Land water storage prediction method and system based on artificial intelligence, and electronic device

Abstract

The application discloses a land water storage prediction method and system based on artificial intelligence and electronic equipment. The prediction method comprises the following steps: collecting atmospheric reanalysis data sets and gravity satellite land water storage data, and obtaining various global climate model output data; a long short-term memory model for simulating long series of land water storage is constructed; the long short-term memory model is driven by the corrected grid global climate model output data to simulate the first prediction result of the land water storage in the future prediction period; a hook-shaped response function of the land water storage and the dew point temperature is fitted, and the second prediction result of the land water storage in the future prediction period is obtained; and an emergent constraint model for correcting the prediction result is constructed. The application can fully reflect the spatio-temporal evolution law of the land water storage under the influence of climate change, significantly prolong the prediction period of the land water storage, improve the prediction effect, and solve the defect that the prediction of the land water storage has great uncertainty in the prior art.

Description

Technical Field

[0001] This invention relates to the field of land water storage prediction technology, specifically to an artificial intelligence-based land water storage prediction method, system, and electronic equipment. Background Technology

[0002] Strengthening scientific research on the mechanisms of water resource response to climate change is of paramount importance.

[0003] Terrestrial Water Storage (TWS) is the total amount of continental water stored in tree canopies, snow, rivers, lakes and reservoirs, wetlands, soil, and groundwater. It represents the total available water volume on land and is crucial to water resource availability and sustainable socio-economic development. In March 2002, the Gravity Recovery and Climate Experiment (GRACE) satellite was successfully launched, providing a continuous and highly accurate direct observation method for acquiring large-scale global observations of material migration on the Earth's surface. Based on the gravity field model calculated from GRACE satellite signals, it can extract information on the variation of the Earth's monthly gravity field on a 300 km × 300 km spatial scale. After deducting the influence of crustal material movement, atmospheric motion, ocean currents, and tides, it can effectively reflect gravity changes caused by snow, surface water, soil water, groundwater, and anthropogenic factors, comprehensively monitoring TWS signals. In May 2018, following a one-year hiatus of the GRACE gravity satellite, the GRACE-FO (GRACE Follow-On) gravity satellite was successfully launched, continuing the scientific observation mission of the GRACE satellite. The GRACE / GRACE-FO gravity satellite effectively solved problems such as shallow ground observation range, uneven spatial distribution, and difficulty in data acquisition, demonstrating great potential in tracking regional and global drought events.

[0004] However, gravity satellites can only provide inversion datasets of land water storage from 2002 to the present. Land water storage is affected by the combined effects of climate change and human activities, making it difficult to predict land water storage under future climate change.

[0005] Existing research generally uses a combination of hydrological models and global climate models to predict future land water storage, but existing prediction methods have the following problems:

[0006] First, the prediction period for future land water storage is relatively short;

[0007] Second, the uncertainty of the forecast was not effectively controlled, resulting in mediocre forecast performance;

[0008] Third, under normal circumstances, it fails to fully reflect the spatiotemporal evolution of subsurface water storage under global climate change. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention proposes an artificial intelligence-based method, system, and electronic device for predicting terrestrial water storage. This method can fully reflect the spatiotemporal evolution of terrestrial water storage under the influence of global climate change, significantly extending the prediction cycle and improving the prediction effect. It also solves the problem of large uncertainty in the prediction of terrestrial water storage in existing technologies.

[0010] To achieve the above objectives, the present invention provides an artificial intelligence-based method for predicting terrestrial water storage, characterized by the following steps:

[0011] S1) Select a watershed for land water storage prediction, collect historical atmospheric reanalysis datasets and gravity satellite land water storage data within the watershed, and obtain various global climate model output data for historical periods and future prediction periods. The atmospheric reanalysis datasets and global climate model output data include meteorological and hydrological observation data.

[0012] S2) Using the meteorological and hydrological observation data in step S1), the relative humidity and specific humidity of each grid point in the watershed are derived, and the relative humidity and specific humidity data within the corresponding spatial range are selected by the spatial sliding window method; a long short-term memory model for simulating long series terrestrial water storage is constructed, and the long short-term memory model is used to reconstruct the long series terrestrial water storage dataset;

[0013] The Long Short-Term Memory model is a model showing the relationship between each driving factor of each grid point and the terrestrial water storage of the gravity satellite.

[0014] The driving factors include meteorological and hydrological observation data for each grid point, as well as the derived relative humidity and specific humidity;

[0015] S3) Compare meteorological and hydrological observation data from the same historical period with the output data of various global climate models, correct the output data of various global climate models for the future prediction period, and use the corrected gridded global climate model output data to drive the long short-term memory model in step S2) to simulate the first prediction result of land water storage for the future prediction period.

[0016] S4) Based on the dew point temperature in the reconstructed long-series terrestrial water storage dataset and meteorological and hydrological observation data from the same historical period, fit the hook-shaped response function of terrestrial water storage and dew point temperature in the watershed; deduce the dew point temperature in the future prediction period, and obtain the second prediction result of terrestrial water storage in the future prediction period based on the hook-shaped response function.

[0017] S5) For each type of global climate model, calculate the average of the first and second prediction results, and denote it as the preliminary prediction value of land water storage in the future prediction period; construct an emergence constraint model to correct the prediction results of watershed land water storage changes in the future prediction period, and solve the parameters of the emergence constraint model. The emergence constraint model is a relationship model between the preliminary prediction value of land water storage in the future prediction period and the historical dew point temperature change trend under each type of global climate model.

[0018] Furthermore, in S1), the meteorological and hydrological observation data are monthly-scale meteorological and hydrological observation data, including 2m air temperature, air pressure, dew point temperature, snowfall, precipitation, runoff depth, shortwave radiation, and longwave radiation; the various global climate model output data include monthly average air temperature, monthly relative humidity, monthly precipitation, monthly snowfall, monthly runoff depth, monthly average specific humidity, monthly shortwave radiation intensity, and monthly longwave radiation for historical periods and future forecast periods.

[0019] Furthermore, in S2), the relative humidity and specific humidity of each grid point are obtained by the following formula.

[0020] ;

[0021] ;

[0022] In the formula,

[0023] RH The relative humidity at each grid point,

[0024] T dew Dew point temperature,

[0025] T 2m The air temperature is 2m.

[0026] e sat The mass of water vapor is obtained using the Clausius-Clapeyron thermodynamic equation.

[0027] q For wetness,

[0028] p represents the near-surface air pressure;

[0029] The Clausius-Clapeyron thermodynamic equation is as follows:

[0030] ;

[0031] In the formula,

[0032] e sat For water vapor quality,

[0033] T Temperature is the input variable to the Clausius-Clapeyron thermodynamic equation.

[0034] e so The second integral constant,

[0035] L v Let be the latent heat of vaporization constant.

[0036] R v The constant of water vapor.

[0037] T 0 is the first integral constant.

[0038] Furthermore, in S2), the Long Short-Term Memory model is represented as follows:

[0039] ;

[0040] In the formula,

[0041] TWS(t) express t The short-term memory model simulates the terrestrial water storage.

[0042] F Representing the Long Short-Term Memory model,

[0043] QM(t) express t Input variables at time,

[0044] QM(t-1) express t-1 Input variables at time,

[0045] QM(t-2) express t-2 Input variables at any given time.

[0046] Furthermore, in S3), the correction of various global climate model output data for future forecast periods is performed using the following formula.

[0047] ;

[0048] ;

[0049] In the formula,

[0050] T Represents air temperature, specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation.

[0051] PRepresents precipitation, snowfall, and runoff depth.

[0052] adj Represents the corrected series,

[0053] d Represents daily data,

[0054] GCM Indicates the type of global climate pattern.

[0055] obs Meteorological and hydrological observation data representing the fifth-generation atmospheric reanalysis dataset.

[0056] Q Representing each quantile,

[0057] ref This represents a historical reference period.

[0058] Furthermore, in S4), the steps for constructing the hook-shaped response function of terrestrial water storage and dew point temperature in the watershed include: representing the nonlinear relationship between terrestrial water storage and dew point temperature through a box scaling function; then fitting the dew point temperature-terrestrial water storage series using a local weighted sliding regression smoothing method, thereby drawing the hook-shaped structure of the predicted dew point temperature-terrestrial water storage in the watershed.

[0059] Furthermore, in S4), the steps for estimating the dew point temperature for the future forecast period include: substituting the daily average air temperature into the Clausius-Clapeyron equation to obtain the saturated vapor pressure; then using the saturated vapor pressure and relative humidity to obtain the actual vapor pressure; substituting the actual vapor pressure into the Clausius-Clapeyron equation; obtaining the historical daily average dew point temperature series output by various global climate models through numerical calculation methods; and then using Thiessen polygons to calculate the dew point temperature for the future forecast period.

[0060] The actual water vapor pressure is obtained by the following formula:

[0061] ;

[0062] In the formula,

[0063] e act To calculate the actual water vapor pressure,

[0064] e sa The saturated vapor pressure is obtained from the daily average temperature using various global climate models.

[0065] RH is the daily average relative humidity output by various global climate models.

[0066] Furthermore, in S5), the emergence constraint model is specifically as follows:

[0067] ;

[0068] In the formula,

[0069] WA Fut This represents the corrected forecast of future changes in watershed terrestrial water storage.

[0070] a and b The parameters represent the emergent constraint model.

[0071] This indicates the annual average dew point temperature trend over the historical period of this grid point.

[0072] An artificial intelligence-based land water storage prediction system, applicable to the aforementioned artificial intelligence-based land water storage prediction method, is characterized by including a data acquisition module, a training module, a correction module, a modeling module, and a prediction module.

[0073] The acquisition module is used to acquire atmospheric reanalysis datasets and global climate model data to form a meteorological and hydrological dataset.

[0074] The training module is used to determine the relative humidity and specific humidity at a set time, and to train a long short-term memory model to simulate long series of terrestrial water storage based on the relative humidity and specific humidity and the meteorological and hydrological dataset. The long short-term memory model is then used to reconstruct the long series of terrestrial water storage dataset.

[0075] The correction module is used to drive the long short-term memory model with the corrected gridded global climate model output data to simulate the first prediction result of land water storage in the future prediction period.

[0076] The establishment module is used to fit a hook-shaped response function of land water storage and dew point temperature in the watershed based on the reconstructed long-series land water storage dataset and temperature data; to deduce the dew point temperature in the future prediction period; and to obtain a second prediction result of land water storage in the future prediction period based on the hook-shaped response function.

[0077] The prediction module is used to construct an emergent constraint model based on the first and second prediction results of the land water storage in the future prediction period. The emergent constraint model is used to correct the prediction results of the changes in watershed land water storage in the future prediction period, thereby reducing the uncertainty of the prediction of land water storage in the future prediction period.

[0078] An electronic device is characterized in that it includes a processor, a communication interface, a memory, a communication bus, and a computer program stored in the memory and executable on the processor. The processor, the communication interface, and the memory communicate with each other via the communication bus. The processor executes the program to implement the artificial intelligence-based land water storage prediction method described above.

[0079] The advantages of this invention are:

[0080] The land water storage prediction method, system, and electronic equipment provided by this invention combine global climate models, deep learning, gravity satellites, bias correction methods, and emergent constraint methods to provide important and highly operable reference for risk assessment and early warning of watershed floods and droughts under climate change, and provide engineering reference value for responding to future climate disasters, scientifically formulating emission reduction strategies, and supporting sustainable socio-economic development.

[0081] This invention takes into account the impact of climate change under global climate models, uses thermodynamic prediction theory, and adopts an emergent constraint model, all of which are pioneering. Moreover, the land water storage prediction in this invention refers to long-term prediction under climate change, with a time span that can reach the end of this century, significantly extending the prediction period and improving the results.

[0082] This invention relates to an artificial intelligence-based method, system, and electronic device for predicting terrestrial water reserves. It can fully reflect the spatiotemporal evolution of terrestrial water reserves under the influence of global climate change. It not only significantly extends the prediction cycle of terrestrial water reserves but also improves the prediction effect, thus solving the problem of large uncertainty in the prediction of terrestrial water reserves in existing technologies. Attached Figure Description

[0083] Figure 1 This is a flowchart of the artificial intelligence-based land water storage prediction method in this invention;

[0084] Figure 2 This is a schematic diagram of the spatial sliding window method in this invention;

[0085] Figure 3 This is a schematic diagram of the hook-shaped structure used in this invention to predict the dew point temperature-land water storage of a watershed.

[0086] Figure 4 This is a schematic diagram illustrating the prediction of watershed terrestrial water storage using global climate models and hook structures in this invention.

[0087] Figure 5 This is a schematic diagram of the structure of the artificial intelligence-based land water storage prediction system in this invention;

[0088] Figure 6 This is a schematic diagram of the electronic device in this invention.

[0089] In the diagram: Acquisition module 1, training module 2, correction module 3, establishment module 4 and prediction module 5, processor 810, communication interface 820, memory 830, and communication bus 840. Detailed Implementation

[0090] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0091] This invention discloses an artificial intelligence-based method for predicting terrestrial water storage, comprising the following steps:

[0092] S1) Select a watershed for land water storage prediction, collect historical atmospheric reanalysis datasets and gravity satellite land water storage data within the watershed, and obtain various global climate model output data for historical periods and future prediction periods. The atmospheric reanalysis datasets and global climate model output data include meteorological and hydrological observation data.

[0093] Specifically, the meteorological and hydrological observation data are monthly-scale meteorological and hydrological observation data, including 2m air temperature, air pressure, dew point temperature, snowfall, precipitation, runoff depth, shortwave radiation, and longwave radiation; the various global climate model output data include historical and future forecast monthly average air temperature, monthly relative humidity, monthly precipitation, monthly snowfall, monthly runoff depth, monthly average specific humidity, monthly shortwave radiation intensity, and monthly longwave radiation.

[0094] In this embodiment, for each grid point of the land water storage prediction basin, monthly meteorological and hydrological observation data from 1950 to 2022 were collected from the European Centre for Medium-Range Weather Forecasts' Generation 5 Atmospheric Reanalysis Dataset (ERA5-Land).

[0095] For each global climate model, monthly mean temperature, monthly relative humidity, monthly precipitation, monthly snowfall, monthly runoff depth, monthly mean specific humidity, monthly shortwave radiation intensity, and monthly longwave radiation for the historical period (1950-2014) are obtained. For the future period (2015-2100), monthly mean temperature, monthly relative humidity, monthly precipitation, monthly snowfall, monthly runoff depth, monthly mean specific humidity, monthly shortwave radiation intensity, and monthly longwave radiation for each global climate model under the shared socio-economic pathway SSP585 are obtained. This embodiment specifically selects 32 global climate models.

[0096] For acquiring satellite gravity data to invert terrestrial water storage, major international institutions such as the Potsdam Geoscience Center in Germany, the Jet Propulsion Laboratory (JPL) at Caltech, the Center for Space Research at the University of Texas at Austin (CSR), and NASA's Goddard Space Flight Center (GSFC) are responsible for data interpretation and publishing monthly-scale global gravity field output data. This invention simultaneously utilizes the latest sixth-generation (RL06) products from JPL, CSR, and GSFC. These three data sources have different spatial resolutions and all provide mascon monthly-scale equivalent water heights (subtracting the 2004–2009 mean field) based on the gravity field solution of lumped mass blocks, ultimately outputting a long-sequence TWSA dataset. To account for the uncertainties that may be caused by different data products, the three sets of GRACE / GRACE-FO gravity satellite datasets were interpolated to a 0.5°×0.5° spatial grid, and the average value of each product was taken at each time step to finally obtain the monthly dataset of land water storage from 2002 to 2022.

[0097] Finally, the output data from ERA5-Land and the global climate model were interpolated to a 0.5°×0.5° spatial grid.

[0098] S2) Using the meteorological and hydrological observation data in step S1), the relative humidity and specific humidity of each grid point in the watershed are derived, and the relative humidity and specific humidity data within the corresponding spatial range are selected by the spatial sliding window method; a long short-term memory model for simulating long series of terrestrial water storage is constructed, and the long short-term memory model is used to reconstruct the long series of terrestrial water storage dataset.

[0099] The Long Short-Term Memory (LSTM) model is a model relating each driving factor at each grid point to the land water storage of gravity satellites. The driving factors include meteorological and hydrological observation data at each grid point, as well as the derived relative humidity and specific humidity.

[0100] By substituting the 2-meter air temperature and dew point temperature of meteorological and hydrological observation samples at each grid point in the land water storage prediction basin into the Clausius-Clapeyron thermodynamic equation as input variables, the relative humidity (RH) at each grid point was calculated. q The ratio of water vapor mass to total air mass is given by the ERA5 near-surface pressure. p Derivation of dew point temperature.

[0101] Specifically, the relative humidity and specific humidity of each grid point are obtained by the following formula.

[0102] ;

[0103] ;

[0104] In the formula,

[0105] RH The relative humidity at each grid point,

[0106] T dew Dew point temperature,

[0107] T 2m The air temperature is 2m.

[0108] e sat The mass of water vapor is obtained using the Clausius-Clapeyron thermodynamic equation.

[0109] q For wetness,

[0110] p represents the near-surface air pressure;

[0111] The Clausius-Clapeyron thermodynamic equation is as follows:

[0112] ;

[0113] In the formula,

[0114] e sat For water vapor quality,

[0115] T Temperature is the input variable to the Clausius-Clapeyron thermodynamic equation.

[0116] e so The second integral constant is taken as 611 Pa.

[0117] L v Let be the latent heat of vaporization constant, taken as 2.5 × 10⁻⁶. 6 J·kg -1 ,

[0118] R v The gas constant for water vapor is taken as 461 J·kg⁻¹. -1 ·K -1 ,

[0119] T 0 Let K be the first integration constant, taken as 273.16 K.

[0120] like Figure 2 As shown, for each grid point within the predicted watershed, a 5×5 spatial sliding window threshold is set, and the window slides sequentially according to the range of this interval, selecting the data in the grid points involved in this spatial range as the input of the long short-term memory model. This method can improve the robustness of the machine learning model.

[0121] In this embodiment, the driving factors are data from 2002 to 2022, specifically including variables obtained or derived from ERA5-Land: including air temperature, relative humidity, snowfall, precipitation, runoff depth, shortwave radiation, and longwave radiation.

[0122] For each grid point, a two-month lag is considered; that is, for each month, the driving factors of that month and the preceding 1-2 months are selected as inputs to the Long Short-Term Memory (LSTM) model. An LSTM model is constructed using driving factors from 2002 to 2022 and GRACE gravity satellite data to simulate terrestrial water storage.

[0123] Specifically, the Long Short-Term Memory model is represented as follows:

[0124] ;

[0125] In the formula,

[0126] TWS(t) express t The short-term memory model simulates the terrestrial water storage.

[0127] F Representing the Long Short-Term Memory model,

[0128] QM(t) express t Input variables at time,

[0129] QM(t-1) express t-1 Input variables at time,

[0130] QM(t-2) express t-2 Input variables at any given time.

[0131] Furthermore, the long short-term memory model of each grid point is calibrated using the gradient descent method to obtain parameters. Then, the long short-term memory model is driven by a long series of datasets since 1950 to reconstruct a long series of terrestrial water storage datasets since 1950.

[0132] S3) Compare meteorological and hydrological observation data from the same historical period with output data from various global climate models, correct the output data of various global climate models for the future prediction period, and use the corrected gridded global climate model output data to drive the long short-term memory model in step S2) to simulate the first prediction result of land water storage for the future prediction period.

[0133] In this embodiment, a series of meteorological simulations under climate change scenarios are obtained based on the Global Climate Model (GCM) ensemble and quantile bias correction method. The differences between the output variables of the GCMs and the observed meteorological variables at each quantile (0.01-0.99) are calculated, and these differences are removed from each quantile of the future scenarios output by the GCMs to obtain the future climate predictions of the corrected GCMs.

[0134] Specifically, the corrections to the output data of various global climate models for future forecast periods are performed using the following formula.

[0135] ;

[0136] ;

[0137] In the formula,

[0138] T Represents air temperature, specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation.

[0139] P Represents precipitation, snowfall, and runoff depth.

[0140] adj Represents the corrected series,

[0141] d Represents daily data,

[0142] GCM Indicates the type of global climate pattern.

[0143] obs This represents meteorological and hydrological observation data collected by the fifth-generation atmospheric reanalysis dataset ERA5-Land.

[0144] Q Representing each quantile,

[0145] ref This represents a historical reference period.

[0146] Based on the above results, the long short-term memory model calibrated in step S2) driven by the corrected gridded simulated meteorological data is used to simulate the land water storage series of each grid point under future scenarios. Then, the multi-year average land water storage value of each grid point under the climate change scenario (2071-2100) is estimated. Finally, the Thiessen polygon method is used to calculate the average land water storage value in the predicted watershed, which is recorded as the first prediction result of land water storage in the future period.

[0147] The Thiessen polygon method is used to derive the watershed average monthly series of ERA5-Land and global climate model output data, and the watershed average monthly series is calculated based on the land water storage reconstruction series obtained from each grid point in step S2). The main idea is to transform the grid point dataset into a watershed average dataset.

[0148] S4) Based on the dew point temperature in the reconstructed long-series terrestrial water storage dataset and meteorological and hydrological observation data from the same historical period, fit the hook-shaped response function of terrestrial water storage and dew point temperature in the watershed; deduce the dew point temperature for the future prediction period, and obtain the second prediction result of terrestrial water storage for the future prediction period based on the hook-shaped response function.

[0149] Specifically, the steps for constructing the hook-shaped response function of terrestrial water storage and dew point temperature in the watershed include: representing the nonlinear relationship between terrestrial water storage and dew point temperature through a box-element scaling function; then fitting the dew point temperature-terrestrial water storage series using a local weighted sliding regression smoothing method to draw the hook-shaped structure of the predicted dew point temperature-terrestrial water storage in the watershed.

[0150] Binning scaling functions can effectively characterize the nonlinear relationships between variables and have been widely used in recent years to analyze the response characteristics of extreme precipitation and runoff to global warming. In this embodiment, the basin-average monthly dew point temperature from the ERA5-Land dataset is used as the sorting criterion to divide the series of dew point temperature-land water storage combinations for the same month during the study period into 12 bins with the same sample size. The median daily air temperature is used to represent the dew point temperature of the bin, and the average watershed land water storage in each dew point temperature bin is used to represent the water resources, ultimately resulting in 12 sets of dew point air temperature-land water storage series.

[0151] Previous studies have found that precipitation and runoff generally exhibit a hook-like structure with a near-surface temperature trajectory of "first rising and then falling." This invention uses a locally weighted moving regression smoothing method to fit 12 sets of dew point temperature-terrestrial water storage series, thereby drawing a hook-like structure for predicting dew point temperature-terrestrial water storage in a watershed. The temperature at the inflection point of the hook-like structure is called the peak temperature. Figure 3 As shown, a schematic diagram of the hook-shaped structure of dew point temperature-land water storage in the predicted watershed is presented. In the diagram, T... pp This indicates the peak temperature.

[0152] Specifically, in S4), the steps for estimating the dew point temperature for the future forecast period include: substituting the daily average air temperature into the Clausius-Clapeyron equation to obtain the saturated vapor pressure; then using the saturated vapor pressure and relative humidity to obtain the actual vapor pressure; substituting the actual vapor pressure into the Clausius-Clapeyron equation; obtaining the historical daily average dew point temperature series output by various global climate models through numerical calculation methods; and then using Thiessen polygons to calculate the dew point temperature for the future forecast period.

[0153] The actual water vapor pressure is obtained by the following formula:

[0154] ;

[0155] In the formula,

[0156] e act To calculate the actual water vapor pressure,

[0157] e sa The saturated vapor pressure is obtained from the daily average temperature using various global climate models.

[0158] RH is the daily average relative humidity output by various global climate models.

[0159] In this embodiment, Thiessen polygons are used to calculate and predict the multi-year average dew point temperature of the watershed over the historical period (1985-2014), denoted as T. obs Then, for each global climate model, the multi-year mean dew point temperature for the historical period (1985–2014) within the basin is predicted using Thiessen polygons, denoted as T. his Furthermore, the multi-year average dew point temperature (T0) within the watershed under future climate change scenarios (2071-2100) was calculated using Thiessen polygons. fut .

[0160] like Figure 4 As shown, a schematic diagram illustrating the prediction of watershed dew point water storage using global climate models and hook structures is presented. (T) obs Projecting the data onto the hook-shaped structure drawn in step S4), the corresponding watershed dew point water storage is obtained, denoted as WA. obs Then, for each global climate model, T is further obtained. obs +(T fut -T his The available water volume in the watershed corresponding to point ) is denoted as WA. fut1 .

[0161] Using the above method, we can obtain the predicted values ​​of dew point water storage for the future forecast period under 32 global climate models, which are denoted as the second prediction result of terrestrial water storage for the future forecast period.

[0162] S5) For each type of global climate model, calculate the average of the first and second prediction results, and denote it as the preliminary prediction value of land water storage in the future prediction period; construct an emergence constraint model to correct the prediction results of watershed land water storage changes in the future prediction period, and solve the parameters of the emergence constraint model. The emergence constraint model is a relationship model between the preliminary prediction value of land water storage in the future prediction period and the historical dew point temperature change trend under each type of global climate model.

[0163] The preliminary forecast of terrestrial water storage for the future forecast period is denoted as WA. fut1 First, the historical annual average dew point temperature variation trend of each global climate model is calculated (denoted as ). Furthermore, by integrating data from 32 global climate models, 32 sets of historical annual mean dew point temperature variation trends and future WA values ​​were obtained. fut1 Pairing combinations.

[0164] Specifically, the constructed emergent constraint model is as follows:

[0165] ;

[0166] In the formula,

[0167] WA Fut This represents the corrected forecast of future changes in watershed terrestrial water storage.

[0168] a and b The parameters represent the emergent constraint model.

[0169] This indicates the annual average dew point temperature trend over the historical period of this grid point.

[0170] In this embodiment, the least squares method is used to solve for the parameters of the emergence constraint model. a and b Based on the reanalysis data of the ERA5-Land dataset, the trend of dew point temperature variation during 1985-2014 was deduced, and this trend term was substituted into the constructed emergent constraint model, as follows:

[0171] ;

[0172] In the formula,

[0173] WAFut This represents the corrected forecast of future changes in watershed terrestrial water storage.

[0174] a and b The parameters represent the emergent constraint model.

[0175] OBStrend This indicates the historical annual average dew point temperature trend obtained from the ERA5-Land dataset within this grid point.

[0176] Finally, the model was corrected using an emergent constraint model. Used to guide the comprehensive management of water resources in predicted watersheds.

[0177] The present invention also designs an artificial intelligence-based land water storage prediction system, which is applicable to the artificial intelligence-based land water storage prediction method described above, and includes a data acquisition module 1, a training module 2, a correction module 3, a setup module 4 and a prediction module 5.

[0178] The acquisition module 1 is used to acquire atmospheric reanalysis datasets and global climate model data to form a meteorological and hydrological dataset.

[0179] The training module 2 is used to determine the relative humidity and specific humidity at a set time, and to train a long short-term memory model for simulating long series of terrestrial water storage based on the relative humidity and specific humidity and the meteorological and hydrological dataset, and to reconstruct the long series of terrestrial water storage dataset through the long short-term memory model.

[0180] The correction module 3 is used to drive the long short-term memory model with the corrected gridded global climate model output data to simulate the first prediction result of land water storage in the future prediction period.

[0181] The establishment module 4 is used to fit the hook-shaped response function of land water storage and dew point temperature in the watershed based on the reconstructed long series land water storage dataset and temperature data; to deduce the dew point temperature in the future prediction period; and to obtain the second prediction result of land water storage in the future prediction period based on the hook-shaped response function.

[0182] The prediction module 5 is used to construct an emergence constraint model based on the first and second prediction results of the land water storage in the future prediction period. The emergence constraint model is used to correct the prediction results of the changes in watershed land water storage in the future prediction period, thereby reducing the uncertainty of the prediction of land water storage in the future prediction period.

[0183] For specific implementation methods of the land water storage prediction system of this invention, please refer to the specific implementation methods of the land water storage prediction method described above. To avoid redundancy, they will not be repeated here.

[0184] This invention also designs an electronic device, including a processor 810, a communication interface 820, a memory 830, a communication bus 840, and a computer program stored in the memory 830 and executable on the processor 810. The processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 executes the program to implement the artificial intelligence-based land water storage prediction method described above. Figure 6 As shown.

[0185] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0186] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0187] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0188] This invention relates to an artificial intelligence-based method, system, and electronic device for predicting terrestrial water reserves. It can fully reflect the spatiotemporal evolution of terrestrial water reserves under the influence of global climate change. It not only significantly extends the prediction cycle of terrestrial water reserves but also improves the prediction effect, thus solving the problem of large uncertainty in the prediction of terrestrial water reserves in existing technologies.

[0189] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for predicting terrestrial water storage based on artificial intelligence, characterized in that, Includes the following steps: S1) Select a watershed for land water storage prediction, collect historical atmospheric reanalysis datasets and gravity satellite land water storage data within the watershed, and obtain various global climate model output data for historical periods and future prediction periods. The atmospheric reanalysis datasets and global climate model output data include meteorological and hydrological observation data. S2) Using the meteorological and hydrological observation data in step S1), the relative humidity and specific humidity of each grid point in the watershed are derived, and the relative humidity and specific humidity data within the corresponding spatial range are selected by the spatial sliding window method; a long short-term memory model for simulating long series terrestrial water storage is constructed, and the long short-term memory model is used to reconstruct the long series terrestrial water storage dataset; The Long Short-Term Memory model is a model of the driving factors of each grid point and the land water storage system of gravity satellites. The driving factors include meteorological and hydrological observation data for each grid point, as well as the derived relative humidity and specific humidity; S3) Compare meteorological and hydrological observation data from the same historical period with the output data of various global climate models, correct the output data of various global climate models for the future prediction period, and use the corrected gridded global climate model output data to drive the long short-term memory model in step S2) to simulate the first prediction result of land water storage for the future prediction period. The specific steps include obtaining a series of meteorological simulations under climate change scenarios based on the Global Climate Model (GCM) ensemble and quantile bias correction method, calculating the differences between the output variables of GCMs and the observed meteorological variables at each quantile of 0.01-0.99, and removing these differences from each quantile of the future scenarios output by GCMs to obtain the future corrected climate predictions of GCMs. Step S2) uses a calibrated long short-term memory model driven by gridded simulated meteorological data to simulate the land water storage series at each grid point under future scenarios. Then, the multi-year average land water storage value at each grid point under the climate change scenario is derived. Finally, the Thiessen polygon method is used to calculate the average land water storage value within the predicted watershed, which is recorded as the first prediction result of land water storage in the future period. Step S4) Based on the dew point temperature in the reconstructed long-series land water storage dataset and meteorological and hydrological observation data from the same historical period, a hook-shaped response function of land water storage and dew point temperature within the watershed is fitted. The dew point temperature in the future prediction period is derived, and the second prediction result of land water storage in the future prediction period is obtained based on the hook-shaped response function. The steps to derive the future forecast dew point temperature include: substituting the daily average air temperature into the Clausius-Clapeyron equation to obtain the saturated vapor pressure; then using the saturated vapor pressure and relative humidity to obtain the actual vapor pressure; substituting the actual vapor pressure into the Clausius-Clapeyron equation; using numerical calculation methods to derive the historical daily average dew point temperature series output by various global climate models; and finally using Thiessen polygons to calculate the future forecast dew point temperature. The actual water vapor pressure is obtained by the following formula: ； In the formula, e act To calculate the actual water vapor pressure, e sa The saturated vapor pressure is obtained from the daily average temperature using various global climate models. RH represents the daily average relative humidity output by various global climate models; The specific steps for calculating the future forecast dew point temperature using Thiessen polygons include: Thiessen polygons were used to calculate and predict the multi-year average dew point temperature of the watershed over historical periods, denoted as T. obs Then, for each global climate model, the Thiessen polygon is used to calculate and predict the multi-year mean dew point temperature of the basin for historical periods, denoted as T. his The multi-year average dew point temperature under future climate change scenarios within the watershed is predicted using Thiessen polygons and denoted as T. fut ; T obs Projecting the data onto the hook-shaped structure drawn in step S4), the corresponding watershed dew point water storage is obtained, denoted as WA. obs Then, for each global climate model, obtain T obs +(T fut -T his The available water volume in the watershed corresponding to point ) is denoted as WA. fut1 ; Using the above method, the predicted values ​​of dew point water storage in the future forecast period under various global climate models can be obtained, which are denoted as the second prediction result of terrestrial water storage in the future forecast period. S5) For each type of global climate model, calculate the average of the first and second prediction results, and denote it as the preliminary prediction value of land water storage in the future prediction period; construct an emergence constraint model to correct the prediction results of watershed land water storage changes in the future prediction period, and solve the parameters of the emergence constraint model. The emergence constraint model is a relationship model between the preliminary prediction value of land water storage in the future prediction period and the historical dew point temperature change trend under each type of global climate model.

2. The method for predicting terrestrial water storage based on artificial intelligence according to claim 1, characterized in that: In S1), the meteorological and hydrological observation data are monthly meteorological and hydrological observation data, including 2m air temperature, air pressure, dew point temperature, snowfall, precipitation, runoff depth, shortwave radiation and longwave radiation; the output data of various global climate models include monthly average air temperature, monthly relative humidity, monthly precipitation, monthly snowfall, monthly runoff depth, monthly average specific humidity, monthly shortwave radiation intensity and monthly longwave radiation for historical periods and future forecast periods.

3. The method for predicting terrestrial water storage based on artificial intelligence according to claim 2, characterized in that: In S2), the relative humidity and specific humidity of each grid point are obtained by the following formula. ； ； In the formula, RH The relative humidity at each grid point, T dew Dew point temperature, T 2m The air temperature is 2m. e sat The mass of water vapor is obtained using the Clausius-Clapeyron thermodynamic equation. q For wetness, p represents the near-surface air pressure; The Clausius-Clapeyron thermodynamic equation is as follows: ； In the formula, e sat For water vapor quality, T Temperature is the input variable to the Clausius-Clapeyron thermodynamic equation. e so The second integral constant, L v Let be the latent heat of vaporization constant. R v The constant of water vapor. T 0 is the first integral constant.

4. The method for predicting terrestrial water storage based on artificial intelligence according to claim 3, characterized in that: In S2), the Long Short-Term Memory model is represented as follows: ； In the formula, TWS(t) express t The short-term memory model simulates the terrestrial water storage. F Representing the Long Short-Term Memory model, QM(t) express t Input variables at time, QM(t-1) express t-1 Input variables at time, QM(t-2) express t-2 Input variables at any given time.

5. The method for predicting terrestrial water storage based on artificial intelligence according to claim 4, characterized in that: In S3, the corrections to the output data of various global climate models for the future forecast period are performed using the following formula. ； ； In the formula, T Represents air temperature, specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation. P Represents precipitation, snowfall, and runoff depth. adj Represents the corrected series, d Represents daily data, GCM Indicates the type of global climate pattern. obs Meteorological and hydrological observation data representing the fifth-generation atmospheric reanalysis dataset. Q Representing each quantile, ref This represents a historical reference period.

6. The method for predicting terrestrial water storage based on artificial intelligence according to claim 1, characterized in that: In S4), the steps for constructing the hook-shaped response function of terrestrial water storage and dew point temperature in the watershed include: representing the nonlinear relationship between terrestrial water storage and dew point temperature through a box scaling function; then fitting the dew point temperature-terrestrial water storage series using a local weighted sliding regression smoothing method to draw the hook-shaped structure of the predicted dew point temperature-terrestrial water storage in the watershed.

7. The method for predicting terrestrial water storage based on artificial intelligence according to claim 6, characterized in that: In S5), the emergence constraint model is specifically as follows: ； In the formula, WA Fut This represents the corrected forecast of future changes in watershed terrestrial water storage. a and b The parameters represent the emergent constraint model. This indicates the annual average dew point temperature trend over the historical period of this grid point.

8. An artificial intelligence-based land water storage prediction system, applicable to the artificial intelligence-based land water storage prediction method as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition module (1), a training module (2), a correction module (3), a setup module (4), and a prediction module (5); The acquisition module (1) is used to acquire atmospheric reanalysis datasets and global climate model data to form a meteorological and hydrological dataset; The training module (2) is used to determine the relative humidity and specific humidity at a set time, and to train a long short-term memory model for simulating long series of terrestrial water storage based on the relative humidity and specific humidity and the meteorological and hydrological dataset, and to reconstruct the long series of terrestrial water storage dataset through the long short-term memory model. The correction module (3) is used to drive the long short-term memory model with the corrected gridded global climate model output data to simulate the first prediction result of land water storage in the future prediction period. The establishment module (4) is used to fit the hook response function of land water storage and dew point temperature in the watershed based on the reconstructed long series land water storage dataset and temperature data; to deduce the dew point temperature in the future prediction period; and to obtain the second prediction result of land water storage in the future prediction period based on the hook response function. The prediction module (5) is used to construct an emergence constraint model based on the first and second prediction results of the land water storage in the future prediction period. The emergence constraint model is used to correct the prediction results of the change of watershed land water storage in the future prediction period, thereby reducing the uncertainty of the prediction of land water storage in the future prediction period.

9. An electronic device, characterized in that: The system includes a processor (810), a communication interface (820), a memory (830), a communication bus (840), and a computer program stored in the memory (830) and executable on the processor (810). The processor (810), the communication interface (820), and the memory (830) communicate with each other through the communication bus (840). The processor (810) executes the program to implement the artificial intelligence-based land water storage prediction method as described in any one of claims 1 to 7.

Images