Regional offshore sea surface temperature inversion method based on LGBM-MLP

Abstract

The invention relates to a regional offshore sea surface temperature inversion method based on LGBM-MLP. The method comprises the following steps: constructing a data set; wherein the data set comprises the microwave radiation data of the earth surface and the atmosphere, the sea surface temperature data and the related data of the reanalyzed ocean atmosphere; the data set is preprocessed; based on the preprocessed data set, training the LGBM-MLP hybrid model to obtain a sea surface temperature inversion model; and carrying out regional offshore sea surface temperature inversion based on the sea surface temperature inversion model. According to the LGBM-MLP-based regional offshore sea surface temperature inversion method, sea air environment information is fully fused, and environment characteristics of brightness temperature data can be better captured, so that the regional offshore sea surface temperature inversion precision is improved.

Description

Technical Field

[0001] This invention relates to the field of marine microwave remote sensing technology, and in particular to a regional nearshore sea surface temperature inversion method based on LGBM-MLP. Background Technology

[0002] Sea surface temperature is a key parameter affecting air-sea coupling and the study of ocean dynamics, providing scientific evidence for marine environmental and ecological monitoring. Methods for observing sea surface temperature include buoys, ship-based measurements, and satellite remote sensing. Among these, satellite remote sensing, with its advantages of high spatiotemporal coverage and high resolution, is widely used in various scientific research and monitoring missions. The AMSR2 (Advanced Microwave Scanning Radiometer), carried on the GCOM-W1 satellite, features multi-band operation and high spatial resolution, enabling the retrieval of various land and ocean surface and atmospheric parameters, providing data support for diverse atmospheric and oceanic data.

[0003] Currently, the most common method for sea surface temperature (SST) retrieval is the physical inversion algorithm. This algorithm analyzes sea surface emissivity, brightness temperature, and atmospheric radiation characteristics, using brightness temperature in a frequency band sensitive to SST, combined with sea surface emissivity models and atmospheric correction models to retrieve SST. While physical methods have clear physical methods and meanings, they require accurate forward models and are computationally complex, with accuracy limited, especially in complex environments. In contrast, empirical methods based on machine learning, through a data-driven approach, can effectively improve retrieval accuracy and computational efficiency. In recent years, machine learning-based SST retrieval methods have become a research hotspot.

[0004] The existing technology is a deep learning-based method for retrieving sea surface temperature. First, training and validation datasets are constructed. An autoencoder is used to adjust for data errors, and a decoding layer connected to a fully connected layer feeds the data into a deep neural network. The fully connected layer is then fine-tuned by minimizing the loss function to retrieve the sea surface temperature.

[0005] The nearshore environment is complex, characterized by river inflow, radio frequency interference, land and sea pollution, and limited data. The 100-130E and 0-45N regions are particularly affected by radio frequency interference and river inflow. Current sea surface temperature (SST) retrieval methods primarily target global or open ocean areas. While deep neural networks can improve the accuracy of SST retrieval compared to physical models, the abundance of data in open ocean areas, coupled with their lower complexity compared to nearshore regions, means that the data relationships fitted by deep learning networks tend to favor open ocean areas. This limits the improvement in accuracy for nearshore SST retrieval and results in insufficient supplementary data. Therefore, it is necessary to establish a SST retrieval model specifically for nearshore areas to improve accuracy and fill the gaps in nearshore temperature data. Summary of the Invention

[0006] The purpose of this invention is to provide a regional nearshore sea surface temperature retrieval method based on LGBM-MLP, to address the current limitations in sea surface temperature data accuracy and insufficient data in this nearshore region due to strong radio frequency interference and numerous estuaries. This invention combines LGBM and MLP network structures to further enhance the ability of machine learning algorithms to retrieve regional nearshore sea surface temperatures.

[0007] To achieve the above objectives, the present invention provides the following solution: A regional nearshore sea surface temperature inversion method based on LGBM-MLP includes: Construct a dataset; wherein the dataset includes: microwave radiation data of the Earth's surface and atmosphere, sea surface temperature data, and reanalysis of ocean-atmosphere related data; The dataset is preprocessed; Based on the preprocessed dataset, the LGBM-MLP hybrid model is trained to obtain the sea surface temperature inversion model; Based on the sea surface temperature inversion model, the nearshore sea surface temperature of the region is inverted.

[0008] Optionally, the microwave radiation data of the Earth's surface and atmosphere includes: Earth's azimuth, Earth's zenith angle, solar azimuth, solar zenith angle, time, latitude and longitude, channel brightness temperature, land-sea markers, and mass markers; The reanalysis of ocean-atmosphere related data includes: time, latitude and longitude, 10m U wind at sea surface, 10m V wind at sea surface, total cloud liquid water, total cloud water vapor, and total average precipitation rate. The sea surface temperature data includes: global satellite in-situ fusion of daily sea surface temperature data and in-situ sea surface temperature quality monitoring data.

[0009] Optionally, preprocessing the dataset includes: The microwave radiation data of the Earth's surface and atmosphere, sea surface temperature data, and related ocean atmospheric data are matched to obtain a matching dataset.

[0010] Optionally, matching the microwave radiation data of the Earth's surface and atmosphere with sea surface temperature data and reanalysis of ocean-atmosphere related data includes: The microwave radiation data of the Earth's surface and atmosphere are matched with the coastline data, and the sea surface temperature data of the nearshore preset area are filtered to obtain nearshore observation data. Based on preset time and space windows, nearshore observation data are matched with reanalysis ocean-atmosphere related data and in-situ sea surface temperature and quality monitoring data; During the matching process, high-quality data were selected based on the quality labels, iQuam quality labels, and land-sea labels in the nearshore observation data. Quality control was also performed based on the AMSR2 L2 sea surface temperature anomalies and the MGDSST sea surface temperature-land sea ice anomalies.

[0011] Optionally, the LGBM-MLP hybrid model adopts a Stacking framework structure, and the LGBM-MLP hybrid model includes: a base model and a meta-model; The basic model includes an LGBM model and an MLP model, and the meta-model includes an MLP model.

[0012] Optionally, when training the LGBM-MLP hybrid model, the inputs to the model are channel brightness temperature, sea surface wind speed and direction, total average precipitation rate, total column water vapor, total column cloud liquid water, and daily sea surface temperature data, and the output of the model is sea surface temperature.

[0013] The beneficial effects of this invention are as follows: The advantages of the LGBM-MLP-based regional nearshore sea surface temperature inversion method of the present invention are as follows: (1) High computational efficiency: Unlike traditional physical inversion methods that rely on complex atmospheric correction and radiative transfer models, the LGBM-MLP-based regional nearshore sea surface temperature inversion method of the present invention simplifies the inversion process through machine learning models, and can efficiently process large-scale data without complex physical modeling. (2) Integration of air-sea environment information: Traditional empirical sea surface temperature inversion methods rely only on brightness temperature data and ignore the importance of air-sea environment information, resulting in low inversion accuracy in areas with complex topography and drastic air-sea environment changes. The LGBM-MLP-based regional nearshore sea surface temperature inversion method of the present invention fully integrates air-sea environment information, and can better capture the environmental characteristics of brightness temperature data, thereby improving the accuracy of regional nearshore sea surface temperature inversion. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a diagram illustrating the architecture of a regional nearshore sea surface temperature inversion model based on LGBM-MLP, according to an embodiment of the present invention. Figure 2 This is a scatter plot showing the comparison between regional nearshore inverted sea surface temperature data and AMSR2 L2 sea surface temperature data from an embodiment of the present invention, and the iQuam sea surface temperature data. Figure 3This is a line graph showing the monthly average sea surface temperature near the coast in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] This embodiment proposes a regional nearshore sea surface temperature inversion method based on LGBM-MLP, including: Construct a dataset; the dataset includes: microwave radiation data of the Earth's surface and atmosphere, sea surface temperature data, and reanalysis of ocean-atmosphere related data; Preprocess the dataset; Based on the preprocessed dataset, the LGBM-MLP hybrid model is trained to obtain the sea surface temperature inversion model; Based on the sea surface temperature inversion model, the nearshore sea surface temperature of the region is inverted.

[0019] Furthermore, microwave radiation data for the Earth's surface and atmosphere includes: Earth's azimuth, Earth's zenith angle, solar azimuth, solar zenith angle, time, latitude and longitude, channel brightness temperature, land-sea markers, and mass markers; Further analysis of ocean-atmosphere related data includes: time, latitude and longitude, 10m U wind at sea surface, 10m V wind at sea surface, total cloud liquid water, total cloud water vapor, and total average precipitation rate; Sea surface temperature data includes: global satellite-in-situ fusion of daily sea surface temperature data and in-situ sea surface temperature quality monitoring data.

[0020] Further preprocessing of the dataset includes: By matching microwave radiation data of the Earth's surface and atmosphere with sea surface temperature data and further analyzing ocean-atmosphere related data, a matching dataset is obtained.

[0021] Furthermore, matching microwave radiation data from the Earth's surface and atmosphere with sea surface temperature data, and reanalyzing ocean-atmosphere related data includes: Microwave radiation data from the Earth's surface and atmosphere are matched with coastline data, and sea surface temperature data for a predetermined nearshore area are filtered to obtain nearshore observation data. Based on preset time and space windows, nearshore observation data are matched with reanalysis ocean-atmosphere related data and in-situ sea surface temperature and quality monitoring data; During the matching process, high-quality data were selected based on the quality labels, iQuam quality labels, and land-sea labels in the nearshore observation data. Quality control was also performed based on the AMSR2 L2 sea surface temperature anomalies and the MGDSST sea surface temperature-land sea ice anomalies.

[0022] Furthermore, the LGBM-MLP hybrid model adopts a stacking framework structure, which includes a base model and a meta-model. The base model consists of an LGBM model and an MLP model, and the meta-model consists of an MLP model.

[0023] Furthermore, when training the LGBM-MLP hybrid model, the inputs to the model are channel brightness temperature, sea surface wind speed and direction, total average precipitation rate, total column water vapor, total column cloud liquid water, and daily sea surface temperature data, and the output of the model is sea surface temperature.

[0024] Specifically, this embodiment provides a regional nearshore sea surface temperature inversion method based on LGBM-MLP, and compares it with AMSR2 L2 sea surface temperature data to illustrate the improvement of the hybrid model's performance in regional nearshore sea surface temperature inversion. The specific steps are as follows: Step S1: Establish a dataset, which includes AMSR2 L1R observation data, ERA5 ocean-atmosphere related data, MGDSST daily sea surface temperature data, iQuam sea surface temperature data, and AMSR2 L2 sea surface temperature data.

[0025] Step S2: Preprocess the dataset. Match the obtained AMSR2 observation data, ERA5 ocean-atmosphere related data, MGDSST daily sea surface temperature data, and iQuam sea surface temperature data according to the matching principle to obtain a matched dataset, and then divide the dataset. Step S3: Construct a sea surface temperature inversion model based on LGBM-MLP, such as... Figure 1 As shown, LGBM-MLP combines the advantages of Multilayer Perceptron (MLP) and Light Gradient Boosting Machine (LGBM). Based on the Stacking model, it uses the prediction results of the base model as input to train the meta-model, thereby further improving the model's prediction accuracy. The base model is set as an LGBM model and an MLP model, and the meta-model is set as an MLP model. Step S4: Train and test the sea surface temperature inversion model based on LGBM-MLP, compare the model inversion results with L2 sea surface temperature data, and evaluate the accuracy of sea surface temperature inversion.

[0026] Step S1 is detailed as follows: Create a dataset covering the entire year of 2020. Download AMSR2 data from the Japan Aerospace Exploration Agency (JAXA) GCOM-W1 website. AMSR2 data includes Earth azimuth, Earth zenith angle, solar azimuth, solar zenith angle, time, latitude and longitude, AMSR2 6-36 GHz channel brightness temperature, AMSR2 L2 sea surface temperature data, land-sea identification, and mass identification. Download MGDSST sea surface temperature data, including sea surface temperature, from the Japan Meteorological Agency (JMA). Also download ERA5 reanalysis ocean-atmosphere related data from the European Centre for Medium-Range Weather Forecasts (ECMWF) website, including time, latitude and longitude, 10m U-wind, 10m V-wind, total column cloud liquid water, total column water vapor, and total average precipitation rate. Download iQuam sea surface temperature data from the In-Situ Sea Surface Temperature and Mass Monitoring System website.

[0027] Step S2 is detailed as follows: S21. First, the AMSR2 L1R observation data was matched with the coastline data, and sea surface temperature data within 200km of the nearshore area (100-130°E, 0-45°N) were filtered to obtain the AMSR2 nearshore observation data. The 10m U-wind and 10m V-wind data from ERA5 were converted to 10m wind speed and direction. Since the MGDSST data only contains sea surface temperature, the MGDSST dataset was constructed according to the data specifications, including longitude, latitude, time, and sea surface temperature. For dataset matching, the AMSR2 nearshore observation data was primarily used for matching with other datasets, including ERA5 data, MGDSST sea surface temperature data, iQuam sea surface temperature, and AMSR2 L2 sea surface temperature. Since there are certain spatiotemporal errors between different datasets, matching was performed with a 30-minute time window and a 25km spatial window to minimize these errors. The matched data underwent quality control. During the matching process, high-quality data was selected based on observation data quality indicators, iQuam quality indicators, and land-sea indicators. Quality control was also performed on AMSR2 L2 sea surface temperature outliers and MGDSST sea surface temperature-land-sea ice outliers. Data was retained according to the following rules: AMSR2 quality indicator of 0 indicates best quality; AMSR2 land-sea indicator of 0 indicates 0% land area within the scan footprint; MGDSST sea surface temperature not equal to 999 or 888 indicates it is neither land nor sea ice; iQuam quality indicator of 5 indicates best quality; and AMSR2 L2 sea surface temperature not equal to -32768 indicates valid data. The resulting matched dataset was then obtained.

[0028] S22. Divide the matched data into training, validation, and test sets. Divide the data from January to December 24, 2020, into a 75% and 25% set, respectively. 75% of the data will be used as the training dataset, and the remaining 25% will be used as the validation dataset to adjust hyperparameters and verify overfitting. Select the data from December 25 to 31, 2020, as a separate test set to test the model's ability to retrieve nearshore sea surface temperature in the region.

[0029] Step S3 is detailed as follows: S31. Determine the model input as latitude and longitude, time, 6-36 GHz horizontal and vertical polarization observed brightness temperature, sea surface 10m wind speed and direction, total column cloud liquid water, total column water vapor, total average precipitation rate, and MGDSST sea surface temperature. The model output is iQuam sea surface temperature data.

[0030] S32. Parameter settings for the basic model: LGBM parameters were selected from the following: maximum depth [6, 7, 8, 9], number of decision tree leaves [32, 64, 128, 256], learning rate [0.1, 0.01, 0.001], and feature ratio [0.7, 0.8, 0.9, 1]. After testing, the parameters were determined to be 8, 32, 0.1, and 1. Regarding the MLP hidden layers [16, 32, 64, 128], after testing, the parameters were determined to be 32 and 16. The hidden layers of the meta-model MLP were also determined to be 32 and 16 after testing.

[0031] The specific steps in step S4 are as follows: S41. To demonstrate the effectiveness of the LGBM-MLP hybrid model in retrieving regional nearshore sea surface temperature, the accuracy was verified by comparing it with the matched AMSR2 L2 sea surface temperature.

[0032] S42. Use regression indicators to evaluate the accuracy of sea surface temperature inversion results, including mean bias (MB), root mean square error (RMSE), and correlation coefficient (R).

[0033] (1) (2) (3) In the formula, The sea surface temperature of AMSR2 L2. To retrieve sea surface temperature, To retrieve the number of sea surface temperatures, This represents the average sea surface temperature of AMSR2 L2. This represents the average value of the retrieved sea surface temperature.

[0034] S43. Figure 2 The image shows a scatter plot comparing the sea surface temperature data obtained from the hybrid model inversion with the AMSR2 L2 and iQuam sea surface temperature data. The black line represents y=x, and the straight line fitted by the scatter plot is the red line. Compared to the AMSR2 L2 sea surface temperature data, the red and black lines of the hybrid model-inverted sea surface temperature data overlap more significantly, indicating that the hybrid model's sea surface temperature data is closer to the iQuam sea surface temperature. Furthermore, both the MB and RMSE are lower than those of the AMSR2 L2 sea surface temperature data, indicating that the hybrid model's sea surface temperature data has higher accuracy.

[0035] S44. After validating the effectiveness of the hybrid model, high-precision sea surface temperature data for Taiwan Province can be retrieved and analyzed. For example, further analysis of the retrieved sea surface temperature data can yield... Figure 3 The data shown is the monthly average sea surface temperature data for Taiwan Province.

[0036] This embodiment proposes a regional nearshore sea surface temperature retrieval method based on LGBM-MLP. The method achieves nearshore sea surface temperature retrieval through steps including data acquisition, preprocessing, model building, training, and evaluation. This technique incorporates LGBM and MLP models into the Stacking framework, primarily using AMSR2 L1R observation data. By adding physical parameters that affect sea surface temperature retrieval, the accuracy of nearshore sea surface temperature retrieval is improved.

[0037] Figure 2 This is a scatter plot showing the comparison between regional nearshore inverted sea surface temperature data and AMSR2 L2 sea surface temperature data from an embodiment of the present invention, and the iQuam sea surface temperature data. Figure 3 This is a line graph showing the monthly average sea surface temperature near the coast in an embodiment of the present invention.

[0038] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A regional nearshore sea surface temperature inversion method based on LGBM-MLP, characterized in that, include: Construct a dataset; wherein the dataset includes: microwave radiation data of the Earth's surface and atmosphere, sea surface temperature data, and reanalysis of ocean-atmosphere related data; The dataset is preprocessed; Based on the preprocessed dataset, the LGBM-MLP hybrid model is trained to obtain the sea surface temperature inversion model; Based on the sea surface temperature inversion model, the nearshore sea surface temperature of the region is inverted.

2. The regional nearshore sea surface temperature inversion method based on LGBM-MLP according to claim 1, characterized in that, The microwave radiation data of the Earth's surface and atmosphere includes: Earth's azimuth, Earth's zenith angle, solar azimuth, solar zenith angle, time, latitude and longitude, channel brightness temperature, land and sea markings, and mass markings. The reanalysis of ocean-atmosphere related data includes: time, latitude and longitude, U wind at 10m sea surface, V wind at 10m sea surface, total cloud liquid water, total cloud water vapor, and total average precipitation rate. The sea surface temperature data includes: global satellite in-situ fusion of daily sea surface temperature data and in-situ sea surface temperature quality monitoring data.

3. The regional nearshore sea surface temperature inversion method based on LGBM-MLP according to claim 2, characterized in that, Preprocessing the dataset includes: The microwave radiation data of the Earth's surface and atmosphere, sea surface temperature data, and related ocean atmospheric data are matched to obtain a matching dataset.

4. The regional nearshore sea surface temperature inversion method based on LGBM-MLP according to claim 3, characterized in that, Matching the microwave radiation data of the Earth's surface and atmosphere with sea surface temperature data and reanalysis of ocean-atmosphere related data includes: The microwave radiation data of the Earth's surface and atmosphere are matched with the coastline data, and the sea surface temperature data of the nearshore preset area are filtered to obtain nearshore observation data. Based on preset time and space windows, nearshore observation data are matched with reanalysis ocean-atmosphere related data and in-situ sea surface temperature and quality monitoring data; During the matching process, high-quality data were selected based on the quality labels, iQuam quality labels, and land-sea labels in the nearshore observation data. Quality control was also performed based on the AMSR2 L2 sea surface temperature anomalies and the MGDSST sea surface temperature-land sea ice anomalies.

5. The regional nearshore sea surface temperature inversion method based on LGBM-MLP according to claim 1, characterized in that, The LGBM-MLP hybrid model adopts a stacking framework structure and includes a base model and a meta-model. The basic model includes an LGBM model and an MLP model, and the meta-model includes an MLP model.

6. The regional nearshore sea surface temperature inversion method based on LGBM-MLP according to claim 1, characterized in that, When training the LGBM-MLP hybrid model, the inputs to the model are channel brightness temperature, sea surface wind speed and direction, total average precipitation rate, total column water vapor, total column cloud liquid water, and daily sea surface temperature data, and the output of the model is sea surface temperature.

Images