H of the target site site , if |H FNL -H site |>50m, correct the temperature according to the vertical temperature lapse rate:

[0021] T site =T FNL +γ(H FNL -H site )

[0022] In the above formula, T site is the corrected temperature of the site, T FNL The temperature data are from the FNL reanalysis data, and the temperature lapse rate is γ = 6.5℃ / km.

[0023] Furthermore, in step 2:

[0024] Substituting the FNL reanalysis data into the following formula yields the true brightness temperature of the atmosphere:

[0025]

[0026] In the above formula, T AP (v,θ) represents the atmospheric brightness temperature at frequency ν with an angle θ, k ν(z) is the atmospheric absorption coefficient at frequency ν at height z, T EXTRA (ν) represents the cosmic radiation before entering the Earth's atmosphere, τ ν is the opacity of the atmosphere at the zenith, T(z) is the temperature of the atmosphere at height z, τ ν (0,z) is the optical thickness of the atmosphere in the zenith direction between the ground and the height z.

[0027] Furthermore, in step 3:

[0028] The ground pressure, atmospheric temperature, water vapor density and atmospheric brightness temperature are used as input parameters, and the atmospheric temperature and humidity profile is used as output parameter. The length L of the input vector X is determined by the input parameters, and the length M of the output vector Y represents the total number of vertical layers of the atmosphere. The output of any single neuron is:

[0029]

[0030] In the above formula: f is the output value of the neuron, ω is the input weight matrix of the neuron, p is the input value of the neuron, b is the deviation of the neuron, and S uses the logarithmic tangent function:

[0031] Furthermore, in step 5, a model file of 100 layers of vertical profiles is outputted from 0 to 10,000 meters, with each layer being 100 meters.

[0032] The beneficial effects of the present invention are:

[0033] The inversion algorithm disclosed in the present invention replaces sounding samples with high-resolution reanalysis data, combines cross-resolution feature matching with dynamic bias correction, and solves the technical problem that the temperature and humidity inversion accuracy of the atmospheric temperature and humidity profiler decreases due to reliance on historical sounding samples in areas without sounding data. High-precision atmospheric temperature, humidity and water vapor vertical distribution parameters can be obtained without relying on sounding data.

[0034] The inversion algorithm disclosed in the present invention overcomes the strong dependence of traditional atmospheric temperature and humidity profile inversion methods on sounding data through breakthroughs in modeling technology, and realizes global indiscriminate modeling. Compared with the existing technology, its core advantage is that it gets rid of the dependence on sounding data. The existing method needs to rely on sounding observation data within 50 kilometers around the station as a training and verification benchmark, resulting in areas without sounding stations (such as plateaus, oceans, and remote mountainous areas) cannot be modeled. The inversion algorithm of the present invention achieves modeling of areas without sounding through global coverage of FNL reanalysis data modeling. This breakthrough enables traditional blind areas such as plateaus, oceans, and polar regions to obtain high-precision data comparable to sounding verification, which can provide key technical support for global climate change research and disaster weather warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic flow chart of the inversion algorithm disclosed in the present invention;

[0036] Figure 2 It is a three-layer fully connected BP network structure diagram;

[0037] Figure 3 It is a flow chart of BP network training and simulation;

[0038] Figure 4 This is a flowchart for implementing the inversion of tropospheric atmospheric parameter profiles;

[0039] Figure 5 This is a comparison chart of the simulation accuracy of the temperature and humidity profile inversion in January;

[0040] Figure 6 This is a comparison chart of the simulation accuracy of the temperature and humidity profile inversion in July. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] Example 1. This example discloses an atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data. To address the problem of insufficient historical sounding data in areas without sounding data, a spatial accurate matching model between reanalysis data and station observation data is constructed. The reanalysis data at the target longitude and latitude are used as historical data for the inversion model. The model is forward modeled and inverted for verification. The inversion model uses a BP neural network to achieve stable output of high-resolution vertical temperature and humidity profiles. Figure 1 As shown, the following steps are included:

[0043] Step 1: Reanalyze data for spatiotemporal matching preprocessing:

[0044] Based on the FNL reanalysis data, combined with the latitude, longitude and altitude information of the target stations, standardized inputs were constructed through spatial interpolation, terrain correction and time synchronization;

[0045] The following parameters of the FNL reanalysis data were obtained with a temporal resolution of 6 hours, a spatial resolution of 1° × 1°, and 26 vertical layers: temperature (T), specific humidity (q), geopotential height (H), and surface pressure (Ps), and then matched with the target station information (latitude, longitude, and altitude);

[0046] Get the average terrain height H of the FNL reanalysis data grid FNL The actual altitude H of the target site site , if |H FNL -Hsite |>50m, correct the temperature according to the vertical temperature lapse rate:

[0047] T site =T FNL +γ(H FNL -H site )

[0048] In the above formula, T site is the corrected temperature of the site, T FNL The temperature data are from the FNL reanalysis data, and the temperature lapse rate is γ = 6.5℃ / km.

[0049] Step 2, brightness temperature forward modeling:

[0050] Based on the temperature, humidity, and pressure information of the reanalysis data, the brightness temperature data for the corresponding period is calculated and the brightness temperature forward modeling is completed;

[0051] The atmospheric brightness temperature is the amount of radiation radiated by the atmosphere onto the main lobe of the atmospheric temperature and humidity profiler receiving antenna. For a downlink link (i.e., atmospheric temperature and humidity profiler) with no scattering (or with a scattering coefficient much smaller than the absorption coefficient), the received brightness temperature is:

[0052]

[0053] In the above formula, T AP (v,θ) represents the atmospheric brightness temperature at frequency ν with an angle θ, k ν (z) is the atmospheric absorption coefficient at frequency ν at height z, T EXTRA (ν) represents the cosmic radiation before entering the Earth's atmosphere, τ ν is the opacity of the atmosphere at the zenith. When f≥10GHz, T EXTRA (ν) is about 2.7K, and the attenuation factor through the atmosphere The attenuation of can be ignored. T(z) is the temperature of the atmosphere at height z, τ ν (0, z) is the optical thickness of the atmosphere in the zenith direction between the ground and the height z. Substituting the reanalysis data into the above formula yields the true brightness temperature of the atmosphere, providing forward results for model inversion.

[0054] Step 3: Complete the neural network training:

[0055] Adopting BP network structure, inputting training data, setting the number of neurons and transfer function of the network, training the neural network, and establishing the inversion model of atmospheric temperature and humidity profiles;

[0056] According to the brightness temperature observation value and detection time obtained in step 2, the training samples and test samples are screened. Figure 2The three-layer fully connected BP network structure shown in the figure uses a back-propagation algorithm to determine the network's weights and biases during training. This algorithm continuously adjusts the network's weights and biases to minimize the discrepancy between the output vector calculated from the network's input vector and the actual training target output vector. Other network parameters are determined through repeated trial and error testing.

[0057] like Figure 3 As shown in the figure, the ground pressure, atmospheric temperature, water vapor density and atmospheric brightness temperature are used as input parameters, and the atmospheric temperature and humidity profile is used as the output parameter. The length L of the input vector X is determined by the input parameters, and the length M of the output vector Y represents the total number of vertical layers of the atmosphere. The output of any single neuron is:

[0058]

[0059] In the above formula: f is the output value of the neuron, ω is the input weight matrix of the neuron, p is the input value of the neuron, and b is the deviation of the neuron, also known as the threshold vector. S uses the logarithmic tangent function:

[0060] Step 4, profile output:

[0061] like Figure 4 As shown, according to the neural network training, combined with the real-time measurement of atmospheric brightness temperature and temperature, humidity and pressure data, the tropospheric atmospheric parameter profile is output;

[0062] Based on the neural network training results obtained in step 3, combined with the atmospheric brightness temperature and temperature, humidity, and pressure data measured by the atmospheric temperature and humidity profiler, the tropospheric atmospheric temperature profile and humidity profile are output.

[0063] Step 5: Business packaging:

[0064] From 0 to 10,000 meters, 100 layers of vertical profile pattern files are output every 100 meters for direct use by the software to improve the calculation speed.

[0065] A location 300 km from the coastline at 114°E, 19°N was selected, and representative months of the year (January and July) were selected to compare the simulation results with the reanalysis data. Figure 5 、 Figure 6 As shown, the horizontal axis is the number of samples and the vertical axis is the number of layers of inversion height.

[0066] According to statistics, the inversion algorithm disclosed in the present invention enables the temperature and humidity inversion accuracy of the atmospheric temperature and humidity profiler in areas without soundings to reach the level of sounding verification (temperature error 1.35K, humidity error 15.7%). The inversion results show high consistency with the reanalysis data, and the temporal and spatial variation characteristics are also relatively consistent.

[0067] From the above test results, it can be seen that the neural network model established using reanalysis data has high reliability and can provide reliable data support for disastrous weather warnings.

Claims

1. An atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data, characterized by: The steps include: Step 1: Reanalyze data for spatiotemporal matching preprocessing: Based on the FNL reanalysis data, combined with the latitude, longitude and altitude information of the target stations, standardized inputs were constructed through spatial interpolation, terrain correction and time synchronization; Step 2, brightness temperature forward modeling: Based on the temperature, humidity, and pressure information of the reanalysis data, the brightness temperature data for the corresponding period is calculated and the brightness temperature forward modeling is completed; Step 3: Complete the neural network training: Adopting BP network structure, inputting training data, setting the number of neurons and transfer function of the network, training the neural network, and establishing the inversion model of atmospheric temperature and humidity profiles; Step 4, profile output: Based on the neural network training and combined with the real-time measurement of atmospheric brightness temperature, temperature, humidity and pressure data, the tropospheric atmospheric parameter profile is output; Step 5: Business packaging: Output schema file.

2. The atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data according to claim 1 is characterized in that: In step 1: The following parameters of the FNL reanalysis data were obtained with a temporal resolution of 6 hours, a spatial resolution of 1° × 1°, and 26 vertical layers: temperature, specific humidity, geopotential height, and surface pressure, and then matched with the target station information; Get the average terrain height H of the FNL reanalysis data grid FNL The actual altitude H of the target site site , if |H FNL -H site |>50m, correct the temperature according to the vertical temperature lapse rate: T site =T FNL +γ(H FNL -H site ) In the above formula, T site is the corrected temperature of the site, T FNL The temperature data are from the FNL reanalysis data, and the temperature lapse rate is γ = 6.5℃ / km.

3. The atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data according to claim 1 is characterized in that: In step 2: Substituting the FNL reanalysis data into the following formula yields the true brightness temperature of the atmosphere: In the above formula, T AP (v,θ) represents the atmospheric brightness temperature at frequency ν with an angle θ, k ν (z) is the atmospheric absorption coefficient at frequency ν at height z, T EXTRA (ν) represents the cosmic radiation before entering the Earth's atmosphere, τ ν is the opacity of the atmosphere at the zenith, T(z) is the temperature of the atmosphere at height z, τ ν (0,z) is the optical thickness of the atmosphere in the zenith direction between the ground and the height z.

4. The atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data according to claim 1 is characterized in that: In step 3: The ground pressure, atmospheric temperature, water vapor density and atmospheric brightness temperature are used as input parameters, and the atmospheric temperature and humidity profile is used as output parameter. The length L of the input vector X is determined by the input parameters, and the length M of the output vector Y represents the total number of vertical layers of the atmosphere. The output of any single neuron is: In the above formula: f is the output value of the neuron, ω is the input weight matrix of the neuron, p is the input value of the neuron, b is the deviation of the neuron, and S uses the logarithmic tangent function:

5. The atmospheric temperature and humidity profiler inversion algorithm based on FNL reanalysis data according to claim 1 is characterized in that: In step 5, a model file of 100 layers of vertical profiles is output from 0 to 10,000 meters, with each layer being 100 meters.

Images