CN115758865B - Underwater three-dimensional warm salt reconstruction method and system based on deep learning - Google Patents
Underwater three-dimensional warm salt reconstruction method and system based on deep learning Download PDFInfo
- Publication number
- CN115758865B CN115758865B CN202211354385.2A CN202211354385A CN115758865B CN 115758865 B CN115758865 B CN 115758865B CN 202211354385 A CN202211354385 A CN 202211354385A CN 115758865 B CN115758865 B CN 115758865B
- Authority
- CN
- China
- Prior art keywords
- data
- salt
- temperature
- remote sensing
- underwater
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 150000003839 salts Chemical class 0.000 title claims abstract description 183
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000013136 deep learning model Methods 0.000 claims abstract description 33
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 101000794020 Homo sapiens Bromodomain-containing protein 8 Proteins 0.000 claims description 13
- 101001006782 Homo sapiens Kinesin-associated protein 3 Proteins 0.000 claims description 13
- 101000615355 Homo sapiens Small acidic protein Proteins 0.000 claims description 13
- WGKGADVPRVLHHZ-ZHRMCQFGSA-N N-[(1R,2R,3S)-2-hydroxy-3-phenoxazin-10-ylcyclohexyl]-4-(trifluoromethoxy)benzenesulfonamide Chemical compound O[C@H]1[C@@H](CCC[C@@H]1N1C2=CC=CC=C2OC2=C1C=CC=C2)NS(=O)(=O)C1=CC=C(OC(F)(F)F)C=C1 WGKGADVPRVLHHZ-ZHRMCQFGSA-N 0.000 claims description 13
- 102100021255 Small acidic protein Human genes 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 238000003908 quality control method Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 6
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 description 26
- 239000011159 matrix material Substances 0.000 description 7
- 239000013535 sea water Substances 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 239000012267 brine Substances 0.000 description 2
- HPALAKNZSZLMCH-UHFFFAOYSA-M sodium;chloride;hydrate Chemical compound O.[Na+].[Cl-] HPALAKNZSZLMCH-UHFFFAOYSA-M 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000001311 chemical methods and process Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000000593 degrading effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005370 electroosmosis Methods 0.000 description 1
- 239000013388 element-organic framework Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 230000008014 freezing Effects 0.000 description 1
- 238000007710 freezing Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Landscapes
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention relates to an underwater three-dimensional warm salt reconstruction method and system based on deep learning, and belongs to the field of ocean remote sensing. The method comprises the following steps: acquiring remote sensing temperature data and remote sensing salinity data; acquiring underwater real-time salt temperature profile data and preprocessing the underwater real-time salt temperature profile data to generate the underwater salt temperature profile data; a data matching method is adopted to construct a corresponding relation between remote sensing sea surface temperature salt data and underwater temperature salt profile data, and a matching data set is established; establishing a multi-layer perceptron model and training by adopting a matching data set; inputting remote sensing sea surface temperature salt data of the accurate point to be predicted into a trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point. The method is based on actual data of satellite remote sensing observation, so that the inversion result is more accurate; meanwhile, the matching data set trained by the model is based on scattered points of accurate buoy points, so that the deep learning model obtained by training can invert scattered point temperature salt profiles of any accurate points on a grid.
Description
Technical Field
The invention relates to the technical field of ocean remote sensing, in particular to an underwater three-dimensional warm salt reconstruction method and system based on deep learning.
Background
The temperature and the salinity are two basic characteristics of seawater, and influence the physical properties of the seawater such as freezing point, density, thermal expansion coefficient, compressibility, conductivity, sound velocity and the like. Meanwhile, chemical processes, ecological processes, biological geological processes, sea-air interaction processes and the like in the sea are also influenced by the temperature and the salinity of the sea water. Seawater temperature and salinity are the most basic marine environmental parameters, and are important physical quantities for describing the properties of seawater. Through the temperature and the salinity of the sea water, other sea environment parameters such as the density of the sea water, the sound velocity and the like can be calculated, and basic data are provided for researching physical sea phenomena such as jump layers, internal waves, vortex and the like.
The conventional method for inverting an underwater three-dimensional structure by using sea surface information mainly comprises the following steps: (1) The dynamic method, such as Hurlburt et al (1986), constructs a numerical ocean model, and can calculate underwater information based on satellite altitude counting, so as to reverse the deep pressure field; (2) An empirical estimation method for the abnormal temperature and volume of the section of the whole water body is provided by an empirical formula method of sound velocity, such as Meinen et al (2000), in combination with historical hydrologic data and sound velocity observation results; (3) Regression statistical methods, such as Carnes et al (1990), use surface altitude and temperature to generate a temperature synthesis profile of the ocean based on regression relations between empirical quadrature function amplitudes (EOFs) between vertical structures of sea surface dynamic altitude and temperature.
Although a great deal of research and great progress are made in the aspect of deep learning inversion of underwater three-dimensional warm salt structures at home and abroad, some problems exist: on one hand, most of data used for learning by the deep learning model are data of a re-analysis product, and the spatial resolution is low, so that the inversion result is inaccurate; on the other hand, the input and output of the constructed model are regular grids, and the model cannot be directly applied to specific accurate point positions, so that the actual requirements cannot be met.
Disclosure of Invention
In order to solve or at least alleviate the problems, the invention provides an underwater three-dimensional warm salt reconstruction method and system based on deep learning, which can invert a scattered point warm salt profile of any accurate point not on a grid and improve the accuracy of an inversion result.
In order to achieve the above object, the present invention provides the following solutions:
an underwater three-dimensional warm salt reconstruction method based on deep learning comprises the following steps:
acquiring remote sensing temperature data and remote sensing salinity data to form remote sensing sea surface temperature salt data;
acquiring underwater real-time salt temperature profile data and preprocessing the underwater real-time salt temperature profile data to generate the underwater salt temperature profile data;
a data matching method is adopted to construct a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data, so that a matching data set is established;
establishing a multi-layer perceptron model, and training by adopting the matched data set to obtain a trained deep learning model;
inputting remote sensing sea surface temperature salt data of an accurate point to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point.
Optionally, the acquiring the remote sensing temperature data and the remote sensing salinity data and forming the remote sensing sea surface temperature salt data includes:
acquiring FY3C VIRR sea surface temperature products as remote sensing temperature data and storing the remote sensing temperature data in an HDF format;
acquiring 8-day sliding average sea surface salinity products of SMAP satellites as remote sensing salinity data and storing the remote sensing salinity data in nc format;
and forming remote sensing sea surface temperature salt data by the remote sensing temperature data and the remote sensing salinity data corresponding to the longitude and latitude.
Optionally, the acquiring and preprocessing the underwater real-time brine temperature profile data to generate the underwater brine temperature profile data specifically includes:
and acquiring underwater real-time salt temperature profile data, performing quality control pretreatment, and performing linear interpolation on the pretreated profile data with 5dbar as vertical resolution in a pressure range of 0-2000dbar to generate the underwater salt temperature profile data.
Optionally, the data matching method is adopted to construct a corresponding relationship between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data, so as to establish a matching data set, which specifically includes:
using a nearest-neighbor matching method to associate remote sensing temperature data closest to the position of the Argo practical observation point at the same time as a remote sensing observation temperature value of the point, associating remote sensing salinity data closest to the position of the Argo practical observation point at the same date as a remote sensing observation salinity value of the point, associating the underwater thermal salt profile data to the position of the Argo practical observation point according to longitude and latitude data of the underwater thermal salt profile data, and constructing a corresponding relation between the remote sensing sea surface thermal salt data and the underwater thermal salt profile data;
and using remote sensing sea surface salt temperature data and underwater salt temperature profile data which are related to the same Argo actual observation point position and corresponding longitude and latitude data as the same line of data of the matching data set, and establishing the matching data set.
Optionally, the building a multi-layer perceptron model, and training by using the matching data set to obtain a trained deep learning model specifically includes:
establishing a multi-layer perceptron model with 5 hidden layers; the node numbers of the 5 hidden layers are 64,128,256,64,32 respectively;
and training the multi-layer perceptron model by adopting the matching data set, wherein an activation function in a training process is set as ReLU, a model optimizer is set as Adam, a learning rate is set as 0.001, and a trained deep learning model is obtained after training is completed.
An underwater three-dimensional warm salt reconstruction system based on deep learning, comprising:
the remote sensing sea surface temperature salt data acquisition module is used for acquiring remote sensing temperature data and remote sensing salinity data and forming remote sensing sea surface temperature salt data;
the underwater temperature and salt profile data acquisition module is used for acquiring underwater real temperature and salt profile data and preprocessing the underwater real temperature and salt profile data to generate the underwater temperature and salt profile data;
the data matching module is used for constructing a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data by adopting a data matching method so as to establish a matching data set;
the model building and training module is used for building a multi-layer perceptron model, and training by adopting the matching data set to obtain a trained deep learning model;
the underwater three-dimensional temperature salt reconstruction module is used for inputting remote sensing sea surface temperature salt data of an accurate point position to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point position.
Optionally, the remote sensing sea surface temperature salt data acquisition module specifically includes:
the remote sensing temperature data acquisition unit is used for acquiring FY3C VIRR sea surface temperature products as remote sensing temperature data and storing the remote sensing temperature data in an HDF format;
the remote sensing salinity data acquisition unit is used for acquiring 8-day moving average sea surface salinity products of the SMAP satellite as remote sensing salinity data and storing the remote sensing salinity data in an nc format;
the remote sensing sea surface temperature salt data construction unit is used for forming remote sensing sea surface temperature salt data together with the remote sensing salinity data corresponding to the longitude and latitude.
Optionally, the underwater warm salt profile data acquisition module specifically includes:
the underwater thermal salt profile data processing unit is used for acquiring underwater real-time thermal salt profile data and performing quality control pretreatment, and linearly interpolating the pretreated profile data with 5dbar as vertical resolution within the pressure range of 0-2000dbar to generate the underwater thermal salt profile data.
Optionally, the data matching module specifically includes:
the data matching unit is used for associating the remote sensing temperature data closest to the position of the practical observation point of the Argo at the same time with a nearest matching method to be a remote sensing observation temperature value of the point, associating the remote sensing salinity data closest to the position of the practical observation point of the Argo at the same date with the remote sensing salinity value of the point, associating the underwater thermal salt profile data with the position of the practical observation point of the Argo according to the longitude and latitude data of the underwater thermal salt profile data, and constructing a corresponding relation between the remote sensing sea surface thermal salt data and the underwater thermal salt profile data;
the matching data set establishing unit is used for establishing the matching data set by taking remote sensing sea surface temperature salt data and underwater temperature salt profile data which are related to the same Argo actual observation point position and corresponding longitude and latitude data together as the same line of data of the matching data set.
Optionally, the model building and training module specifically includes:
the model building unit is used for building a multi-layer perceptron model with 5 hidden layers; the node numbers of the 5 hidden layers are 64,128,256,64,32 respectively;
the model training unit is used for training the multi-layer perceptron model by adopting the matching data set, the activation function in the training process is set to be ReLU, the model optimizer is set to be Adam, the learning rate is set to be 0.001, and a trained deep learning model is obtained after training is completed.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides an underwater three-dimensional warm salt reconstruction method and system based on deep learning, wherein the method comprises the following steps: acquiring remote sensing temperature data and remote sensing salinity data to form remote sensing sea surface temperature salt data; acquiring underwater real-time salt temperature profile data and preprocessing the underwater real-time salt temperature profile data to generate the underwater salt temperature profile data; a data matching method is adopted to construct a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data, so that a matching data set is established; establishing a multi-layer perceptron model, and training by adopting the matched data set to obtain a trained deep learning model; inputting remote sensing sea surface temperature salt data of an accurate point to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point. The data used in the method is actual data based on satellite remote sensing observation, and compared with the data of a re-analysis product, the data is closer to an actual value, so that the inversion result is more accurate; meanwhile, the training data set is based on scattered points of accurate buoy points, so that the deep learning model obtained through training can invert scattered point temperature salt profiles of any accurate points on a grid, can meet actual needs and has wide application prospects.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of an underwater three-dimensional warm salt reconstruction method based on deep learning according to an embodiment of the invention;
FIG. 2 is a schematic diagram of an underwater three-dimensional warm salt reconstruction method based on deep learning according to an embodiment of the invention;
FIG. 3 is a schematic diagram of remote sensing temperature data according to an embodiment of the present invention;
fig. 4 is a schematic diagram of remote sensing salinity data according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide an underwater three-dimensional warm salt reconstruction method based on deep learning, which can invert a scattered point warm salt profile of any accurate point not on a grid and improve the accuracy of an inversion result.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
FIG. 1 is a flow chart of an underwater three-dimensional warm salt reconstruction method based on deep learning according to an embodiment of the invention; fig. 2 is a schematic diagram of an underwater three-dimensional warm salt reconstruction method based on deep learning according to an embodiment of the invention. Referring to fig. 2, in order to realize direct inversion of an underwater three-dimensional temperature and salinity structure based on remote sensing temperature and salinity in the global ocean range, the invention constructs a matching data set between global scale remote sensing temperature salt and an underwater temperature salt profile, builds a multi-layer perceptron model for training, learns to obtain the corresponding relation between sea surface temperature salt and the underwater three-dimensional temperature salt, and realizes prediction of corresponding underwater three-dimensional structure information by inputting remote sensing temperature and salinity of a certain accurate point. Referring to fig. 1, the invention relates to an underwater three-dimensional warm salt reconstruction method based on deep learning, which specifically comprises the following steps:
step 1: and acquiring remote sensing temperature data and remote sensing salinity data to form remote sensing sea surface temperature salt data.
According to the method, FY3CVIRR (visible light infrared scanning radiometer) sea surface temperature products are selected as remote sensing temperature data, SMAP satellite 8-day moving average sea surface salinity products are selected as remote sensing salinity data, and the remote sensing temperature data and the remote sensing salinity data corresponding to longitude and latitude jointly form remote sensing sea surface temperature salinity data.
The FY3CVIRR sea surface temperature product is clear sky sea surface temperature information obtained by inversion of a VIRR splitting window channel or a splitting window and a medium wave infrared three channel, and is divided into a sea surface temperature product (daytime) and a sea surface temperature product (night) according to the observation period. The SMAP satellite 8-day sliding average seasurface salinity product is sea surface salinity information obtained by averaging observations of near-polar orbit SMAP satellites covering the global sea for 3 days with a period of 8 days.
FY3C sea surface temperature data is stored in an HDF format, a data set is a sea surface temperature matrix, the matrix size is 7200 x 3600, and a visual image of the data set is shown in FIG. 3. The SMAP satellite 8-day moving average sea surface salinity data is stored in nc format, the data set used is sss_smap matrix with matrix size 1440 x 720, and the data set visualization is shown in fig. 4.
Step 2: and acquiring underwater real-time salt temperature profile data and preprocessing the data to generate the underwater salt temperature profile data.
The method selects global ocean Argo scattered point data set controlled by the quality of the China Argo real-time data center as real temperature and salt profile data, and carries out further quality control pretreatment on the downloaded data based on the following criteria:
(1) The quality sign is 1;
(2) The shallowest data lie between the sea surface and 10 meters deep, the deepest data are below 1000 meters;
(3) At least 30 effective values are arranged at the upper 1000 meters of each section;
(4) The characteristic value should be within the effective range (temperature 35 to-2 ℃, salinity 38 to 30psu (Practical salinity unit));
(5) The pressure profile should increase monotonically.
All profiles pretreated by the above criteria were linearly interpolated (constant extrapolation to the surface) at 5dbar as vertical resolution over the pressure range of 0-2000dbar, yielding underwater warm salt profile data.
In one embodiment, 185800 underwater warm salt profile data of the Argo scatter points meeting the requirements are finally obtained and stored in csv format. Each csv file contains longitude, latitude, observation time for that point, temperature and salinity observations for 45 standard layers at depths of 0-1000 meters.
Step 3: and constructing a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data by adopting a data matching method, thereby establishing a matching data set.
And (3) synthesizing the data obtained in the step (1) and the step (2), and constructing the corresponding relation between the remote sensing sea surface temperature salt and the underwater three-dimensional temperature salt.
The matching of the remote sensing temperature data needs to distinguish between daytime and nighttime in time, the daytime is defined as 6:00 to 18:00, the nighttime is defined as 18:00 to 6:00 of the next day, and the matching with the sea surface temperature product (daytime) or sea surface temperature product (night) of the same day is determined according to the specific observation time of the Argo buoy. And spatially using a nearest-neighbor matching method to correlate the remote sensing temperature data closest to the position of the Argo practical observation point into a remote sensing observation temperature value of the point.
The matching of the remote sensing salinity data only needs to be corresponding to the date in time, and the closest matching method is also used in space, so that the remote sensing salinity data closest to the position of the practical observation point of Argo on the same date is correlated to the remote sensing observation salinity value of the point.
And using remote sensing sea surface salt temperature data and underwater salt temperature profile data which are related to the same Argo actual observation point position and corresponding longitude and latitude data as the same line of data of the matching data set, and establishing the matching data set. Only profiles meeting both the sensed temperature and the sensed salinity are recorded. In a specific embodiment, 66193 effective value pairs are obtained through the step 3 total matching and stored in a csv format, each row of matching data in the matching data set is in the form of longitude, latitude, remote sensing temperature data, remote sensing salinity data, 45 standard layers of temperatures and 45 standard layers of salinity, and the whole matching data set is in the form of a 66193 x 94 matrix.
Step 4: and establishing a multi-layer perceptron model, and training by adopting the matched data set to obtain a trained deep learning model.
And (3) establishing a multi-layer perceptron model with 5 hidden layers, training the matched data set in the step (3) to obtain a corresponding deep learning model, and predicting corresponding underwater three-dimensional structure information by inputting remote sensing temperature and salinity of a certain accurate point location through the deep learning model.
The realization principle of the multi-layer perceptron is as follows:
input matrix X ε R n×d (matching number in the present invention)A 661193 x 94 matrix of data sets) represents n samples, where each sample has d input features. For a single hidden layer multi-layer perceptron with H hidden units, H E R is used n×h Representing the output of the hidden layer. Since the hidden layer and the output layer are all fully connected, the hidden layer weight W (1) ∈R d×h Hidden layer bias b (1) ∈R 1×h Output layer weight W (2) ∈R h×q Output layer bias b # 2 )∈R 1×q . Formally, the output O E R of the single hidden layer multi-layer perceptron is calculated as follows n×q ：
H＝XW (1) +b (1) (1)
O＝HW (2) +b (2) (2)
Wherein, q representing the number of output units.
To exploit the potential of multi-layer architecture, a nonlinear activation function σ is applied to each hidden unit after affine transformation to prevent the multi-layer perceptron from degrading to a linear model:
H＝σ(XW (1) +b (1) ) (3)
O＝HW (2) +b (2) (4)
to make the multi-layer perceptron model more generalized, such hidden layers, e.g., H, may be further stacked (1) ＝σ 1 (XW (1) +b (1) )，H (2) ＝σ 2 (XW (2) +b (2) ) ... Until the model meets the requirements.
In the invention, the multi-layer perceptron model structure is set to be 5 hidden layers, and the node number is 64,128,256,64,32. The activation function of the model training process was set to ReLU, the model optimizer was set to Adam, and the learning rate was set to 0.001. And obtaining a trained deep learning model after training is completed.
When the trained deep learning model is called, the input data format is as follows: longitude, latitude, remote sensing temperature data, remote sensing salinity data; the output data format is: longitude, latitude, predicted underwater three-dimensional temperature salt profile data (i.e., underwater three-dimensional temperature salt structure information).
Step 5: inputting remote sensing sea surface temperature salt data of an accurate point to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point.
According to the underwater three-dimensional warm salt reconstruction method based on deep learning, a matching data set between global scale remote temperature sensing salt and an underwater warm salt profile is constructed, a multi-layer perceptron model is built for training, the corresponding relation between sea surface warm salt and the underwater three-dimensional warm salt is obtained through learning, and the effective method for predicting the corresponding underwater three-dimensional warm salt structure by inputting remote sensing temperature and salinity of a certain accurate point position is realized. According to the method, the remote sensing temperature and salt product data with higher precision are used as the original data, and the multi-layer perceptron model is trained based on the original data, so that the inversion result obtained according to the trained deep learning model can be more accurate.
In addition, by calling a deep learning model of the temperature or the salinity of the corresponding layer, the temperature and the salinity information of each layer of 45 standard layers on the 0-1000 meter layer surface under water can be predicted by remote sensing of the temperature and the salinity, so that a three-dimensional temperature salt structure field under water is constructed.
Most of the data used for learning by the existing deep learning model are data of a re-analysis product, the spatial resolution is low, and the data used in the method are actual data based on satellite remote sensing observation, compared with the data of the re-analysis product, the data are closer to an actual value, so that the inversion result is more accurate. The input and output of the model constructed by the traditional method are regular grids, and the model cannot be directly applied to specific accurate point positions, so that the actual requirements cannot be met; the training data set (namely the matching data set) is the Argo scattered points based on the accurate buoy point positions, so that the deep learning model obtained by training can invert the scattered point temperature salt profile of any accurate point position on the grid, can meet the actual needs, and has wide application prospect.
Based on the method provided by the invention, the invention also provides an underwater three-dimensional warm salt reconstruction system based on deep learning, which comprises the following steps:
the remote sensing sea surface temperature salt data acquisition module is used for acquiring remote sensing temperature data and remote sensing salinity data and forming remote sensing sea surface temperature salt data;
the underwater temperature and salt profile data acquisition module is used for acquiring underwater real temperature and salt profile data and preprocessing the underwater real temperature and salt profile data to generate the underwater temperature and salt profile data;
the data matching module is used for constructing a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data by adopting a data matching method so as to establish a matching data set;
the model building and training module is used for building a multi-layer perceptron model, and training by adopting the matching data set to obtain a trained deep learning model;
the underwater three-dimensional temperature salt reconstruction module is used for inputting remote sensing sea surface temperature salt data of an accurate point position to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point position.
The remote sensing sea surface temperature salt data acquisition module specifically comprises:
the remote sensing temperature data acquisition unit is used for acquiring FY3CVIRR sea surface temperature products as remote sensing temperature data and storing the remote sensing temperature data in an HDF format;
the remote sensing salinity data acquisition unit is used for acquiring 8-day moving average sea surface salinity products of the SMAP satellite as remote sensing salinity data and storing the remote sensing salinity data in an nc format;
the remote sensing sea surface temperature salt data construction unit is used for forming remote sensing sea surface temperature salt data together with the remote sensing salinity data corresponding to the longitude and latitude.
The underwater warm salt profile data acquisition module specifically comprises:
the underwater thermal salt profile data processing unit is used for acquiring underwater real-time thermal salt profile data and performing quality control pretreatment, and linearly interpolating the pretreated profile data with 5dbar as vertical resolution within the pressure range of 0-2000dbar to generate the underwater thermal salt profile data.
The data matching module specifically comprises:
the data matching unit is used for associating the remote sensing temperature data closest to the position of the practical observation point of the Argo at the same time with a nearest matching method to be a remote sensing observation temperature value of the point, associating the remote sensing salinity data closest to the position of the practical observation point of the Argo at the same date with the remote sensing salinity value of the point, associating the underwater thermal salt profile data with the position of the practical observation point of the Argo according to the longitude and latitude data of the underwater thermal salt profile data, and constructing a corresponding relation between the remote sensing sea surface thermal salt data and the underwater thermal salt profile data;
the matching data set establishing unit is used for establishing the matching data set by taking remote sensing sea surface temperature salt data and underwater temperature salt profile data which are related to the same Argo actual observation point position and corresponding longitude and latitude data together as the same line of data of the matching data set.
The model building and training module specifically comprises:
the model building unit is used for building a multi-layer perceptron model with 5 hidden layers; the node numbers of the 5 hidden layers are 64,128,256,64,32 respectively;
the model training unit is used for training the multi-layer perceptron model by adopting the matching data set, the activation function in the training process is set to be ReLU, the model optimizer is set to be Adam, the learning rate is set to be 0.001, and a trained deep learning model is obtained after training is completed.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (4)
1. An underwater three-dimensional warm salt reconstruction method based on deep learning is characterized by comprising the following steps:
acquiring remote sensing temperature data and remote sensing salinity data to form remote sensing sea surface temperature salt data;
the obtaining of the remote sensing temperature data and the remote sensing salinity data to form remote sensing sea surface temperature salt data comprises the following steps:
acquiring FY3C VIRR sea surface temperature products as remote sensing temperature data and storing the remote sensing temperature data in an HDF format; FY3C VIRR sea surface temperature products are clear sky sea surface temperature information obtained by inversion of a VIRR splitting window channel or a splitting window and a medium wave infrared three channel, and are divided into daytime sea surface temperature products and nighttime sea surface temperature products according to observation time periods;
acquiring 8-day sliding average sea surface salinity products of SMAP satellites as remote sensing salinity data and storing the remote sensing salinity data in nc format; the 8-day sliding average sea surface salinity product of the SMAP satellite is sea surface salinity information obtained by averaging the observation result of the near-polar orbit SMAP satellite covering the global sea for 3 days with the period of 8 days;
the remote sensing temperature data and the remote sensing salinity data corresponding to longitude and latitude are formed into remote sensing sea surface temperature salt data together;
acquiring underwater real-time salt temperature profile data and preprocessing the underwater real-time salt temperature profile data to generate the underwater salt temperature profile data;
the method for acquiring the underwater real-time salt temperature profile data and preprocessing the underwater real-time salt temperature profile data to generate the underwater real-time salt temperature profile data specifically comprises the following steps:
acquiring underwater real-time salt temperature profile data, performing quality control pretreatment, and performing linear interpolation by taking 5dbar as vertical resolution in the pressure range of 0-2000dbar on the pretreated profile data to generate the underwater salt temperature profile data;
the downloaded real temperature salt profile data is subjected to quality control pretreatment based on the following criteria:
(1) The quality sign is 1;
(2) The shallowest data lie between the sea surface and 10 meters deep, the deepest data are below 1000 meters;
(3) At least 30 effective values are arranged at the upper 1000 meters of each section;
(4) The characteristic value is in the effective range, the temperature is 35 to minus 2 ℃, and the salinity is 38 to 30psu;
(5) The pressure profile should increase monotonically;
a data matching method is adopted to construct a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data, so that a matching data set is established;
the method for constructing the corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data by adopting a data matching method, thereby establishing a matching data set, specifically comprising the following steps:
using a nearest-neighbor matching method to associate remote sensing temperature data closest to the position of the Argo practical observation point at the same time as a remote sensing observation temperature value of the point, associating remote sensing salinity data closest to the position of the Argo practical observation point at the same date as a remote sensing observation salinity value of the point, associating the underwater thermal salt profile data to the position of the Argo practical observation point according to longitude and latitude data of the underwater thermal salt profile data, and constructing a corresponding relation between the remote sensing sea surface thermal salt data and the underwater thermal salt profile data;
remote sensing sea surface salt temperature data and underwater salt temperature profile data which are related to the same Argo actual observation point position are taken as the same line of data of the matching data set together with corresponding longitude and latitude data, and the matching data set is established;
establishing a multi-layer perceptron model, and training by adopting the matched data set to obtain a trained deep learning model;
inputting remote sensing sea surface temperature salt data of an accurate point to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point.
2. The method of claim 1, wherein the establishing a multi-layer perceptron model and training using the matching dataset to obtain a trained deep learning model comprises:
establishing a multi-layer perceptron model with 5 hidden layers; the node numbers of the 5 hidden layers are 64,128,256,64,32 respectively;
and training the multi-layer perceptron model by adopting the matching data set, wherein an activation function in a training process is set as ReLU, a model optimizer is set as Adam, a learning rate is set as 0.001, and a trained deep learning model is obtained after training is completed.
3. An underwater three-dimensional warm salt reconstruction system based on deep learning, which is characterized by comprising:
the remote sensing sea surface temperature salt data acquisition module is used for acquiring remote sensing temperature data and remote sensing salinity data and forming remote sensing sea surface temperature salt data;
the remote sensing sea surface temperature salt data acquisition module specifically comprises:
the remote sensing temperature data acquisition unit is used for acquiring FY3C VIRR sea surface temperature products as remote sensing temperature data and storing the remote sensing temperature data in an HDF format; FY3C VIRR sea surface temperature products are clear sky sea surface temperature information obtained by inversion of a VIRR splitting window channel or a splitting window and a medium wave infrared three channel, and are divided into daytime sea surface temperature products and nighttime sea surface temperature products according to observation time periods;
the remote sensing salinity data acquisition unit is used for acquiring 8-day moving average sea surface salinity products of the SMAP satellite as remote sensing salinity data and storing the remote sensing salinity data in an nc format; the 8-day sliding average sea surface salinity product of the SMAP satellite is sea surface salinity information obtained by averaging the observation result of the near-polar orbit SMAP satellite covering the global sea for 3 days with the period of 8 days;
the remote sensing sea surface temperature salt data construction unit is used for jointly forming remote sensing sea surface temperature salt data by the remote sensing temperature data and the remote sensing salinity data corresponding to the longitude and latitude;
the underwater temperature and salt profile data acquisition module is used for acquiring underwater real temperature and salt profile data and preprocessing the underwater real temperature and salt profile data to generate the underwater temperature and salt profile data;
the underwater warm salt profile data acquisition module specifically comprises:
the underwater temperature salt profile data processing unit is used for acquiring underwater real-time temperature salt profile data and performing quality control pretreatment, and linearly interpolating the pretreated profile data within the pressure range of 0-2000dbar by taking 5dbar as the vertical resolution to generate the underwater temperature salt profile data;
the downloaded real temperature salt profile data is subjected to quality control pretreatment based on the following criteria:
(1) The quality sign is 1;
(2) The shallowest data lie between the sea surface and 10 meters deep, the deepest data are below 1000 meters;
(3) At least 30 effective values are arranged at the upper 1000 meters of each section;
(4) The characteristic value is in the effective range, the temperature is 35 to minus 2 ℃, and the salinity is 38 to 30psu;
(5) The pressure profile should increase monotonically;
the data matching module is used for constructing a corresponding relation between the remote sensing sea surface temperature salt data and the underwater temperature salt profile data by adopting a data matching method so as to establish a matching data set;
the data matching module specifically comprises:
the data matching unit is used for associating the remote sensing temperature data closest to the position of the practical observation point of the Argo at the same time with a nearest matching method to be a remote sensing observation temperature value of the point, associating the remote sensing salinity data closest to the position of the practical observation point of the Argo at the same date with the remote sensing salinity value of the point, associating the underwater thermal salt profile data with the position of the practical observation point of the Argo according to the longitude and latitude data of the underwater thermal salt profile data, and constructing a corresponding relation between the remote sensing sea surface thermal salt data and the underwater thermal salt profile data;
the matching data set establishing unit is used for establishing the matching data set by taking remote sensing sea surface salt temperature data and underwater salt temperature profile data which are related to the same Argo actual observation point position and corresponding longitude and latitude data together as the same line of data of the matching data set;
the model building and training module is used for building a multi-layer perceptron model, and training by adopting the matching data set to obtain a trained deep learning model;
the underwater three-dimensional temperature salt reconstruction module is used for inputting remote sensing sea surface temperature salt data of an accurate point position to be predicted into the trained deep learning model, and predicting the underwater three-dimensional temperature salt structure information of the accurate point position.
4. The system of claim 3, wherein the model building and training module specifically comprises:
the model building unit is used for building a multi-layer perceptron model with 5 hidden layers; the node numbers of the 5 hidden layers are 64,128,256,64,32 respectively;
the model training unit is used for training the multi-layer perceptron model by adopting the matching data set, the activation function in the training process is set to be ReLU, the model optimizer is set to be Adam, the learning rate is set to be 0.001, and a trained deep learning model is obtained after training is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211354385.2A CN115758865B (en) | 2022-11-01 | 2022-11-01 | Underwater three-dimensional warm salt reconstruction method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211354385.2A CN115758865B (en) | 2022-11-01 | 2022-11-01 | Underwater three-dimensional warm salt reconstruction method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115758865A CN115758865A (en) | 2023-03-07 |
| CN115758865B true CN115758865B (en) | 2023-11-14 |
Family
ID=85355998
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211354385.2A Active CN115758865B (en) | 2022-11-01 | 2022-11-01 | Underwater three-dimensional warm salt reconstruction method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115758865B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116822381B (en) * | 2023-08-30 | 2023-11-21 | 中国海洋大学 | Marine temperature salt structure inversion method based on artificial intelligence |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109145494A (en) * | 2018-09-11 | 2019-01-04 | 北京师范大学 | A kind of Sea surface temperature method and system |
| CN109543768A (en) * | 2018-11-30 | 2019-03-29 | 福州大学 | Ocean interior thermohaline information intelligent extracting method based on multi-source satellite remote sensing |
| CN113984178A (en) * | 2021-09-18 | 2022-01-28 | 中国人民解放军国防科技大学 | Method for constructing mesoscale vortex underwater sound velocity disturbance general structure |
| CN114238847A (en) * | 2021-10-29 | 2022-03-25 | 中国人民解放军61540部队 | Surface layer accurate-rotation reconstruction method and system based on ocean measured data |
| CN114758080A (en) * | 2022-06-14 | 2022-07-15 | 中国科学院空天信息创新研究院 | Sea surface salinity gridding inversion method and device |
| CN115062527A (en) * | 2022-03-14 | 2022-09-16 | 北京华云星地通科技有限公司 | Geostationary satellite sea temperature inversion method and system based on deep learning |
-
2022
- 2022-11-01 CN CN202211354385.2A patent/CN115758865B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109145494A (en) * | 2018-09-11 | 2019-01-04 | 北京师范大学 | A kind of Sea surface temperature method and system |
| CN109543768A (en) * | 2018-11-30 | 2019-03-29 | 福州大学 | Ocean interior thermohaline information intelligent extracting method based on multi-source satellite remote sensing |
| CN113984178A (en) * | 2021-09-18 | 2022-01-28 | 中国人民解放军国防科技大学 | Method for constructing mesoscale vortex underwater sound velocity disturbance general structure |
| CN114238847A (en) * | 2021-10-29 | 2022-03-25 | 中国人民解放军61540部队 | Surface layer accurate-rotation reconstruction method and system based on ocean measured data |
| CN115062527A (en) * | 2022-03-14 | 2022-09-16 | 北京华云星地通科技有限公司 | Geostationary satellite sea temperature inversion method and system based on deep learning |
| CN114758080A (en) * | 2022-06-14 | 2022-07-15 | 中国科学院空天信息创新研究院 | Sea surface salinity gridding inversion method and device |
Non-Patent Citations (2)
| Title |
|---|
| A Deep Learning Network to Retrieve Ocean Hydrographic Profiles from Combined Satellite and In Situ Measurements;Bruno Buongiorno Nardelli;remote sensing;第12卷;第1-14页 * |
| Mesoscale Eddies in the Northwestern Pacific Ocean: Three-Dimensional Eddy Structures and Heat/Salt Transports;Di Dong et al.;Journal of Geophysical Research: Oceans;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115758865A (en) | 2023-03-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cummings et al. | Ocean data assimilation systems for GODAE | |
| Stewart et al. | JRA55-do-based repeat year forcing datasets for driving ocean–sea-ice models | |
| Chapman et al. | Arctic sea ice variability: Model sensitivities and a multidecadal simulation | |
| CN115758865B (en) | Underwater three-dimensional warm salt reconstruction method and system based on deep learning | |
| CN112598113A (en) | Ocean sound velocity profile acquisition method based on self-organizing competitive neural network | |
| CN109978279A (en) | Marine surface temperature regional prediction method | |
| Ma et al. | MM-RNN: A multimodal RNN for precipitation nowcasting | |
| CN115248992A (en) | Ocean three-dimensional temperature and salt space-time prediction method based on compressed excitation PredRNN | |
| Grayek et al. | Assessment of the Black Sea observing system. A focus on 2005-2012 Argo campaigns | |
| Pauthenet et al. | Four-dimensional temperature, salinity and mixed-layer depth in the Gulf Stream, reconstructed from remote-sensing and in situ observations with neural networks | |
| He et al. | Projecting three‐dimensional ocean thermohaline structure in the North Indian Ocean from the satellite sea surface data based on a variational method | |
| Wang et al. | The construction of sound speed field based on back propagation neural network in the global ocean | |
| CN103514377A (en) | Urban agglomeration land environment influence estimation method based on sky-land-biology | |
| Imbrie et al. | Transfer functions: calibrating micropaleontological data in climatic terms | |
| CN116739161A (en) | Sea surface temperature prediction method integrating empirical mode decomposition and Informir model | |
| CN114186483B (en) | Inversion method fusing buoy data and ocean satellite remote sensing image | |
| Yu et al. | Inversion of the three-dimensional temperature structure of mesoscale eddies in the Northwest Pacific based on deep learning | |
| CN112085779B (en) | Wave parameter estimation method and device | |
| Takano et al. | A method to estimate three-dimensional thermal structure from satellite altimetry data | |
| CN108920799B (en) | Two-dimensional hydrology-hydrodynamic unidirectional coupling method based on square adaptive grid | |
| Udaya Bhaskar et al. | Comparison between Argo-derived sea surface temperature and microwave sea surface temperature in tropical Indian Ocean | |
| ECCO Consortium | A Twenty-Year Dynamical Oceanic Climatology: 1994-2013. Part 1: Active Scalar Fields: Temperature, Salinity, Dynamic Topography, Mixed-Layer Depth, Bottom Pressure. | |
| Tokunaga et al. | OSnet: a neural network to predict ocean temperature and salinity | |
| Guo et al. | An inversion method of subsurface thermohaline field based on deep learning and remote sensing data | |
| Lu et al. | Dynamic Prediction of Full-Ocean Depth SSP by A Hierarchical LSTM: An Experimental Result |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |