Estimation device, estimation method, and estimation program

Abstract

This estimation device 1 comprises: a simulation unit 11 for time-evolving state data indicating the state of the sea; a first extraction unit 13 for extracting sea surface information relating to the state of the sea surface from the time-evolved state data of the sea; and an estimation unit 14 for learning sub-sea surface information relating to the sub-sea surface state on the basis of the sea surface information and spatio-temporal information about the sea. The estimation unit 14 estimates the sub-sea surface information at a desired time on the basis of the sea surface information at the desired time and the spatio-temporal information about the sea.

Description

Estimation Device, Estimation Method, and Estimation Program

[0001] The present disclosure relates to an estimation device, an estimation method, and an estimation program.

[0002] There is a technique for estimating the upwelling area of the sea surface using sea surface information (see Non-Patent Document 1).

[0003] Zineb El Abidi, et al., “An Efficient Detection of Moroccan Coastal Upwelling Based on Fusion of Chlorophyll-a and Sea Surface Temperature Images With a New Validation Index”, IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 18, NO. 8, August 2021, p.1322-p.1326

[0004] However, it only estimated the state of the sea surface.

[0005] The present disclosure has been made in view of the above circumstances, and an object of the present disclosure is to provide a technique capable of estimating the state under the sea surface.

[0006] An estimation device according to an aspect of the present disclosure includes a simulation unit that time-evolves state data indicating the state of the sea, a first extraction unit that extracts sea surface information regarding the state of the sea surface from the time-evolved sea state data, and an estimation unit that learns subsea information regarding the state under the sea surface based on the sea surface information and the spatio-temporal information of the sea. The estimation unit estimates the subsea information at a desired time based on the sea surface information and the spatio-temporal information of the sea at the desired time.

[0007] An estimation method according to an aspect of the present disclosure is an estimation method performed by an estimation device, in which state data indicating the state of the sea is time-evolved, sea surface information regarding the state of the sea surface is extracted from the time-evolved sea state data, subsea information regarding the state under the sea surface is learned based on the sea surface information and the spatio-temporal information of the sea, and the subsea information at a desired time is estimated based on the sea surface information and the spatio-temporal information of the sea at the desired time.

[0008] An estimation program according to one aspect of the present disclosure causes a computer to perform the following processes: evolving state data indicating the state of the sea over time; extracting sea surface information relating to the state of the sea surface from the evolved state data of the sea; learning subsurface information relating to the state below the sea surface based on the sea surface information and spatiotemporal information of the sea; and estimating subsurface information at a desired time based on the sea surface information and spatiotemporal information of the sea at a desired time.

[0009] According to this disclosure, it is possible to estimate the conditions below the sea surface.

[0010] Figure 1 shows an example of the functional block configuration of the estimation device. Figure 2 shows an example of the operation of the estimation device during the learning process. Figure 3 shows an example of the operation of the estimation device during the inference process. Figure 4 shows an example of the hardware configuration of the estimation device.

[0011] Embodiments of this disclosure will be described below with reference to the drawings. In the drawings, the same parts are denoted by the same reference numerals and their descriptions are omitted.

[0012] [Summary of this Disclosure] This disclosure discloses a technology for estimating the internal state below the sea surface from sea surface information. For example, it estimates a three-dimensional ocean current field as the internal state below the sea surface based on satellite observation data such as sea surface height (SSH), sea surface temperature (SST), and chlorophyll a (Chl-a), as well as seabed topography data recorded by government agencies, etc. This can improve the range of applications for sea surface information.

[0013] Furthermore, this disclosure discloses a technique that enables the determination of upwelling origins from various sources by reanalyzing inference data. Conventionally, when a neural network is trained to determine only the presence or absence of upwelling, it is only possible to distinguish upwelling based on a specific mechanism / origin included in the dataset used for training. However, by retaining the three-dimensional ocean current field below the sea surface (i.e., the neural network variables for estimating the ocean current field) as an inference result, it becomes possible to determine upwelling due to different mechanisms / origins through reanalysis.

[0014] [Configuration of the Estimation Device] Figure 1 is a diagram showing an example of the functional block configuration of the estimation device 1 according to this embodiment.

[0015] Figure 1 also shows the input device 2, storage device 3, and display device 4. The input device 2 is an observation device such as a satellite device for observing sea surface temperature, etc., and a recording device for recording seabed topography, etc. The storage device 3 is a database. The display device 4 is a display or monitor.

[0016] The functions of the estimation device 1 will now be described. The estimation device 1 comprises a simulation unit 11, a preprocessing unit 12, a first extraction unit 13, an estimation unit 14, a second extraction unit 15, and a learning unit 16.

[0017] The simulation unit 11 receives state data (such as potential temperature) and spatiotemporal information of the ocean (such as seabed topography) from the input device 2, and has the function of evolving the ocean state data over time using ocean simulation technology.

[0018] The first extraction unit 13 has the function of extracting sea surface information (SST, etc.) related to the state of the sea surface from the sea state data that has been evolved over time by the simulation unit 11.

[0019] The estimation unit 14 is a deep neural network.

[0020] During the learning process, the estimation unit 14 has the function of learning subsurface information (ocean current field, velocity field, etc.) relating to the state below the sea surface, based on the sea surface information extracted by the first extraction unit 13 and the spatiotemporal information of the sea input from the input device 2.

[0021] The estimation unit 14 has the function of estimating sub-sea information at a desired time based on sea surface information and spatiotemporal information of the sea at that desired time during the inference process.

[0022] The estimation unit 14 has the function of reading the learned neural network variables from the storage device 3 during inference processing and using the learned neural network variables to estimate the underwater information at the desired time.

[0023] In other words, the estimation unit 14, which is a deep neural network, takes sea surface information and spatiotemporal information of the ocean as input, and uses the variables of the trained neural network to estimate sub-sea information within the neural network and outputs it as the estimated sub-sea information.

[0024] The second extraction unit 15 has the function of extracting subsurface information regarding the state below the sea surface from the ocean state data that has been evolved over time by the simulation unit 11.

[0025] The learning unit 16 uses the subsea information extracted by the second extraction unit 15 as training data, learns the variables of the neural network for learning or estimating the subsea information so that the difference between the subsea information learned or estimated by the estimation unit 14 and the training data becomes small, and stores the learned neural network variables in the storage device 3.

[0026] The preprocessing unit 12 has a function to perform preprocessing (such as normalization) on sea surface information at a desired time.

[0027] [Operation of the estimation device during the learning process] Figure 2 shows an example of the operation of the estimation device 1 during the learning process.

[0028] Step S101; The simulation unit 11 receives ocean condition data observed at time Tx and spatiotemporal information of the ocean from the input device 2. Ocean condition data includes, for example, velocity field, salinity, potential temperature, wave height, and Chl-a. Spatiotemporal information of the ocean includes, for example, seabed topography, latitude, longitude, and date and time (year, month, day, hour, minute, second).

[0029] The simulation unit 11 generates three-dimensional state data that has evolved over time from time Tx to time Ty using ocean simulation technology. Examples of ocean simulation technology include MITgcm, ROMS, and HYCOM. Examples of three-dimensional state data include velocity field, salinity, potential temperature, wave height, and Chl-a.

[0030] Step S102; The first extraction unit 13 extracts sea surface information relating to the state of the sea surface from the three-dimensional state data that has been time-evolved by the simulation unit 11. Sea surface information includes, for example, SSH, SST, and Chl-a.

[0031] In other words, the first extraction unit 13 extracts only the first layer (sea surface) from the three-dimensional state data in the depth direction from time Tx to time Ty. That is, the first extraction unit 13 converts the three-dimensional state data containing multiple layers into two-dimensional state data of only the first layer, and outputs two-dimensional state data of only the sea surface from time Tx to time Ty.

[0032] Step S103; The estimation unit 14 learns subsurface information about the conditions below the sea surface based on the sea surface information extracted by the first extraction unit 13 and the spatiotemporal information of the sea input from the input device 2. Subsurface information includes, for example, the velocity field, potential temperature, salinity, and Chl-a.

[0033] Step S104; The second extraction unit 15 extracts subsurface information relating to the state below the sea surface from the three-dimensional state data that has been time-evolved by the simulation unit 11. Subsurface information includes, for example, the velocity field, potential temperature, salinity, and Chl-a.

[0034] In other words, the second extraction unit 15 outputs three-dimensional state data that extracts everything except the first layer (sea surface) in the depth direction from time Tx to time Ty.

[0035] Step S105; The learning unit 16 uses the subsea information extracted by the second extraction unit 15 as training data, and learns the variables of the neural network for learning or estimating the subsea information so that the difference between the subsea information learned or estimated by the estimation unit 14 and the training data becomes small, and stores the learned neural network variables in the storage device 3.

[0036] [Operation of the estimation device during inference processing] Figure 3 shows an example of the operation of the estimation device during inference processing.

[0037] Step S201; The preprocessing unit 12 performs preprocessing on the sea surface information at time Tx. Preprocessing includes, for example, unifying the resolution, normalizing the values, and handling missing values. The sea surface information is, for example, SSH, SST, and Chl-a.

[0038] Step S202; The estimation unit 14 reads the learned neural network variables from the storage device 3 and uses the learned neural network variables to estimate the subsurface information at time Tx based on the preprocessed sea surface information and the spatiotemporal information of the sea. The spatiotemporal information of the sea includes, for example, seabed topography, latitude, longitude, and date and time (year, month, day, hour, minute, second).

[0039] In other words, the deep neural network, which is the estimation unit 14, takes two-dimensional state data of the sea surface at time Tx as input and outputs the internal state below the sea surface at time Tx.

[0040] The estimation unit 14 outputs the estimated subsea information to the display device 4. Subsea information includes, for example, velocity field, potential temperature, salinity, and Chl-a.

[0041] [Effects] According to this embodiment, state data indicating the state of the sea is evolved over time, sea surface information regarding the state of the sea surface is extracted from the time-evolved state data of the sea, subsurface information regarding the state below the sea surface is learned based on the sea surface information and spatiotemporal information of the sea, and subsurface information at a desired time is estimated based on the sea surface information and spatiotemporal information of the sea at a desired time, so that the state below the sea surface can be estimated.

[0042] [Other] This disclosure is not limited to the embodiments described above. This disclosure can be modified in numerous ways within the scope of the gist of this disclosure.

[0043] The estimation device 1 of this embodiment described above can be realized using a general-purpose computer system, for example, as shown in Figure 4, which includes a CPU 901, a memory 902, a storage 903, a communication device 904, an input device 905, and an output device 906. The memory 902 and the storage 903 are storage devices. In this computer system, each function of the estimation device 1 is realized when the CPU 901 executes a predetermined program loaded onto the memory 902.

[0044] The estimation device 1 may be implemented on a single computer. The estimation device 1 may be implemented on multiple computers. The estimation device 1 may be a virtual machine implemented on a computer.

[0045] The program for the estimation device 1 can be stored in a computer-readable recording medium such as a HDD, SSD, USB memory, CD, DVD, etc. The computer-readable recording medium is, for example, a non-transitory recording medium. The program for the estimation device 1 can also be distributed via a communication network.

[0046] 1 Estimation device 2 Input device 3 Storage device 4 Display device 11 Simulation unit 12 Preprocessing unit 13 First extraction unit 14 Estimation unit 15 Second extraction unit 16 Learning unit 901 CPU 902 Memory 903 Storage 904 Communication device 905 Input device 906 Output device

Claims

1. An estimation device comprising: a simulation unit that evolves state data indicating the state of the sea over time; a first extraction unit that extracts sea surface information relating to the state of the sea surface from the evolved state data of the sea over time; and an estimation unit that learns subsurface information relating to the state below the sea surface based on the sea surface information and spatiotemporal information of the sea, wherein the estimation unit estimates subsurface information at a desired time based on sea surface information and spatiotemporal information of the sea at a desired time.

2. The estimation device according to claim 1, further comprising: a second extraction unit that extracts subsurface information relating to the state below the sea surface from the sea state data; and a learning unit that learns variables for learning or estimating subsurface information such that the difference between the learned or estimated subsurface information and the extracted subsurface information becomes small, and stores the learned variables in a memory device, wherein the estimation unit estimates the subsurface information at the desired time using the learned variables read from the memory device.

3. An estimation method performed by an estimation device, comprising: evolving state data indicating the state of the sea over time; extracting sea surface information relating to the state of the sea surface from the evolved state data of the sea; learning subsurface information relating to the state below the sea surface based on the sea surface information and spatiotemporal information of the sea; and estimating subsurface information at a desired time based on the sea surface information and spatiotemporal information of the sea at a desired time.

4. An estimation program that causes a computer to perform the following processes: evolving state data indicating the state of the sea over time; extracting sea surface information regarding the state of the sea surface from the evolved state data of the sea over time; learning subsurface information regarding the state below the sea surface based on the sea surface information and spatiotemporal information of the sea; and estimating subsurface information at a desired time based on the sea surface information and spatiotemporal information of the sea at a desired time.

Images