Improving partial ice cover forecasts using machine learning, satellite, thermodynamic, and in situ observations
By combining satellite data and thermodynamic models with machine learning models, ice cover predictions are generated and combined. Optimization is achieved using error statistics and in-situ observation equipment, thus solving the inaccuracy problem of fluid dynamics models in ice cover prediction and realizing more accurate ice cover predictions.
Patent Information
- Application Number
- CN202111433899.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-12-01
- Filing Date
- 2021-11-29
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-11-29
AI Technical Summary
Existing fluid dynamics models lack accurate predictions of the spatial extent of ice cover when simulating water movement during periods of partial ice cover, resulting in inaccurate simulations.
By combining satellite data, thermodynamic models, and machine learning models, ice cover predictions are generated and combined. Error statistics are used to update model parameters, and in-situ observation equipment is used to optimize predictions, thereby improving the accuracy of ice cover predictions.
It improves the accuracy of partial ice cover prediction, especially the prediction of its expansion and contraction, and enhances the model's self-improvement capability.
Smart Images

Figure CN114582440B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates generally to artificial intelligence and machine learning systems and, more particularly, to methods and systems for predicting partial ice cover on a body of water. Background Art
[0002] Many lakes and parts of the oceans freeze over during the winter. Ice acts as a buffer between the atmosphere and the water, affecting heat flux and momentum, the exchange of gases, and the penetration of light. This has profound consequences for the movement of water and nutrients and their biological evolution. Reliable predictions of ice cover ensure more accurate simulations of fluid dynamics as well as nutrients, chemistry, and biology during the ice-covered months.
[0003] However, ice cover is usually partial, that is, part of the body of water is covered by ice, while the remainder is open water. Partial ice cover can exist for days, weeks, or even months, and the ice cover expands and contracts as the weather changes. During this period, some parts of the body of water are exposed to the atmosphere, while other parts of the body of water have this ice cover buffer. This has local effects on the water cycle, and if partial ice cover exists for many days, there are more general effects.
[0004] Currently, fluid dynamic models are used to simulate the movement of water. In order to accurately simulate the fluid dynamics during extended periods of partial ice cover, fluid dynamic models require knowledge of the spatial extent of the ice cover. Summary of the invention
[0005] According to various embodiments, a computing device, non-transitory computer-readable storage medium, and method are provided for making predictions with machine learning to determine fractional ice cover using at least one of satellite data, thermodynamics, and in situ observation systems.
[0006] In one embodiment, a computer-implemented method for predicting ice cover on a body of water includes: generating a first ice cover prediction with a thermodynamic module, and generating a second ice cover prediction with a machine learning module. Combining the first ice cover prediction with the second ice cover prediction to generate a combined ice cover prediction. Computing error statistics based on a comparison of the combined ice cover prediction with ice cover observations, and updating the combined ice cover prediction based on the error statistics.
[0007] In some embodiments, the thermodynamic module includes a fluid dynamics model and thermodynamic equations for generating the first ice cover prediction. In some embodiments, ice cover in the fluid dynamics model is updated based on error statistics.
[0008] In some embodiments, the machine learning module includes a trained model and a module for generating predicted local weather conditions. In some embodiments, a feedback loop is provided to the trained model based on error statistics.
[0009] In some embodiments, the method further comprises: applying weights and thresholds to the first ice cover prediction and the second ice cover prediction to generate a combined ice cover prediction. In some embodiments, the weights and thresholds are applied to each grid cell defining the surface area of the body of water. In some embodiments, the weights and thresholds for each grid cell are separately defined based on a predefined period of the year.
[0010] In some embodiments, a tree-based model is used to optimize the combined ice cover prediction.
[0011] In some embodiments, the method further comprises performing direct observations of ice cover at the body of water with an observation device. In some embodiments, a travel path for the observation device is optimized based on a combined ice cover prediction and observations of whether ice is present at grid cells defining a surface area of the body of water. In some embodiments, the observation device is an in-situ automated observation device.
[0012] According to various embodiments, a computer-implemented method for predicting ice cover on a body of water includes: generating a first ice cover prediction with a thermodynamic module, and generating a second ice cover prediction with a machine learning module. Combining the first ice cover prediction with the second ice cover prediction based on an initial set of weights and thresholds to generate a combined ice cover prediction. Computing error statistics based on a comparison of the combined ice cover prediction with ice cover observations from an automated observation device, and updating the weights and thresholds based on the error statistics. Generating an updated combined ice cover prediction using the updated weights and thresholds.
[0013] By means of the concepts discussed herein, systems and methods are provided that improve upon current methods for determining fractional ice cover. The systems and methods discussed herein can improve the accuracy of fractional ice cover predictions, including predictions of expansion and contraction of the fractional ice cover. As described below, the systems and methods can further improve predictions by using in situ observation systems, thereby confirming predictions from the model and including the possibility of improving the model itself.
[0014] These and other features will become apparent from the following detailed description of illustrative embodiments of the invention, which is to be read in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The accompanying drawings are illustrative embodiments. They do not show all embodiments. In addition or alternatively, other embodiments may be used. Details that may be obvious or unnecessary may be omitted to save space or for more effective description. Some embodiments may be implemented with additional components or steps and / or without all components or steps shown. When the same number appears in different drawings, it refers to the same or similar components or steps.
[0016] Figure 1 is a representation of various components used in a system for predicting ice cover on a body of water in accordance with an embodiment of the present disclosure.
[0017] Figure 2 is a graphical representation of a water body divided into grids, where a combined fluid dynamics model and machine learning model are used to predict an indication of ice cover.
[0018] Figure 3 is a diagram according to an illustrative embodiment. Figure 2 A graphical representation of how the fluid dynamics model and the machine learning model can be combined to generate a combined prediction.
[0019] Figure 4 is a diagram according to an illustrative embodiment. Figure 3 A graphical representation of the combined forecast and further illustrating how in situ autonomous observations can be used to update the combined forecast.
[0020] Figure 5 is a flow diagram showing actions involved in making and updating an ice cover forecast in accordance with an illustrative embodiment.
[0021] Figure 6 It can be used to implement Figure 2 Functional block diagram of the computer hardware platform of the ice cover prediction engine. DETAILED DESCRIPTION
[0022] In the following detailed description, many specific details are set forth by way of example to provide a thorough understanding of the relevant teachings. However, it should be appreciated that the present teachings may be practiced without these details. In other examples, well-known methods, processes, components and / or circuits have been described at a relatively high level without detail to avoid unnecessarily obscuring aspects of the present teachings.
[0023] Unless otherwise specifically stated, and as will be apparent from the following description and claims, it should be understood that throughout the specification, descriptions utilizing terms such as “processing,” “computing,” “calculating,” “determining,” etc. refer to actions and / or processes of a computer or computing system or similar electronic computing device that manipulates and / or transforms data represented as physical (such as electronic) quantities within the computing system's registers and / or memories into other data similarly represented as physical quantities within the computing system's memories, registers, or other such information storage, transmission, or display devices.
[0024] As discussed in more detail below, the present disclosure generally relates to a system for predicting ice cover on a body of water by using a hybrid model approach that utilizes both a physics-based model and a machine learning model to predict actual ice cover. The machine learning model can be enhanced by observations. The model can include feedback (as described in more detail below) to improve the predictions over time. The present disclosure further relates to a method for combining predictions from a machine learning model trained on satellite-measured ice cover concentrations with predictions from a thermodynamic-based ice model driven by a fluid dynamics model to determine ice concentrations and mark grid cells in the fluid dynamics as having ice cover, thereby simulating fluid dynamics during periods of ice cover.
[0025] There are existing methods that use machine learning to predict ice concentrations using satellite observations and weather data. Further, there are existing methods that predict ice formation in fluid dynamic models via numerical discretization of water thermodynamics using differential equations. The methods and systems of the present invention combine both machine learning models and fluid dynamic models and use optimized feedback (as described below) to improve simulations that employ fluid dynamic models.
[0026] Reference will now be made in detail to the details illustrated in the accompanying drawings and discussed below.
[0027] refer to Figure 1 and Figure 2 , the body of water 100 may include ice cover 102, which is represented by a slash on the body of water 100. Thermodynamic module 104 of ice prediction engine 106 may receive weather data from weather station 112 and may use this data in fluid dynamics model 114 along with thermodynamic equations 116 to make ice cover predictions. Machine learning module 108 may use satellite data from satellite 110 to correlate observed ice concentrations with local climate to make predictions about future ice cover. Thermodynamic module 104 may perform calculations of fluid dynamics model 114 up to every time step (such as every three seconds), resulting in a high temporal resolution of thermodynamic module 104. Information for ice cover predictions may be obtained at a high spatial resolution.
[0028] Machine learning module 108 may use the correlation between observed ice concentration and local climate to make predictions. Calculations by machine learning module 108 may be performed as new weather information becomes available and may be gated by the selection of feature values used in training the model, such as with average daily temperature. In some embodiments, trained model 202 of machine learning module 108 may generate predicted local weather conditions 204 to generate ice cover prediction 230.
[0029] Reference now Figure 2 and Figure 3, ice prediction engine 106 may generate ice cover predictions 220, 230 for each of fluid dynamics module 104 and machine learning module 108. These ice cover predictions 220, 230 may be combined (as discussed in more detail below) to generate a combined prediction 240. Combined prediction 240 may account for ice cover on body of water 100, as indicated by the grid portions marked with Xs.
[0030] refer to Figure 4 The combined prediction 240 can be compared with observations of actual ice cover and error statistics can be calculated in action 400. Figure 3 ) thresholds and parameters (as described below) may be adjusted based on the calculated error statistics in act 402. At act 404, combined prediction 240 may be updated.
[0031] In some embodiments, at action 406, the in-situ autonomous observation device 120 (see Figure 1 , also referred to as observation equipment 120 ) may be used to observe actual ice cover, and such observations may be used in act 400 to compare these observations with combined predictions 240 to calculate error statistics.
[0032] The observation device 120 can be used to supplement remote sensing data (e.g., from satellites 110). The observation device 120 can take various forms, such as an underwater observer (UAV), a surface floating device (such as a floating dock), an aircraft (such as a drone), a snowmobile, an amphibious vehicle, etc. The observation device 120 can be equipped with sensors to detect the presence of ice cover at any given location. In some embodiments, the observation device 120 can be optimized to explore areas based on the combined forecast 240, where areas adjacent to ice cover can be selectively explored, which can be advantageous for larger bodies of water where exploration of the entire body of water may be impractical or unavailable within a given time period. In some embodiments, the observation device 120 may include a machine learning algorithm that provides the ability to use weather data in conjunction with the combined forecast 240 to determine a desired search space.
[0033] With reference to the foregoing description of an exemplary ice prediction engine 106 (see Figure 2 ), it may be helpful to now consider a high-level discussion of the example process. To this end, Figure 5Illustrative processes associated with generating ice cover forecasts through both a thermodynamic module and a machine learning module are presented. Process 500 is shown in a logical flow diagram as a collection of boxes, which represent a sequence of operations that can be implemented in hardware, software, or a combination thereof. In the context of software, boxes represent computer-executable instructions that, when executed by one or more processors, perform the recited operations. Generally, computer-executable instructions may include routines, programs, objects, components, data structures, etc. that perform functions or implement abstract data types. In each process, the order in which the operations are described is not intended to be construed as a limitation, and any number of the described boxes may be combined in any order and / or executed in parallel to implement the process. For purposes of discussion, reference is made to FIG. Figure 2 The process 500 is described with reference to the ice prediction engine 106 .
[0034] refer to Figure 5 At act 502, process 500 for predicting ice cover of a body of water may generate an ice / ice-free prediction at each grid cell at each time step using thermodynamic module 104. At act 504, process 500 may use machine learning module 106 to generate an ice / ice-free prediction at each grid cell using, for example, a deep neural network.
[0035] At act 506, weights and thresholds may be applied to combine the predictions of acts 502 and 504 to generate a combined prediction at act 508. The problem of applying weights and thresholds at act 506 may be suitable for a tree-based model (such as random forest, XGboost, etc.) to optimize the predictions, although other algorithmic approaches may also be used. Random weights and thresholds may be selected to provide a first prediction. The tree-based model may be trained using observations, and at act 510, the weights and thresholds for each grid cell are adjusted based on error statistics. At act 512, the weights and thresholds may be adjusted based on, for example, the L2 norm to update the predictions at act 514. In some embodiments, hyperparameter adjustment of the tree-based model (e.g., adjusting the number of branches) may be performed for optimizing the performance of the tree-based model, thereby adjusting the weights and thresholds at act 512. The adjusted weights and thresholds may be defined for a given grid cell of the surface area of a body of water, for example, at Figure 3 Grid cells shown in the combined ice cover forecast 240 in FIG. Subsequent training can be performed as more observations become available. In some embodiments, weights and thresholds for a given grid cell may be defined for a given time of year, where, as an example, different weights and thresholds may be provided for each month because the predictive skill of ice cover from the fluid dynamics model and the satellite-based machine learning model may differ seasonally.
[0036] When the combined forecast is generated (at acts 506 and 508) or updated (at act 514), the updated ice cover forecast may be fed back to the thermodynamic module 104 at act 516 for use in the fluid dynamics model 114. In some embodiments, based on the error statistics determined at act 510, the ice forecast engine 106 (see Figure 2 ) updates the trained model of the machine learning module 106 so that the predictions made at action 504 are more accurate.
[0037] In some embodiments, process 500 may use an in-situ automated observation device, wherein at act 518, an optimal path of travel for the observation device may be determined, and at act 520, the observation device may be used to observe whether ice is present at a given location on the body of water. The optimal path of travel may be adjusted to accurately define the boundary between ice cover and absence of ice on the body of water.
[0038] In some embodiments, the trained model 202 may be trained with hindcasts performed by the fluid dynamics model 114. In other words, the output from the fluid dynamics model 114 may be used as feature values in the machine learning module 106, creating a link between the two modules 104, 106.
[0039] In some embodiments, ice cover prediction engine 106 may be used for iceberg tracking. For example, satellite-derived trajectories (target values) may be used to train machine learning module 106 on possible iceberg trajectories given local weather information (feature values). Predictions of iceberg trajectories may be obtained by combining a fluid dynamics model and a machine learning module in a manner similar to that described above.
[0040] Figure 6 A functional block diagram of a computer hardware platform 600 that can be used to implement a specially configured computing device that can host an ice cover prediction engine 106 is provided. As described above, the ice cover prediction engine 106 can include a thermodynamic module 104 and a machine learning module 108. The thermodynamic module 104 can include a fluid dynamics model 114 and thermodynamic equations 116 as described above. The machine learning module 108 can include a trained model 202 and a predicted weather condition 204, as described above. In particular, Figure 6 A network or host computer platform 600 is shown, which can be used to implement Figure 2 A suitably configured engine 106 is shown.
[0041] Computer platform 600 may include a central processing unit (CPU) 610 , a hard disk drive (HDD) 620 , random access memory (RAM) and / or read-only memory (ROM) 630 , a keyboard 650 , a mouse 660 , a display 670 , and a communication interface 680 , which are connected to a system bus 640 .
[0042] In one embodiment, HDD 620 has the capability to include stored programs that can perform various processes in the manner described herein, such as ice cover prediction engine 106 .
[0043] In one embodiment, the communication interface 680 may be configured to receive data from the satellite 110, the weather station 112, the observation device 120, etc. Figure 1 shown.
[0044] The description of various embodiments of the present teachings has been provided for the purpose of illustration, but is not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terms used herein are selected to best explain the principles of the embodiments, practical applications, or technical improvements existing in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.
[0045] Although the above has described what is considered to be the best state and / or other examples, it should be understood that various modifications may be made therein, and the subject matter disclosed herein may be implemented in various forms and examples, and the teachings may be applied to many applications, only some of which are described herein. The appended claims are intended to claim any and all applications, modifications and variations that fall within the true scope of the present teachings.
[0046] The components, steps, features, objectives, benefits and advantages discussed herein are merely illustrative. None of them or the discussion related to them is intended to limit the scope of protection. Although various advantages have been discussed herein, it will be understood that not all embodiments must include all advantages. Unless otherwise stated, all measurements, values, ratings, positions, sizes, dimensions and other specifications set forth in this specification, including the attached claims, are approximate and not exact. They are intended to have a reasonable range consistent with the functions to which they are related and the customary practices of the field to which they belong.
[0047] Many other embodiments are also contemplated. These include embodiments with fewer, additional and / or different components, steps, features, objects, benefits and advantages. These also include embodiments in which components and / or steps are arranged and / or ordered in different ways.
[0048] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It will be understood that each frame of the flowchart and / or block diagram and the combination of frames in the flowchart and / or block diagram can be implemented by computer-readable program instructions.
[0049] These computer-readable program instructions can be provided to a processor of a suitably configured computer, a special-purpose computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device create a device for implementing the functions / actions specified in one or more boxes of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, which can guide the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable storage medium having the instructions stored therein includes an article of manufacture, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes of the flowchart and / or block diagram.
[0050] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps will be performed on the computer, other programmable apparatus, or other device to produce a computer-implemented process, so that the instructions executed on the computer, other programmable apparatus, or other device implement the functions / actions specified in one or more boxes of the flowchart and / or block diagram.
[0051] The call flow, flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each frame in the flow chart or block diagram can represent a module, segment or part of an instruction, which includes one or more executable instructions for realizing the specified logical function. In some alternative embodiments, the function noted in the frame may not occur in the order noted in the figure. For example, the two frames shown in succession can actually be performed substantially simultaneously, or these frames can sometimes be performed in reverse order, depending on the function involved. It will also be noted that each frame of the block diagram and / or flow chart illustration and the combination of the frames in the block diagram and / or flow chart illustration can be realized by a dedicated hardware-based system that performs a specified function or action or performs a combination of special hardware and computer instructions.
[0052] Although the foregoing has been described in conjunction with exemplary embodiments, it should be understood that the term "exemplary" is meant only as an example, rather than the best or optimal. Except as immediately stated above, no component, step, feature, object, benefit, advantage, or equivalent is intended or should be construed as being dedicated to the public, whether or not stated in the claims.
[0053] It should be understood that, unless specific meanings are elaborated in addition herein, the terms and expressions used herein have the common meanings consistent with these terms and expressions of respective investigation and research fields corresponding to them. Relational terms such as "first" and "second" etc. can be used only to distinguish one entity or action from another, without requiring or implying any actual such relationship or order between these entities or actions. The terms "include", "comprise" or any other variation thereof are intended to cover non-exclusive inclusions, so that the process, method, article or device comprising a series of elements not only include those elements, but also may include other elements that are not clearly listed or are inherent to such process, method, article or device. In the absence of further constraints, the element with "one" or "an" in front does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.
[0054] The abstract of the present disclosure is provided to allow the reader to quickly determine the essence of the present technical disclosure. It should be understood that it is not used to interpret or limit the scope or meaning of the claims. In addition, in the foregoing detailed description, it can be seen that in order to streamline the present disclosure, various features are grouped together in various embodiments. This method of disclosure should not be interpreted as reflecting the intention that the claimed embodiments have more features than the features explicitly stated in each claim. On the contrary, as reflected in the attached claims, the subject matter of the invention lies in less than all the features of a single disclosed embodiment. Therefore, the attached claims are incorporated into the detailed description here, wherein each claim is independently a subject matter claimed separately.
Claims
1. A computer-implemented method for predicting ice cover on a body of water, comprising: Generate the first ice cover forecast using the thermodynamic module; Generate a second ice cover forecast using a machine learning module; combining the first ice cover prediction and the second ice cover prediction based on an initial set of weights and thresholds to generate a combined ice cover prediction; calculating error statistics based on a comparison of the combined ice cover forecast with ice cover observations from automated observing equipment; Based on the error statistics, updating the weights and thresholds; as well as Use the updated weights and thresholds to generate an updated combined ice cover forecast.
2. The computer-implemented method of claim 1 , further comprising: When a discrepancy regarding ice cover occurs based on the combined ice cover prediction and observations of ice cover made with the automated observation device, a travel path of the automated observation device is updated.
3. The computer-implemented method of claim 1 , wherein: The weights and thresholds are applied to each grid cell defining the surface area of the body of water.
4. The computer-implemented method of claim 1 , wherein: The weights and thresholds for each grid cell are defined separately based on predefined periods of the year.
5. The computer-implemented method of claim 1 , wherein: A tree-based model is used to optimize the combined ice cover forecast.
6. A computer program product comprising a computer readable storage medium tangibly embodying computer readable program code having computer readable instructions, the computer readable instructions when executed causing a computer device to perform the method of predicting ice cover on a body of water as recited in any one of claims 1-5.
7. A computer system for predicting ice cover on a body of water, comprising means for implementing the steps of the method according to any one of claims 1 to 5.
Citation Information
Patent Citations
NRIET fog forecasting method based on machine learning
CN108375808A
Wide and deep machine learning models
CN109196527A
Multivariate statistical prediction method for polar region fixed ice thickness
CN109829569A
Overhead transmission line icing prediction method based on weather and geographical environments
CN111539842A
Unmanned underwater vehicles, locations of their docking stations, and their programmed routes
US20160167755A1
