Drought early warning method and device based on artificial intelligence

Abstract

The invention belongs to the technical field of data processing, and particularly provides a drought early warning method and device based on artificial intelligence, and the method comprises the steps: collecting the data of a drought early warning drainage basin; constructing a deep learning model considering a hydrological process for daily runoff simulation, and reconstructing a long series of daily runoff data; constructing a monthly-scale vegetation-meteorological hydrological drought regression model; calculating a comprehensive meteorological and hydrological drought index, extracting drought events, and constructing K artificial intelligence models; inputting meteorological and hydrological observation data into K artificial intelligence-based models for simulation, and deducing the weight of each artificial intelligence model by adopting a Bayesian mode averaging method; a joint distribution function of drought index observation and forecast values is constructed based on a Copula function, and an artificial intelligence-Bayesian coupling model is adopted to carry out drought forecast. According to the method, the problems of data scarcity, model uncertainty and the like in drought early warning are solved, and high-precision and quantifiable decision support is provided for water resource management and disaster prevention and reduction.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and in particular, relates to a drought early warning method and device based on artificial intelligence. Background Art

[0002] Droughts, with their complex causes, long timescales, and high destructive power, are a major constraint on the sustainable development of natural ecosystems and socio-economic development. They are often categorized as meteorological drought, hydrological drought, agricultural drought, and socio-economic drought. Meteorological and hydrological droughts significantly impact water resource management and various water-related activities, and are the most prominent categories of drought events that have garnered significant attention. Meteorological drought primarily refers to low precipitation, often triggered by abnormal atmospheric circulation. Meteorological drought is a trigger for hydrological drought. Low precipitation and high temperatures can cause droughts in soil moisture, river and lake runoff, and groundwater, further triggering hydrological drought. Severe droughts are often the result of the gradual development of meteorological and hydrological droughts. Drought is influenced by numerous factors, including hydrological, meteorological, and vegetation factors, and these factors are often closely correlated. Therefore, to quantitatively characterize the extent of water deficit, scholars at home and abroad have proposed a number of single-factor and multi-factor drought indices, including the Standardized Precipitation Index, Standardized Precipitation Evapotranspiration Index, Palmer Drought Index, and Standardized Runoff Index. Although the research objects and physical processes of each drought index are different, they mainly consider one or more meteorological and hydrological elements such as precipitation, evapotranspiration, runoff and soil moisture content, and cannot fully characterize the intrinsic physical characteristics of drought events.

[0003] Currently, how to fully consider diverse meteorological and hydrological factors and apply them to drought early warning remains a research challenge. Some researchers have commonly used meteorological forecast data to feed hydrological models for drought early warning. However, the deterministic forecasts produced by a single model often suffer from high uncertainty. Combining the forecasts of multiple hydrological models and employing statistical post-processing methods to construct hydrological ensemble forecasting methods is an important approach to quantifying or reducing the uncertainty of hydrological models. In recent years, some researchers have attempted to use data-driven models to establish the response relationship between factors such as climate variables and runoff data, demonstrating potential for application in drought early warning. Overall, the application of artificial intelligence technology to drought early warning remains underdeveloped. Furthermore, existing literature rarely considers the comprehensive consideration of multiple meteorological and hydrological factors for drought early warning and fails to fully consider the impact of human activities on drought formation mechanisms. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an artificial intelligence-based drought early warning method and device, which solves the problems of data scarcity, complex mechanisms, and model uncertainty in drought early warning, and provides high-precision and quantifiable decision support for water resources management and disaster prevention and mitigation, and has significant scientific significance and engineering application value.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a drought early warning method based on artificial intelligence, comprising the following steps: Step 1: Collect data for the drought warning basin: Collect ground observation data, including observation data from ground meteorological stations in the basin and daily runoff data from hydrological stations at the outlet section, collect meteorological and hydrological observation data and vegetation data from the reanalysis dataset, collect satellite-derived precipitation datasets, construct a statistical downscaling model, correct for precipitation magnitude and temperature simulation bias, and obtain a long series of downscaled meteorological observation data; Step 2: Based on the short-series runoff observation data of the watershed with scarce data and the downscaled long-series meteorological observation data, a deep learning model considering the hydrological process is constructed to simulate daily runoff and reconstruct the long-series daily runoff data; Step 3: Construct a monthly vegetation-meteorological-hydrological drought regression model and obtain model parameters; Step 4: Based on the model parameters of the vegetation-meteorological-hydrological drought regression model, the integrated meteorological-hydrological drought index is calculated, and then drought events are extracted and K artificial intelligence models are constructed to simulate drought events. Step 5: Construct an AI-Bayesian coupled model: Input the meteorological and hydrological observation data of the reanalysis dataset into K AI-based models for simulation, and use the Bayesian model averaging method to derive the weights of each AI model; Step 6: Construct a joint distribution function of drought index observation and forecast values based on the Copula function, and use the artificial intelligence-Bayesian coupling model to carry out drought forecasting.

[0006] In a preferred solution, in step 1, an equal rate correction method is used to construct a statistical downscaling model, and the operation steps are as follows: Assuming that the monthly biases between satellite-derived precipitation data or temperature data from reanalysis datasets and ground-based observations are consistent over time, we first calculate the correction factor for each month based on the ground-based observations and then apply this correction factor to the long series of simulated datasets for the same month. The correction factors include the precipitation deviation rate and the absolute temperature deviation. After obtaining the corresponding correction factors, the temperature series of the satellite-retrieved precipitation data and the reanalysis dataset are corrected using the following formula: (1); Where: N Indicates site m Total number of observation days per month; i Represents a daily precipitation or temperature series; Indicates that the satellite-retrieved precipitation m The original data series for the month, Indicates the corrected satellite-retrieved precipitation in mMonthly series; Indicates that the satellite-retrieved precipitation m Month i The original data of the day, Indicates that the weather station is m Month i Daily observed precipitation data; Reanalysis temperature in the m The original data series for the month, The corrected reanalysis temperature product is m Monthly series; Reanalysis temperature product in the m Month i The original data of the day, Indicates that the weather station is m Month i Observed temperature data for the day.

[0007] In a preferred solution, in step 2, performing daily runoff simulation by using a deep learning model considering hydrological processes includes the following steps: S2.1. Construct and calibrate the HBV model based on short-series runoff observations and downscaled long-series meteorological observations in data-scarce basins. S2.2. The first simulated daily runoff process is simulated based on the calibrated HBV model. The lag time that affects the measured daily runoff is determined by statistically analyzing the first simulated daily runoff process and the measured daily runoff process. N ; S2.3. Use the corrected long series of satellite-retrieved precipitation data and temperature data from the reanalysis dataset to drive the calibrated HBV model to obtain a long series of daily runoff simulations. The simulations are then divided into three types of daily runoff series with different magnitudes. S2.4. Based on the three types of daily runoff series with different magnitudes, a long-short-term memory neural network model is constructed respectively. The long series of daily runoff simulation is corrected to obtain the second simulated runoff series, which is expressed as: ] (2); Where, express t The second simulated runoff after time correction, Indicates that the HBV model is t Daily runoff simulation series at each moment, Indicates that the HBV model is t- 1 moment daily runoff simulation series, N Indicates the lag time that affects the daily measured runoff; F LSTM Represents an LSTM model.

[0008] In the preferred embodiment, in step S2.3, the method for dividing the simulation series into three types of daily runoff series of different magnitudes is: calculating the 30% quantile and 60% quantile of the daily runoff series, and dividing the daily runoff series into a first type below the 30% quantile, a second type between the 30% quantile and the 60% quantile, and a third type above the 60% quantile.

[0009] In a preferred solution, in step 3, the monthly runoff is calculated based on the long series of daily runoff data reconstructed in step 2, and the monthly precipitation is calculated based on the precipitation data of the reanalysis dataset, and a vegetation-meteorological-hydrological drought regression model is further constructed: (3); in, For the m Monthly relative NDVI anomaly for the entire study period, which is the vegetation data of the reanalysis dataset obtained in step 1; For the m reconstructed monthly runoff for the month; For the m Monthly precipitation totals; and are model parameters, and the least square method is used to solve these two parameters.

[0010] In a preferred embodiment, step 4 comprises the following steps: S4.1. Based on the long series of monthly runoff data reconstructed in step 3 and the basin-averaged monthly precipitation data, use the parameters of the obtained vegetation-meteorological-hydrological drought regression model to obtain the meteorological-hydrological comprehensive moisture index: (4); Where: For the m Monthly meteorological and hydrological comprehensive moisture index, and are the model parameters of the vegetation-meteorological-hydrological drought regression model; S4.2. The drought index is calculated using the meteorological and hydrological comprehensive moisture index as follows: (5); Where: Indicates the j The drought index for the month is as follows: the larger the index, the more severe the drought. MHwater j Indicates the j Monthly MHwater index; Respectively m The mean and standard deviation of monthly MHwater, k =1,2…,12, according to jDetermine the calendar month in which the month falls; S4.3. Based on the drought index derived in step 4.2, the run theory is used to extract drought events. Drought events are extracted when the set threshold is exceeded, and relative humidity and specific humidity are calculated. The calculation process is as follows: The Lausius-Clapeyron thermodynamic equation is defined as follows: (6); Where: is the first integration constant, which is 273.16 K; is the second integral constant, which is 611 Pa; is the latent heat of vaporization constant, take ; is the water vapor gas constant, take ; T is the air temperature, which is the input variable of the Clausius-Clapeyron thermodynamic equation; The 2-meter air temperature and dew point temperature of the meteorological and hydrological observation samples of each grid point in the study basin are substituted as input variables into the Clausius-Clapeyron thermodynamic equation to calculate the relative humidity of each grid point. RH , , where T dew is the dew point temperature, T 2m The air temperature at 2m; Specific humidity q is the ratio of water vapor mass to total mass of air mass, using the ground pressure in the reanalysis dataset p And dew point temperature, the formula is as follows: (7); S4.4. Use the Thiessen polygon method to derive average meteorological variables for watersheds with scarce data. Combine the drought events in the watershed with the meteorological variables for the watershed. The meteorological variables include the average monthly temperature, specific humidity, relative humidity, snowfall, precipitation, wind speed, shortwave radiation, and longwave radiation for the month of the drought event and the month before. Use a random forest model to screen key meteorological factors. S4.5. Based on the reanalysis dataset, K artificial intelligence models are constructed using the key meteorological factors selected in step S4.4 as input and the drought index during the drought event as the target variable.

[0011] In the preferred solution, in step S4.5, the K artificial intelligence models include: linear regression, logistic regression, decision tree, random forest, support vector machine, K nearest neighbor algorithm, naive Bayes classifier, neural network, convolutional neural network, recurrent neural network, long short-term memory network, generative adversarial network, Transformer, XGBoost and Markov decision process model.

[0012] In a preferred embodiment, step 5 comprises the following steps: S5.1. Use the meteorological data from the reanalysis dataset as input to drive the K artificial intelligence models constructed in step 4 to obtain a series of simulated magnitudes of drought events in the data-scarce basin. S5.2. Based on the simulation magnitude results of K artificial intelligence models and the measured drought index, the probability density function of the simulated drought index is constructed according to the Bayesian total probability formula. Let S be the simulated drought index, R = [D, O] represents the model input data, where D is the simulation series of each drought simulation model during the training period, and O is the measured series. The output results of K different artificial intelligence models are obtained by the Bayesian total probability formula. The probability density function of S is as follows: (8); Where: For the k AI models The probability density function of the simulated value S under the given data R; For a given training data R k The posterior probability density function of the artificial intelligence model, K represents the number of artificial intelligence models; S5.3. Determine the corresponding weights based on the relative contributions of each AI model to the drought index simulation effect, thereby establishing a Bayesian mode average correction model. The specific operations are as follows: First, the drought index observation series and the simulation series obtained by each artificial intelligence model are transformed into normality using the Box-Cox function. Then, the estimation results of multiple models are weighted averaged based on the normal linear distribution assumption: (9); Where: Indicates the mean , the variance is Normal distribution; E represents the expected value of the function, For the k The weight of an AI model.

[0013] In a preferred embodiment, step 6 includes the following steps: S6.1. Obtain weather forecast information for the basin where data are scarce from relevant weather forecasting agencies or departments, including monthly average temperature, specific humidity, relative humidity, snowfall, precipitation, wind speed, shortwave radiation, and longwave radiation for the next month; S6.2. Use the Thiessen polygon method to calculate the regional average weather forecast information for the watershed where data are scarce; S6.3. Reanalysis meteorological data for the past month are obtained from the reanalysis dataset. The Thiessen polygon method is used to obtain the regional average reanalysis meteorological data for the data-scarce basin. S6.4. Use the regional average meteorological forecast information for the next month for the data-scarce watershed obtained in step 6.2 and the regional average reanalysis meteorological data for the recent month for the data-scarce watershed obtained in step 6.3 as input to drive the artificial intelligence-Bayesian coupling model established in step 5 to obtain a drought index forecast for the next month. S6.5. Use the retrospective forecast method to construct the joint distribution function of the historical drought index observation and forecast values, and let H k 、 S k Respectively represent k The real drought index and the deterministic drought index at each moment, h k 、 s k Then H k 、 S k The realized value of and They are H k 、 S k The marginal distribution function of and The corresponding H k 、 S k The probability density function of , according to Sklar's theorem, there exists a n -Copula function C , so that the following equation holds: (10); Where: Copula function parameters; is the density function of the two-dimensional Copula function; represents the joint distribution function of the real drought index and the predicted drought index; S6.7, based on n - Use the Copula function to derive the posterior distribution function of the drought index and conduct probabilistic forecasting; Given forecast value s k ,exist S k = s k hour, Hk The conditional distribution function of for: (11); H k The density function is expressed as: (12); and When the deterministic forecast drought index information is given, the measured drought index h k The posterior probability distribution function and posterior probability density function are used to describe the uncertainty of the forecast; the expected value of the posterior probability density function is taken as the post-processing deterministic forecast result, and the confidence level is obtained at the same time The expected value of the posterior probability density function of the drought index under the predicted interval of the drought index h ke Solve the following equation: (13); S6.8. Based on the drought index probability forecast results of step S6.7, carry out drought warning.

[0014] The present invention also provides an artificial intelligence-based drought early warning device for executing the above-mentioned drought early warning method, comprising: The first determination module is used to collect data on drought warning basins; An analysis module is used to establish a watershed hydrological model and use a long-short-term memory model for correction, thereby constructing a deep learning model that considers hydrological processes; A module is constructed to build a monthly-scale vegetation-meteorological-hydrological drought regression model; The second determination module is used to build an artificial intelligence-Bayesian coupling model; The prediction module constructs the joint distribution function of drought index observation and forecast values based on the Copula function, and uses the artificial intelligence-Bayesian coupling model to carry out drought forecasting.

[0015] The artificial intelligence-based drought early warning method and device provided by the present invention have the following beneficial effects: 1. By coupling the physical mechanism of meteorological and hydrological drought formation, the uncertainty of hydrological processes, artificial intelligence and remote sensing inversion methods, it provides an important and highly operational reference for drought early warning in basins with scarce data, and provides engineering reference value for responding to climate disasters, disaster prevention and mitigation, and integrated water resources management.

[0016] 2. In step 1, multi-source data fusion and correction address data bottlenecks in data-scarce basins. By integrating ground-based observations, satellite-derived precipitation data, and reanalysis datasets (such as ERA5), a statistical downscaling model is constructed using an equal-rate correction method to effectively correct for biases in precipitation magnitude and temperature simulation. This operation addresses the problem of short, discontinuous series of ground-based observations in data-scarce basins, generating a long-term, continuous meteorological and hydrological dataset, providing a reliable data foundation for subsequent modeling.

[0017] 3. In step 2, a deep learning model was constructed by combining the HBV hydrological model with a long short-term memory (LSTM) neural network. This model calibrates daily runoff at different magnitudes, effectively capturing the impact of human activities on runoff. Hysteresis analysis was used to determine the optimal simulation duration, making the reconstructed long series of daily runoff data more accurate than actual hydrological processes, providing highly accurate hydrological input for drought event identification.

[0018] 4. In step 3, a vegetation-meteorological-hydrological drought regression model was constructed, coupling NDVI anomalies with monthly runoff and precipitation to quantify vegetation responses to meteorological-hydrological drought. This model, for the first time, incorporates vegetation indices into a drought assessment system, addressing the shortcomings of traditional drought indices that rely solely on meteorological or hydrological factors and more comprehensively reflecting the integrated response of ecosystems to water deficit.

[0019] 5. In step 4, the meteorological-hydrological integrated moisture index is calculated based on the regression model parameters, and the drought index VPDD is further derived. Drought events are extracted by combining the run theory, and thermodynamic parameters such as relative humidity and specific humidity are incorporated to comprehensively characterize the meteorological-hydrological-ecological coupling process of drought.

[0020] 6. In step 5, 15 AI models were constructed to simulate drought events, and the Bayesian Model Averaging (BMA) method was used to assign model weights. This method avoids the limitations of a single model and significantly improves simulation accuracy by weighted averaging the results of multiple models using the Box-Cox transformation and the normal linear assumption.

[0021] 7. In step 6, the posterior distribution of the drought index is constructed based on the Copula function, providing deterministic forecast values and probabilistic forecast intervals. This method addresses the blindness of traditional deterministic forecasts. By solving the Copula parameters using the Kendall rank correlation coefficient, it quantifies forecast uncertainty and provides more scientific early warning information to decision-makers. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A schematic diagram of the method flow provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the data of ground meteorological station observations, satellite remote sensing data and runoff data; Figure 3 Schematic diagram of the change in correlation coefficient between daily measured runoff and simulated runoff at different lag times; Figure 4 A block diagram of the drought early warning device provided in the embodiment; Figure 5 A block diagram of the execution device provided in the embodiment. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] Example 1: like Figure 1 As shown, a drought early warning method based on artificial intelligence includes the following steps: Step 1: Collect data for the drought warning basin: Collect ground observation data, including observation data from ground meteorological stations in the basin and daily runoff data from outlet section hydrological stations, collect meteorological and hydrological observation data and vegetation data from reanalysis datasets, collect satellite-derived precipitation datasets, construct a statistical downscaling model, correct precipitation magnitude and temperature simulation bias, and obtain a long series of downscaled meteorological observation data.

[0027] like Figure 2 As shown in FIG, a schematic diagram of ground meteorological station observations, reanalysis data and runoff data is given. For the watershed in the embodiment, there is a mismatch between ground meteorological data and runoff data, and they cannot be used directly to extract drought events. Therefore, it is necessary to use relatively complete satellite telemetry data and statistical downscaling models to obtain a long series of corrected satellite telemetry data, so as to establish an artificial intelligence model with the scarce runoff observation series and finally realize the reconstruction of the long series of drought events.

[0028] Therefore, collecting data on drought warning basins specifically includes the following steps: S1.1. Collect ground observation data from ground stations in data-scarce areas. Ground observation data include limited meteorological observation data, hydrological observation series data sets, and reanalysis data sets.

[0029] The reanalysis dataset uses the ERA5 product. For each grid point in the drought warning basin, hourly meteorological and hydrological observation data are collected from the fifth-generation atmospheric reanalysis dataset (ERA5) of the European Centre for Medium-Range Weather Forecasts. These data include 2-meter air temperature, dew point temperature, snowfall, precipitation, runoff depth, shortwave radiation, and leaf area index (LAI). These meteorological and hydrological observation data are then integrated into daily data.

[0030] The monthly average NDVI (Normalized Difference Vegetation Index) dataset of the drought warning basin was further collected as vegetation data. This dataset was obtained by averaging the data of the first 15 days and the last 15 days of each month from satellite inversion data, and then reconstructing the monthly average NDVI data using the maximum synthesis method. It was obtained through relevant public vegetation remote sensing data sources at home and abroad.

[0031] This embodiment collects the daily runoff observation data of the outlet section of the data-scarce watershed and the observation data of the ground meteorological station as the ground observation data.

[0032] Step 1.2: Use the equal-rate correction method to construct a statistical downscaling model to correct the precipitation magnitude and temperature simulation bias on a daily basis.

[0033] Specifically, the isorelative correction method assumes that the monthly biases between satellite-retrieved precipitation or reanalysis temperature data and ground-based observation series are consistent over different periods. First, the correction factor for each month is calculated based on ground-based observation information, and then the factor is applied to the long series of simulated data sets for the same month. In this embodiment, for each month at a meteorological station, the deviation ratio (precipitation) or absolute deviation (temperature) of that month is calculated based on the observation data and the satellite remote sensing inversion series. That is, the correction factor includes the precipitation deviation ratio and the absolute temperature deviation. After obtaining the corresponding correction factors, the satellite inverted precipitation data and the temperature series of the reanalysis dataset are corrected using the following formula: (1); Where: N Indicates site m Total number of observation days per month; i Represents a daily precipitation or temperature series; Indicates that the satellite-retrieved precipitation m The original data series for the month, Indicates the corrected satellite-retrieved precipitation in m Monthly series; Indicates that the satellite-retrieved precipitation m Month i The original data of the day, Indicates that the weather station is m Month i Daily observed precipitation data; Reanalysis temperature in the m The original data series for the month, The corrected reanalysis temperature product is m Monthly series; Reanalysis temperature product in the m Month i The original data of the day, Indicates that the weather station is m Month i Observed temperature data for the day.

[0034] The equal-rate correction method is used to establish the relationship between the surface meteorological station observation series and the ERA5 reanalysis dataset, and then this relationship is applied to the long series process to obtain the long series precipitation and temperature datasets at the surface meteorological station locations; the Thiessen polygon method is further used to obtain the basin-averaged long series precipitation and temperature data.

[0035] Step 2: Based on the short-series runoff observation data of the basin with scarce data and the downscaled long-series meteorological observation data, a deep learning model considering the hydrological process is constructed to simulate daily runoff and reconstruct the long-series daily runoff data.

[0036] Daily runoff simulation using a deep learning model that considers hydrological processes includes the following steps: S2.1. Based on the short-series runoff observation data and the downscaled long-series meteorological observation data in the data-scarce basin, construct the HBV model and calibrate the model parameters.

[0037] The HBV model is a conceptual hydrological model that simplifies actual hydrological processes using conceptual functions. By dividing the watershed into subbasins, the HBV can be considered a fully distributed or semi-distributed hydrological model. The HBV model primarily consists of four submodels: a snowmelt and snow accumulation model, a soil moisture and effective precipitation model, an evapotranspiration model, and a runoff model.

[0038] Calibration of HBV models and HBV models are routine techniques in the art.

[0039] S2.2. The first simulated daily runoff process is simulated based on the calibrated HBV model. The lag time that affects the measured daily runoff is determined by statistically analyzing the first simulated daily runoff process and the measured daily runoff process. N .

[0040] like Figure 3 As shown in the figure, a schematic diagram of the change in the correlation coefficient between the daily measured runoff and the first simulated daily runoff under different lag times is given; the correlation coefficient between the first simulated runoff and the measured runoff generally decreases gradually with the extension of the lag time; further, a suitable correlation threshold is selected to determine the simulated runoff time for establishing an artificial intelligence model with the measured runoff; for example, 0.5 can be taken.

[0041] S2.3. Use the corrected long series of satellite-retrieved precipitation data and the temperature data of the reanalysis dataset to drive the calibrated HBV model to obtain a long series of daily runoff simulations, and divide the simulation series into three categories of daily runoff series with different magnitudes.

[0042] The method for dividing the simulation series into three types of daily runoff series of different magnitudes is as follows: calculate the 30% quantile and 60% quantile of the daily runoff series, and divide the daily runoff series into the first type below the 30% quantile, the second type between the 30% quantile and the 60% quantile, and the third type above the 60% quantile.

[0043] S2.4. Based on three types of daily runoff series of different magnitudes, long short-term memory neural network (LSTM) models are constructed respectively to correct the long series of daily runoff simulation series obtained in step S2.3 to reduce the hydrological model error caused by human activities.

[0044] This embodiment constructs a long short-term memory neural network (LSTM) model with a three-layer neural network architecture to generalize the regulating and storage effects of dams, reservoirs, or water diversion projects on watersheds and improve the accuracy of hydrological simulations. This embodiment uses a neural network interval simulation mean method to independently run the neural network model multiple times and take the average value as the final simulation result to reduce uncertainty.

[0045] To solve the problem of nonlinear autoregressive exogenous input model (NARX) dynamic neural network in deep learning process (number of hidden layers) To address the exploding and vanishing gradient problems caused by the NARX neural network (layer 2), the LSTM (Long Short-Term Memory) neural network enhances the long-term memory capacity of the NARX neural network by introducing storage units—namely, input gates, forget gates, internal feedback connections, and output gates—into the hidden layers of the NARX neural network. This allows the network to selectively remember current information or forget past information (such as the rainfall-runoff mapping). In short, the LSTM neural network replaces each hidden layer in the NARX dynamic neural network with a storage unit with memory capabilities, referred to as an LSTM unit. Its input and output layers are the same as those of the NARX dynamic neural network.

[0046] Furthermore, the LSTM model is trained using the conventional technique in this field, the minimum batch gradient descent method.

[0047] Based on the long short-term memory neural network (LSTM) model, the corrected second simulated runoff series is expressed as: ] (2); Where, express t The second simulated runoff after time correction, Indicates that the HBV model is t Daily runoff simulation series at each moment, Indicates that the HBV model is t- 1 moment daily runoff simulation series, N Indicates the lag time that affects the daily measured runoff; F LSTM Represents an LSTM model.

[0048] Based on the above steps, a method for obtaining a second simulated runoff was constructed by using the HBV model to simulate the first runoff, and further correcting it using the response LSTM model according to the type of runoff on each day. In this embodiment, the combination of the HBV model, runoff type classification, and LSTM model is referred to as a deep learning model that considers hydrological processes.

[0049] Step 3: Construct a monthly-scale vegetation-meteorological-hydrological drought regression model and obtain model parameters.

[0050] The long series of precipitation and temperature data from step 1 are input into the HBV model and the deep learning model to reconstruct the long series of daily runoff processes in the data-scarce basin. In this embodiment, the reconstructed long series of daily runoff is approximated to the measured runoff series.

[0051] Based on the long series of daily runoff data reconstructed in step 2, the monthly runoff is calculated, and the monthly precipitation is calculated based on the daily precipitation data of ERA5. The Thiessen polygon method is used to obtain the spatially averaged monthly precipitation and NDVI data, and a vegetation-meteorological-hydrological drought regression model is constructed: (3); in, For the m Monthly relative NDVI anomaly for the entire study period, which is the vegetation data of the reanalysis dataset obtained in step 1; For the m reconstructed monthly runoff for the month; For the m Monthly precipitation totals; are model parameters, and the least square method is used to solve these two parameters.

[0052] Step 4: Based on the model parameters of the vegetation-meteorological-hydrological drought regression model, the integrated meteorological-hydrological drought index is calculated, and then drought events are extracted. K artificial intelligence models are constructed to simulate drought events, including the following steps: S4.1. Based on the long series of monthly runoff data reconstructed in step 3 and the basin-averaged monthly precipitation data, use the parameters of the obtained vegetation-meteorological-hydrological drought regression model to obtain the meteorological-hydrological comprehensive moisture index: (4); Where: is the meteorological-hydrological comprehensive moisture index of the mth month, and α and β are the model parameters of the vegetation-meteorological-hydrological drought regression model.

[0053] S4.2. The drought index is calculated using the meteorological and hydrological comprehensive moisture index as follows: (5); Where: Indicates the j The drought index for the month is as follows: the larger the index, the more severe the drought. MHwater j Indicates the j Monthly MHwater index; Respectively m The mean and standard deviation of monthly MHwater, k =1,2…,12, according to jDetermine the calendar month in which the month falls.

[0054] S4.3. Based on the drought index derived in step 4.2, the run theory is used to extract drought events. Drought events are extracted when the set threshold is exceeded, and relative humidity and specific humidity are calculated.

[0055] Run-length theory is a common technique in this field. Drought events were extracted using a threshold exceeding 0.5, and months below this threshold were defined as drought months.

[0056] The calculation process of relative humidity and specific humidity is: The Lausius-Clapeyron thermodynamic equation is defined as follows: (6); Where: is the first integration constant, which is 273.16 K; is the second integral constant, which is 611 Pa; is the latent heat of vaporization constant, take ; is the water vapor gas constant, take ; T is the air temperature, which is the input variable of the Clausius-Clapeyron thermodynamic equation; The 2-meter air temperature and dew point temperature of the meteorological and hydrological observation samples of each grid point in the study basin are substituted as input variables into the Clausius-Clapeyron thermodynamic equation to calculate the relative humidity of each grid point. RH , , where T dew is the dew point temperature, T 2m The air temperature at 2m; Specific humidity q is the ratio of water vapor mass to total air mass, using the surface pressure in ERA5 p And dew point temperature, the formula is as follows: (7); S4.4. Use the Thiessen polygon method to derive the average meteorological variables of the basin with scarce data, and combine the drought events in the basin with the meteorological variables of the basin. The meteorological variables include the average monthly temperature, specific humidity, relative humidity, snowfall, precipitation, wind speed, shortwave radiation and longwave radiation in the month when the drought event occurred and the month before. Use the random forest model to screen the key meteorological factors.

[0057] S4.5. Based on the reanalysis dataset, K artificial intelligence models are constructed using the key meteorological factors selected in step S4.4 as input and the drought index during the drought event as the target variable.

[0058] The K artificial intelligence models include: linear regression, logistic regression, decision tree, random forest, support vector machine, K nearest neighbor algorithm, naive Bayes classifier, neural network, convolutional neural network, recurrent neural network, long short-term memory network, generative adversarial network, Transformer, XGBoost and Markov decision process model.

[0059] That is, in this embodiment, 15 artificial intelligence models are constructed, and the 15 calibrated artificial intelligence models are obtained by optimizing and training using the gradient descent method.

[0060] Step 5: Construct an AI-Bayesian coupled model: Input the meteorological and hydrological observation data of the reanalysis dataset into K AI-based models for simulation, and use the Bayesian model averaging method to derive the weights of each AI model, including the following steps: S5.1. Use ERA5 meteorological data as input to drive the 15 artificial intelligence models constructed in step 4 to obtain a series of simulated magnitudes of drought events in the data-scarce basin.

[0061] S5.2. Based on the simulation magnitude results of 15 artificial intelligence models and the measured drought index, a probability density function of the simulated drought index was constructed according to the Bayesian total probability formula. Let S be the simulated drought index, R = [D, O] represents the model input data, where D is the simulation series of each drought simulation model during the training period, and O is the measured series. The output results of K different artificial intelligence models are obtained by the Bayesian total probability formula. The probability density function of S is as follows: (8); Where: For the k AI models The probability density function of the simulated value S under the given data R; For a given training data R k The posterior probability density function of the artificial intelligence models is calculated, and K represents the number of artificial intelligence models, which is 15 in this embodiment.

[0062] S5.3. Determine the corresponding weights based on the relative contributions of each AI model to the drought index simulation effect, thereby establishing a Bayesian mode average correction model. The specific operations are as follows: First, the drought index observation series and the simulation series obtained by each artificial intelligence model are transformed into normality using the Box-Cox function. Then, the estimation results of multiple models are weighted averaged based on the normal linear distribution assumption: (9); Where: Indicates the mean , the variance is Normal distribution; E represents the expected value of the function, is the weight of the k-th artificial intelligence model.

[0063] The long series of meteorological data sets were input into the 15 artificial intelligence models constructed in step 4 respectively, and then the simulation results of the 15 artificial intelligence models were input into the Bayesian model averaging model, and high-precision drought index simulation results were obtained through the weighted averaging method of optimal weights.

[0064] In this embodiment, the coupling method of artificial intelligence and Bayesian model averaging model is called building an artificial intelligence-Bayesian coupling model, and the input of the coupling model is meteorological variables.

[0065] Step 6: Construct a joint distribution function of drought index observation and forecast values based on the Copula function, and use the artificial intelligence-Bayesian coupling model to carry out drought forecasting.

[0066] The following steps are involved: S6.1. Obtain meteorological forecast information for the basin with scarce data from relevant meteorological forecasting agencies or departments, including the average monthly temperature, specific humidity, relative humidity, snowfall, precipitation, wind speed, shortwave radiation, and longwave radiation for the next month.

[0067] S6.2. Use the Thiessen polygon method to calculate the regional average weather forecast information for the watershed where data are scarce; S6.3. Reanalysis meteorological data for the past month are obtained from the reanalysis dataset, and the Thiessen polygon method is used to obtain the regional average reanalysis meteorological data for the data-scarce basin.

[0068] S6.4. Use the regional average meteorological forecast information for the next month of the scarce data basin obtained in step 6.2 and the regional average reanalysis meteorological data for the past month of the scarce data basin obtained in step 6.3 as input to drive the artificial intelligence-Bayesian coupling model established in step 5 to obtain the drought index forecast value for the next month.

[0069] S6.5. Use the retrospective forecast method to construct the joint distribution function of the historical drought index observation and forecast values, and let H k 、 S k Respectively represent k The real drought index and the deterministic drought index at each moment, h k 、 s k Then H k 、 Sk The realized value of and They are H k 、 S k The marginal distribution function of and The corresponding H k 、 S k The probability density function of , according to Sklar's theorem, there exists a n -Copula function C , so that the following equation holds: (10); Where: is the Copula function parameter, often using the Kendall rank correlation coefficient Solution; is the density function of the two-dimensional Copula function; Represents the joint distribution function of the actual drought index and the predicted drought index.

[0070] In this embodiment, Gumbel-Hougaard, Clayton and Frank Copula functions from the common Archimedean Copula function family are selected to construct H k 、 S k The joint distribution of , and its related mathematical expressions are detailed in Table 1.

[0071]

[0072] S6.7, based on n -Copula function is used to deduce the posterior distribution function of the drought index and conduct probabilistic forecasting.

[0073] Given forecast value s k ,exist S k = s k hour, H k The conditional distribution function of for: (11); H k The density function is expressed as: (12); and When the deterministic forecast drought index information is given, the measured drought index h k The posterior probability distribution function and posterior probability density function are used to describe the uncertainty of the forecast; the expected value of the posterior probability density function is taken as the post-processing deterministic forecast result, and the confidence level is obtained at the same time The expected value of the posterior probability density function of the drought index under the predicted interval of the drought index h ke Solve the following equation: (13); when When H k The confidence level is 90%. H k The lower and upper confidence limits are taken as the 5% and 95% quantiles, respectively, to quantitatively describe the uncertainty of drought forecasts, and the information within the interval is used as the probabilistic forecast result.

[0074] S6.8. Based on the drought index probability forecast results of step S6.7, carry out drought warning.

[0075] When the expected value of the drought index exceeds 0.5, a drought warning information will be issued; when the expected value of the drought index is lower than 0.5, but the 95% quantile exceeds 1.0, a warning information will also be issued; in other cases, no warning information will be issued, but a probability forecast interval value of the drought index will be given for reference in disaster prevention, mitigation and engineering decision-making.

[0076] Example 2: The present invention also provides a drought early warning device based on artificial intelligence, which is used to execute the drought early warning method described in Example 1. Figure 4 As shown, the drought early warning device 200 includes: The first determination module 201 is used to collect data on drought warning basins; Analysis module 202, used to establish a watershed hydrological model and use a long short-term memory model for correction, thereby constructing a deep learning model that considers the hydrological process; Construction module 203, for constructing a monthly scale vegetation-meteorological-hydrological drought regression model; The second determination module 204 is used to construct an artificial intelligence-Bayesian coupling model; The prediction module 205 constructs a joint distribution function of drought index observation and forecast values based on the Copula function, uses an artificial intelligence-Bayesian coupling model to carry out drought forecasting, and uses a Bayesian probability forecast processor based on the Copula function to quantify forecast uncertainty.

[0077] Example 3: The present invention also provides an execution device, see Figure 5 , Figure 5 This is a schematic diagram of the structure of the execution device provided in the embodiment of the present application. The execution device 300 can be specifically manifested as an autonomous driving vehicle, a mobile phone, a tablet, a laptop computer, a desktop computer, a monitoring data processing device, etc., which is not limited here. Figure 1 The execution device 300 includes a receiver 301, a transmitter 302, a processor 303, and a memory 304 (the number of the processor 303 in the execution device 300 can be one or more, Figure 3 (taking one processor as an example), the processor 303 may include an application processor 3031 and a communication processor 3032. In some embodiments of the present application, the receiver 301, the transmitter 302, the processor 303 and the memory 304 may be connected via a bus or other means.

[0078] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 303. A portion of the memory 304 may also include non-volatile random access memory (NVRAM). The memory 304 stores processor and operation instructions, executable modules, or data structures, or subsets or extended sets thereof. The operation instructions may include various operation instructions for implementing various operations.

[0079] Processor 303 controls the operation of the execution device. In specific applications, the various components of the execution device are coupled together via a bus system. In addition to a data bus, the bus system may also include a power bus, a control bus, and a status signal bus. However, for clarity, all of these buses are referred to as a bus system in the figure.

[0080] The methods disclosed in the above embodiments of the present application can be applied to or implemented by processor 303. Processor 303 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be performed by hardware integrated logic circuits or software instructions in processor 303. The above processor 303 can be a general-purpose processor, a digital signal processor, a microprocessor, or a microcontroller. It can also further include an application-specific integrated circuit, a field-programmable gate array, or other programmable logic device, discrete gate or transistor logic devices, or discrete hardware components. The processor 303 can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software modules can be located in storage media well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 304 , and the processor 303 reads the information in the memory 304 and completes the steps of the above method in combination with its hardware.

[0081] Receiver 301 can be used to receive input data or rare character information and generate signal input related to executing device-related settings and function control. Transmitter 302 can be used to output data or rare character information via a first interface. Transmitter 302 can also be used to send instructions to the disk pack via the first interface to modify data in the disk pack. Transmitter 302 can also include a display device such as a display screen.

[0082] In the embodiment of the present application, the processor 303 is used to execute Figure 1 The specific manner in which the application processor 3031 in the processor 303 performs the above steps is the same as that in the present application. Figure 1 The corresponding method embodiments are based on the same concept, and the technical effects they bring are the same as those in this application. Figure 1 The corresponding method embodiments are the same. For specific contents, please refer to the description in the method embodiments shown above in this application, which will not be repeated here.

[0083] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A drought early warning method based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Collect data for the drought warning basin: Collect ground observation data, including observation data from ground meteorological stations in the basin and daily runoff data from hydrological stations at the outlet section, collect meteorological and hydrological observation data and vegetation data from the reanalysis dataset, collect satellite-derived precipitation datasets, construct a statistical downscaling model, correct for precipitation magnitude and temperature simulation bias, and obtain a long series of downscaled meteorological observation data; Step 2: Based on the short-series runoff observation data of the watershed with scarce data and the downscaled long-series meteorological observation data, a deep learning model considering the hydrological process is constructed to simulate daily runoff and reconstruct the long-series daily runoff data; Step 3: Construct a monthly vegetation-meteorological-hydrological drought regression model and obtain model parameters; Step 4: Based on the model parameters of the vegetation-meteorological-hydrological drought regression model, the integrated meteorological-hydrological drought index is calculated, and then drought events are extracted and K artificial intelligence models are constructed to simulate drought events. Step 5: Construct an AI-Bayesian coupled model: Input the meteorological and hydrological observation data of the reanalysis dataset into K AI-based models for simulation, and use the Bayesian model averaging method to derive the weights of each AI model; Step 6: Construct a joint distribution function of drought index observation and forecast values based on the Copula function, and use the artificial intelligence-Bayesian coupling model to carry out drought forecasting.

2. The artificial intelligence-based drought early warning method according to claim 1, characterized in that: In step 1, the equal rate correction method is used to construct a statistical downscaling model, and the operation steps are as follows: Assuming that the monthly biases between satellite-derived precipitation data or temperature data from reanalysis datasets and ground-based observations are consistent over time, we first calculate the correction factor for each month based on the ground-based observations and then apply this correction factor to the long series of simulated datasets for the same month. The correction factors include the precipitation deviation rate and the absolute temperature deviation. After obtaining the corresponding correction factors, the temperature series of the satellite-retrieved precipitation data and the reanalysis dataset are corrected using the following formula: （1）； Where: N Indicates site m Total number of observation days per month; i Represents a daily precipitation or temperature series; Indicates that the satellite-retrieved precipitation m The original data series for the month, Indicates the corrected satellite-retrieved precipitation in m Monthly series; Indicates that the satellite-retrieved precipitation m Month i The original data of the day, Indicates that the weather station is m Month i Daily observed precipitation data; Reanalysis temperature in the m The original data series for the month, The corrected reanalysis temperature product is m Monthly series; Reanalysis temperature product in the m Month i The original data of the day, Indicates that the weather station is m Month i Observed temperature data for the day.

3. The drought early warning method based on artificial intelligence according to claim 1, characterized in that: In step 2, daily runoff simulation using a deep learning model that considers hydrological processes includes the following steps: S2.

1. Construct and calibrate the HBV model based on short-series runoff observations and downscaled long-series meteorological observations in data-scarce basins. S2.

2. The first simulated daily runoff process is simulated based on the calibrated HBV model. The lag time that affects the measured daily runoff is determined by statistically analyzing the first simulated daily runoff process and the measured daily runoff process. N ; S2.

3. Use the corrected long series of satellite-retrieved precipitation data and temperature data from the reanalysis dataset to drive the calibrated HBV model to obtain a long series of daily runoff simulations. The simulations are then divided into three types of daily runoff series with different magnitudes. S2.

4. Based on the three types of daily runoff series with different magnitudes, a long-short-term memory neural network model is constructed respectively. The long series of daily runoff simulation is corrected to obtain the second simulated runoff series, which is expressed as: ] （2）； Where, express t The second simulated runoff after time correction, Indicates that the HBV model is t Daily runoff simulation series at each moment, Indicates that the HBV model t- 1 moment daily runoff simulation series, N Indicates the lag time that affects the daily measured runoff; F LSTM Represents an LSTM model.

4. The artificial intelligence-based drought early warning method according to claim 3, characterized in that: In step S2.3, the method for dividing the simulation series into three types of daily runoff series of different magnitudes is: calculating the 30% quantile and 60% quantile of the daily runoff series, and dividing the daily runoff series into a first type below the 30% quantile, a second type between the 30% quantile and the 60% quantile, and a third type above the 60% quantile.

5. The drought early warning method based on artificial intelligence according to claim 1, characterized in that: In step 3, the monthly runoff is calculated based on the long series of daily runoff data reconstructed in step 2, and the monthly precipitation is calculated based on the precipitation data of the reanalysis dataset, and a vegetation-meteorological-hydrological drought regression model is further constructed: （3）； in, For the m Monthly relative NDVI anomaly for the entire study period, which is the vegetation data of the reanalysis dataset obtained in step 1; For the m reconstructed monthly runoff for the month; For the m Monthly precipitation totals; and are model parameters, and the least square method is used to solve these two parameters.

6. The artificial intelligence-based drought early warning method according to claim 5, characterized in that: The step 4 comprises the following steps: S4.

1. Based on the long series of monthly runoff data reconstructed in step 3 and the basin-averaged monthly precipitation data, use the parameters of the obtained vegetation-meteorological-hydrological drought regression model to obtain the meteorological-hydrological comprehensive moisture index: （4）； Where: For the m Monthly meteorological and hydrological comprehensive moisture index, and are the model parameters of the vegetation-meteorological-hydrological drought regression model; S4.

2. The drought index is calculated using the meteorological and hydrological comprehensive moisture index as follows: （5）； Where: Indicates the j The drought index for the month is as follows: the larger the index, the more severe the drought; MHwater j Indicates the j Monthly MHwater index; Respectively m The mean and standard deviation of monthly MHwater, k =1,2…,12, according to j Determine the calendar month in which the month falls; S4.

3. Based on the drought index derived in step 4.2, the run theory is used to extract drought events. Drought events are extracted when the set threshold is exceeded, and relative humidity and specific humidity are calculated. The calculation process is as follows: The Lausius-Clapeyron thermodynamic equation is defined as follows: (6)； Where: is the first integration constant, which is 273.16 K; is the second integral constant, which is 611 Pa; is the latent heat of vaporization constant, take ; is the water vapor gas constant, take ; T is the air temperature, which is the input variable of the Clausius-Clapeyron thermodynamic equation; The 2-meter air temperature and dew point temperature of the meteorological and hydrological observation samples of each grid point in the study basin are substituted as input variables into the Clausius-Clapeyron thermodynamic equation to calculate the relative humidity of each grid point. RH , , where T dew is the dew point temperature, T 2m The air temperature at 2m; Specific humidity q is the ratio of water vapor mass to total air mass mass, using the ground pressure in the reanalysis dataset. p And dew point temperature, the formula is as follows: (7)； S4.

4. Use the Thiessen polygon method to derive average meteorological variables for watersheds with scarce data. Combine the drought events in the watershed with the meteorological variables for the watershed. The meteorological variables include the average monthly temperature, specific humidity, relative humidity, snowfall, precipitation, wind speed, shortwave radiation, and longwave radiation for the month of the drought event and the month before. Use a random forest model to screen key meteorological factors. S4.

5. Based on the reanalysis dataset, K artificial intelligence models are constructed using the key meteorological factors selected in step S4.4 as input and the drought index during the drought event as the target variable.

7. The artificial intelligence-based drought early warning method according to claim 6, characterized in that: In step S4.5, the K artificial intelligence models include: linear regression, logistic regression, decision tree, random forest, support vector machine, K nearest neighbor algorithm, naive Bayes classifier, neural network, convolutional neural network, recurrent neural network, long short-term memory network, generative adversarial network, Transformer, XGBoost and Markov decision process model.

8. The artificial intelligence-based drought early warning method according to claim 1, characterized in that: The step 5 comprises the following steps: S5.

1. Use the meteorological data from the reanalysis dataset as input to drive the K artificial intelligence models constructed in step 4 to obtain a series of simulated magnitudes of drought events in the data-scarce basin. S5.

2. Based on the simulation magnitude results of K artificial intelligence models and the measured drought index, the probability density function of the simulated drought index is constructed according to the Bayesian total probability formula. Let S be the simulated drought index, R = [D, O] represents the model input data, where D is the simulation series of each drought simulation model during the training period, and O is the measured series. The output results of K different artificial intelligence models are obtained by the Bayesian total probability formula. The probability density function of S is as follows: （8）； Where: For the k AI models The probability density function of the simulated value S under the given data R; For a given training data R k The posterior probability density function of the artificial intelligence model, K represents the number of artificial intelligence models; S5.

3. Determine the corresponding weights based on the relative contributions of each AI model to the drought index simulation effect, thereby establishing a Bayesian mode average correction model. The specific operations are as follows: First, the drought index observation series and the simulation series obtained by each artificial intelligence model are transformed into normality using the Box-Cox function. Then, the estimation results of multiple models are weighted averaged based on the normal linear distribution assumption: （9）； Where: Indicates the mean , the variance is Normal distribution; E represents the expected value of the function, For the k The weight of an AI model.

9. The artificial intelligence-based drought early warning method according to claim 1, characterized in that: The step 6 includes the following steps: S6.

1. Obtain weather forecast information for the basin where data are scarce from relevant weather forecasting agencies or departments, including monthly average temperature, specific humidity, relative humidity, snowfall, precipitation, wind speed, shortwave radiation, and longwave radiation for the next month; S6.

2. Use the Thiessen polygon method to calculate the regional average weather forecast information for the watershed where data are scarce; S6.

3. Reanalysis meteorological data for the past month are obtained from the reanalysis dataset. The Thiessen polygon method is used to obtain the regional average reanalysis meteorological data for the data-scarce basin. S6.

4. Use the regional average meteorological forecast information for the next month for the data-scarce watershed obtained in step 6.2 and the regional average reanalysis meteorological data for the recent month for the data-scarce watershed obtained in step 6.3 as input to drive the artificial intelligence-Bayesian coupling model established in step 5 to obtain a drought index forecast for the next month. S6.

5. Use the retrospective forecast method to construct the joint distribution function of the historical drought index observation and forecast values, let H k 、 S k Respectively represent k The real drought index and the deterministic drought index at each moment, h k 、 s k Then H k 、 S k The realized value of and They are H k 、 S k The marginal distribution function of and The corresponding H k 、 S k The probability density function of , according to Sklar's theorem, there exists a n -Copula function C , so that the following formula holds: (10)； Where: Copula function parameters; is the density function of the two-dimensional Copula function; represents the joint distribution function of the real drought index and the predicted drought index; S6.7, based on n - Use the Copula function to derive the posterior distribution function of the drought index and conduct probabilistic forecasting; Given forecast value s k ,exist S k = s k hour, H k The conditional distribution function of for: (11)； H k The density function is expressed as: (12)； and When the deterministic forecast drought index information is given, the measured drought index h k The posterior probability distribution function and posterior probability density function are used to describe the uncertainty of the forecast; the expected value of the posterior probability density function is taken as the post-processing deterministic forecast result, and the confidence level is obtained at the same time The expected value of the posterior probability density function of the drought index under the predicted interval of the drought index h ke Solve the following equation: (13)； S6.

8. Based on the drought index probability forecast results of step S6.7, carry out drought warning.

10. A drought early warning device based on artificial intelligence, characterized in that: The method for executing the drought early warning method according to any one of claims 1 to 9 comprises: The first determination module is used to collect data on drought warning basins; An analysis module is used to establish a watershed hydrological model and use a long-short-term memory model for correction, thereby constructing a deep learning model that considers hydrological processes; Build a module to construct a monthly-scale vegetation-meteorological-hydrological drought regression model; The second determination module is used to build an artificial intelligence-Bayesian coupling model; The prediction module constructs the joint distribution function of drought index observation and forecast values based on the Copula function, and uses the artificial intelligence-Bayesian coupling model to carry out drought forecasting.

Images