A flood and drought warning method and system for coupling satellite and ground base station signals

Abstract

The application provides a drought-flood sudden change early warning method and system for coupling satellite and ground base station signals, and comprises the following steps: acquiring multi-source data of a research area; loading a hydrological climate reconstruction model to reconstruct long-series gridded rainfall data and soil moisture data of the research area in a historical period; optimizing a plurality of rain attenuation rain intensity models for rainfall inversion by a microwave link signal; constructing a correction module of the rain attenuation rain intensity model to correct real-time regional rainfall series inverted by the microwave link; constructing a drought-flood sudden change event intensity correction module to generate a calibrated historical drought-flood sudden change event library; constructing a drought-flood sudden change dynamic early warning module fusing a circulation signal to output early warning results of drought-flood sudden change in a future period; and generating and releasing drought-flood sudden change risk early warning information based on the early warning results output by the drought-flood sudden change dynamic early warning module. The application provides a real-time, dynamic and high-precision drought-flood sudden change early warning mode by fusing satellite remote sensing and mobile communication base station signals.

Description

Technical Field

[0001] This invention belongs to the field of disaster early warning technology, and in particular relates to a method and system for early warning of sudden shifts between drought and flood by coupling satellite and ground base station signals. Background Technology

[0002] A rapid shift from drought to flood is an extreme climate event, referring to the phenomenon where a region drastically changes from a drought state to a flood state, or vice versa, within a short period. These events are characterized by their suddenness, destructive power, and difficulty in forecasting and early warning. For example, after a period of drought, soil compaction and reduced infiltration capacity make the area highly susceptible to secondary disasters such as flash floods, landslides, and mudslides if subjected to sudden heavy rainfall. The damage from these events often far exceeds that of a prolonged drought or a single flood event. Therefore, developing precise and efficient early warning technologies for rapid shifts from drought to flood is of great significance for improving disaster prevention and mitigation capabilities and ensuring the stable operation of the social economy.

[0003] Currently, the monitoring and early warning of rapid shifts between drought and flood mainly rely on traditional meteorological station networks, satellite remote sensing, and numerical weather prediction. However, each of these methods has significant limitations. The first category is early warning methods based on ground-based station networks. Although meteorological stations and rain gauges can provide accurate single-point data, their spatial distribution is uneven, and their density is insufficient in remote mountainous and rural areas, making it difficult to capture the fine spatial structure of localized heavy rainfall and soil moisture changes. Rapid shifts between drought and flood often occur in these areas with weak monitoring. The second category is early warning methods based on satellite remote sensing. Satellite remote sensing (such as TRMM and GPM) has the advantage of wide spatial coverage, but its temporal resolution is low (from several hours to several days), making it difficult to meet the real-time monitoring needs of minute- to hourly "rapid shift" processes. At the same time, the spatial resolution and accuracy of rainfall and soil moisture products retrieved by satellite still need improvement, especially under the influence of clouds and rain, where signal attenuation and errors increase. The third category is early warning methods based on numerical weather prediction. Numerical weather prediction models can provide future trends, but they have significant uncertainties in predicting extreme weather processes at small and medium scales. Furthermore, the accuracy of the model's initial field heavily depends on sparse observation data, resulting in limited forecasting capabilities for rapidly evolving events such as sudden shifts between drought and flood.

[0004] In recent years, some scholars have proposed using commercial mobile communication networks for rainfall monitoring. Microwave signals are affected by rainfall attenuation during propagation; by analyzing the degree of signal attenuation along the link, the rainfall intensity along the path can be inferred. For example, patent CN202511648327.4 discloses a rainfall observation method that integrates satellite remote sensing and mobile communication base station signals. This method effectively utilizes the wide-area coverage of satellites and the high density of base stations, improving the accuracy of rainfall observation. However, directly applying this technology to early warning of abrupt shifts between drought and flood still faces many challenges. First, existing research mostly focuses on the observation of rainfall itself, failing to effectively couple rainfall information with key indicators characterizing drought (such as soil moisture), and thus failing to systematically depict the complete process chain from "drought" to "flood." Second, there is a lack of quantification and identification mechanisms for the characteristics of "abrupt shifts." Existing technologies mostly provide state data such as rainfall or soil moisture, but fail to automatically and accurately extract the spatiotemporal patterns and early signals of abrupt shifts between drought and flood states from these massive amounts of data using intelligent algorithms. Finally, the adaptability and robustness of early warning models are insufficient. The trigger threshold for rapid shifts between drought and flood varies depending on the region, season, and underlying surface conditions, while existing methods mostly use fixed thresholds, making them difficult to apply universally. Furthermore, they fail to fully utilize artificial intelligence technologies to deeply mine the complex nonlinear relationships between satellite data, large-scale circulation anomalies, reanalysis data, Earth system models, and ground-based base station signals, in order to achieve self-calibration and dynamic optimization of early warning models. Summary of the Invention

[0005] This invention provides a novel early warning method that can integrate multi-source data and be optimized for the characteristics of rapid drought-flood transition events, in order to overcome the shortcomings of existing technologies and achieve high timeliness and high accuracy in early warning of rapid drought-flood transition events.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A method for early warning of sudden shifts between drought and flood by coupling satellite and ground base station signals includes: acquiring multi-source data of the study area and performing fusion preprocessing;

[0008] Load the hydro-climate reconstruction model, and reconstruct long-series gridded rainfall data and soil moisture data of the study area for historical periods based on the multi-source data;

[0009] Based on reconstructed long-series gridded rainfall data, various rainfall attenuation and intensity models retrieved from microwave link signals are optimized. Based on the optimized rainfall attenuation and intensity models, a correction module for these models is constructed using rainfall data retrieved from meteorological satellites to correct the real-time regional rainfall sequence retrieved from microwave links. Furthermore, a real-time drought-flood abrupt change index is calculated based on the reconstructed long-series gridded rainfall data and soil moisture data, and a drought-flood abrupt change event intensity correction module is constructed to generate a calibrated historical drought-flood abrupt change event database. Finally, a dynamic early warning module for drought-flood abrupt changes, incorporating circulation signals, is constructed. Using the corrected real-time regional rainfall sequence, real-time drought-flood abrupt change index, real-time microwave signal characteristics, and large-scale circulation anomaly signals as input, the module outputs early warning results for future drought-flood abrupt changes.

[0010] Based on the early warning results output by the dynamic early warning module for rapid shifts between drought and flood, early warning information on the risk of rapid shifts between drought and flood, including risk level, intensity of the rapid shift, scope of impact, and warning time, is generated and released.

[0011] Furthermore, the multi-source data includes real-time monitoring data, historical and background field data, and large-scale climate driving signals; the real-time monitoring data includes rainfall data retrieved from meteorological satellites and real-time microwave signal correlation data from mobile communication base stations in the study area; the historical and background field data includes meteorological and hydrological data provided by reanalysis datasets and meteorological and hydrological forecast products output by Earth system models; the large-scale climate driving signals include sea surface temperature anomaly data and large-scale circulation anomaly indices in key sea areas.

[0012] Furthermore, the fusion preprocessing includes a spatiotemporally aligned dataset formed by performing spatiotemporal matching, standardization, and quality control on multi-source data.

[0013] Furthermore, the reconstructed long-series gridded rainfall and soil moisture data for the historical period of the study area includes:

[0014] Using the reanalysis dataset, meteorological and hydrological data output from the Earth system model, and large-scale circulation index as inputs, the first deep learning model is used to reconstruct long-series gridded rainfall data of the historical period in the study area;

[0015] Driven by meteorological and hydrological data and large-scale circulation indices provided by long-series reanalysis datasets, and combined with short-series soil moisture reference fields generated by site observation interpolation and their physical constraints, a second deep learning model was used to reconstruct long-series gridded soil moisture data for historical periods in the study area.

[0016] Furthermore, the optimization of various rainfall attenuation and intensity models derived from microwave link signals includes:

[0017] For each microwave link within the study area, the relationship between rainfall attenuation and rainfall intensity is established using multiple candidate rainfall attenuation and rainfall intensity models, and the rainfall intensity of the corresponding microwave link is estimated. Based on the rainfall intensity of the microwave link, the regional rainfall for the corresponding historical period is calculated for each candidate rainfall attenuation and rainfall intensity model.

[0018] Using long-series gridded rainfall data from historical periods in the reconstructed study area as the observed true values, and the regional rainfall corresponding to each candidate rainfall attenuation and intensity model as the simulated values, the Nash efficiency coefficient, root mean square error, and / or Pearson correlation coefficient were used as evaluation indicators to select the best rainfall attenuation and intensity model.

[0019] Furthermore, the rain attenuation and rainfall intensity correction module includes:

[0020] The regional average rainfall for historical periods and the microwave signal at the corresponding time, retrieved based on the optimized rainfall attenuation and rainfall intensity model, were used as explanatory variables.

[0021] Using the regional average hourly rainfall sequence retrieved from meteorological satellites as the target variable, a convolutional neural network model is constructed and trained to correct the historical regional average rainfall sequence retrieved from the optimized rainfall attenuation and intensity model.

[0022] Furthermore, the module for correcting the intensity of sudden shifts between drought and flood events includes:

[0023] Based on the reconstruction of long-series gridded rainfall and soil moisture data of the historical period in the study area, the drought-flood transition index of the historical period is defined and calculated, and historical drought-flood transition events are extracted based on preset thresholds.

[0024] Using the corrected regional average rainfall series from historical periods, the intensity of the calculated drought-flood transition events is recalibrated, an intensity correction relationship is established, and a calibrated database of historical drought-flood transition events is generated.

[0025] Furthermore, the dynamic early warning module for sudden shifts between drought and flood adopts an encoder-decoder network architecture that incorporates an attention mechanism; the encoder is used to encode the multidimensional feature sequence of the input, including corrected real-time rainfall, satellite-retrieved rainfall, real-time drought-flood shift index, microwave signal features, and large-scale circulation anomaly signals; the decoder is used to predict the probability and intensity index of sudden shifts between drought and flood in future periods based on the encoding results.

[0026] Furthermore, the specific rules for generating the aforementioned drought-flood transition risk warning information include:

[0027] Risk levels are dynamically classified based on the predicted probability of a sudden shift from drought to flood.

[0028] Map the predicted drought-flood transition intensity index to transition intensity;

[0029] Spatial analysis is used to identify and delineate continuous areas that reach a predetermined risk level, as the potential impact range;

[0030] Clearly indicate the effective period of the warning results.

[0031] On the other hand, the present invention provides a drought and flood emergency warning system that couples satellite and ground base station signals, comprising:

[0032] Multi-source data acquisition module: It is used to acquire multi-source data of the study area and perform fusion preprocessing;

[0033] Data reconstruction module: It is used to load the hydro-climate reconstruction model and reconstruct long-series gridded rainfall data and soil moisture data of the study area based on the multi-source data.

[0034] The rain attenuation and intensity model optimization module is used to optimize various rain attenuation and intensity models retrieved from microwave link signals based on reconstructed long-series gridded rainfall data. The calibration module is used to construct a calibration module for the rain attenuation and intensity models based on the optimized models and rainfall data retrieved from meteorological satellites, calibrating the real-time regional rainfall sequence retrieved from microwave links. It also calculates the real-time drought-flood abrupt change index based on the reconstructed long-series gridded rainfall data and soil moisture data, constructs a drought-flood abrupt change event intensity calibration module, and generates a calibrated historical drought-flood abrupt change event database. The drought-flood abrupt change dynamic early warning module is used to construct a drought-flood abrupt change dynamic early warning module that integrates circulation signals. It takes the calibrated real-time regional rainfall sequence, real-time drought-flood abrupt change index, real-time microwave signal characteristics, and large-scale circulation anomaly signals as inputs, and outputs early warning results for future drought-flood abrupt changes.

[0035] The results output module is used to generate and publish drought and flood rapid change risk warning information based on the warning results output by the drought and flood rapid change dynamic warning module. This information includes the risk level, rapid change intensity, impact range and warning time.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. This invention is scientifically sound and closely aligned with engineering practice. It fully leverages the advantages of artificial intelligence, satellite remote sensing rainfall observation technology, reanalysis datasets, and microwave link real-time rainfall intensity observation technology, overcoming the shortcomings of low spatial resolution in satellite observations and the difficulty in verifying microwave link observation data. By integrating satellite remote sensing and mobile communication base station signals, it provides a real-time, dynamic, and high-precision early warning mode for rapid shifts between drought and flood.

[0038] 2. This invention integrates satellite remote sensing and artificial intelligence technologies to propose a high-reliability early warning method for rapid shifts between drought and flood, which can provide important and highly operable reference for flood control, disaster reduction, and watershed planning. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a method for early warning of sudden shifts in drought and flood conditions by coupling satellite and ground base station signals according to an embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram of the spatial distribution of mobile communication base stations in one embodiment of the present invention.

[0042] Figure 3 This is a schematic diagram of the observation area divided into small grids and microwave links in one embodiment of the present invention.

[0043] Figure 4 This is a schematic diagram of a microwave link within a small grid in another embodiment of the present invention.

[0044] Figure 5 This is a schematic diagram of a small grid that does not contain microwave links in another embodiment of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] In specific implementation, the method proposed in the technical solution of this invention can be automatically executed by those skilled in the art using computer software technology. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of this invention and computer equipment including the computer program running the corresponding computer program, should also be within the protection scope of this invention.

[0047] Example 1

[0048] This invention collects real-time monitoring data, rainfall data retrieved from meteorological satellites, real-time microwave signal data from mobile communication base stations in the study area, historical and background field data, and large-scale climate-driven signals to form a unified multi-source data cube. Then, it reconstructs a long-term, high-resolution gridded rainfall and soil moisture data of the study area to accurately characterize the region's historical drought and flood background. Different rainfall attenuation and intensity models are used to establish the relationship between rainfall attenuation and intensity, and the corresponding link rainfall intensity is derived. A rainfall attenuation and intensity model suitable for the study area is selected. Based on the selected model, a correction module for the model is constructed, taking into account rainfall data retrieved from meteorological satellites. A correction module for the intensity of drought-flood transition events is constructed by extracting drought-flood transition events from the reconstructed dataset. Finally, a dynamic early warning module for drought-flood transitions, incorporating circulation signals, is constructed to generate and release real-time drought-flood transition risk warnings. This provides a method and device for early warning of drought-flood transitions that couples satellite and ground base station signals. The specific process is detailed in [link to detailed process]. Figure 1 ,include:

[0049] Step 1: Acquire multi-source data for the study area and perform fusion preprocessing;

[0050] Step 2: Load the hydro-climate reconstruction model and reconstruct long-series gridded rainfall data and soil moisture data of the study area based on the multi-source data;

[0051] Step 3: Based on the reconstructed long series of gridded rainfall data, the optimal rain attenuation and rainfall intensity models for retrieving rainfall from microwave link signals are selected;

[0052] Step 4: Based on the selected rainfall attenuation and intensity model, a correction module for the rainfall attenuation and intensity model is constructed by combining the rainfall data retrieved from meteorological satellites to correct the real-time regional rainfall sequence retrieved from the microwave link; and based on the reconstructed long-series gridded rainfall data and soil moisture data, the real-time drought-flood transition index is calculated, and a drought-flood transition event intensity correction module is constructed to generate a calibrated historical drought-flood transition event database.

[0053] Step 5: Construct a dynamic early warning module for rapid drought-flood transition by integrating circulation signals. The module takes the corrected real-time regional rainfall sequence, real-time drought-flood transition index, real-time microwave signal characteristics, and large-scale circulation anomaly signal as inputs and outputs early warning results for rapid drought-flood transition in the future period.

[0054] Step 6: Based on the early warning results output by the dynamic early warning module for rapid drought-flood transition, generate and release early warning information on the risk of rapid drought-flood transition, including risk level, intensity of rapid transition, scope of impact, and warning time.

[0055] In step 1 of this embodiment, multi-source data of the study area is acquired, including: 1) real-time monitoring data: rainfall data retrieved from meteorological satellites, real-time microwave signal correlation data from mobile communication base stations in the study area, and soil moisture data observed in the study area; 2) historical and background field data: meteorological and hydrological data provided by reanalysis datasets, and meteorological and hydrological forecast products output by Earth system models; 3) large-scale climate driving signals: sea surface temperature anomalies in key sea areas, and large-scale circulation anomaly indices such as ENSO and IOD. After obtaining the above three types of data, all data undergoes spatiotemporal matching, standardization, and quality control processing to form a unified multi-source data cube.

[0056] For each grid point in the study area, hourly meteorological and hydrological data for the period 1950–2025 were first collected from the European Centre for Medium-Range Weather Forecasts' Generation 5 Atmospheric Reanalysis Dataset (ERA5-Land). Specific variables included air temperature, air pressure, dew point temperature, snowfall, precipitation, surface runoff depth, soil moisture, evaporation, shortwave radiation, and longwave radiation. Simultaneously, precipitation data retrieved from the Global Precipitation Measurement Mission (GPM) satellite system were acquired, for example, using its GPM IMERG-Final version, which provides hourly precipitation datasets with a spatial resolution of 0.1 degrees for the study area from 2000–2025.

[0057] Furthermore, soil moisture observation data for the study area from 2000 to 2025 were obtained. This data consists of station observation data and is hourly soil moisture data.

[0058] Furthermore, historical and future scenario data from Earth system models will be obtained, for example, using the CAS-ESM2.0 model. Specifically, this includes simulation data for historical periods (1950-2014) and hourly-scale data for the period 2015-2025 under the SSP370 path, with variables covering temperature, specific humidity, relative humidity, snowfall, precipitation, surface runoff depth, soil moisture, evaporation, shortwave radiation, and longwave radiation.

[0059] Furthermore, complete microwave signal data of all mobile communication base stations in the study area since their establishment were collected. This data includes dynamic parameters with precise timestamps, such as Received Signal Level (RSL), Bit Error Rate (BER), and Signal-to-Noise Ratio (SNR), as well as fixed characteristic parameters of the links, such as length, frequency, polarization, location coordinates, and elevation angle.

[0060] Furthermore, in order to capture large-scale climate-driving signals, it is also necessary to collect key circulation anomaly indices from the same period, including: various ENSO indices (such as ONI and SOI), the Indian Ocean Dipole (IOD) index, and the Pacific Decadal Oscillation (PDO) index.

[0061] Finally, all acquired multi-source data are subjected to unified spatiotemporal matching, standardization, and quality control processing to form a spatiotemporally aligned multi-source data cube, laying the foundation for subsequent analysis.

[0062] Step 2 of this embodiment involves the reconstruction of the drought and flood background field based on deep learning. The "hydrological climate reconstruction model" constructed by the first deep learning model is loaded and run. This model uses the reanalysis data fused in Step 1, the Earth system model output, and the large-scale circulation index as inputs to reconstruct a long-term, high-resolution gridded rainfall and soil moisture data of the study area, thereby accurately depicting the region's historical drought and flood background.

[0063] The Clausius-Clapeyron thermodynamic equation is defined as follows:

[0064]

[0065] in, Let K be the first integration constant, taken as 273.16; The second integration constant is set to 611 Pa. Let be the latent heat of vaporization constant, and take . ; Let be the water vapor gas constant, and take . ; T For air temperature, it is an input variable to the Clausius-Clapeyron thermodynamic equation; exp represents the natural exponential function. Indicates temperature T The saturated water vapor pressure below.

[0066] By substituting the air temperature and dew point temperature from the ERA5-Land dataset for each grid point in the study area into the Clausius-Clapeyron thermodynamic equation as input variables, the relative humidity (RH) for each grid point was calculated. ,in T dew Dew point temperature, T 2m The temperature is 2m.

[0067] wet q The ratio of water vapor mass to total air mass mass is given by ERA5-Land pressure. p The formula for calculating the dew point temperature is as follows:

[0068]

[0069] The bilinear interpolation method was used to interpolate the meteorological and hydrological variables from the ERA5-Land dataset and the Earth System Model output to a spatial resolution of 0.1 degrees.

[0070] Based on this, a first deep learning model for reconstructing the hydroclimate field is constructed (in this embodiment, a temporal convolutional network TCN is used). The driving factors of this model are expanded into three categories: 1) variables obtained or derived from ERA5-Land, including air temperature, relative humidity, specific humidity, snowfall, precipitation, runoff depth, soil moisture, shortwave radiation, and longwave radiation; 2) corresponding variables output by Earth system models; and 3) large-scale circulation anomaly indices (including ONI, SOI, IOD, and PDO indices) that serve as key climate forcing signals.

[0071] Based on the driving factors and rainfall data retrieved from the GPM satellite system, an artificial neural network is constructed, considering a lag of several hours. Specifically, the GPM-retrieved rainfall at each grid point within the study area, the driving factors at the same time, and the driving factors at several lag times (e.g., 3 hours) are all used as calibration data for the artificial neural network to calibrate an artificial intelligence model (rainfall reconstruction model) covering the study area. When calibrating the rainfall reconstruction model, physical constraints on the rainfall variables are further added, as follows:

[0072]

[0073] In the formula: These represent the simulations of the rainfall reconstruction model. k The grid point at the th grid point t +1 and t Hourly rainfall at any given time, in mm; express , The smaller value in the model is used; specifically, the gradient descent method is employed to calibrate the rainfall reconstruction model and obtain the optimal parameters.

[0074] To achieve the long-term, high-resolution soil moisture data necessary for early warning of rapid shifts between drought and flood, the hourly-scale station soil moisture observation data obtained in Step 1 for the period 2000-2025 were further subjected to quality control and interpolation. A co-kriging method based on topography and soil characteristics was used to spatially interpolate the discrete station observation data onto a 0.1-degree grid consistent with the ERA5-Land data, generating a short-term (2000-2025) but highly reliable gridded soil moisture reference field.

[0075] Furthermore, a second deep learning model (using a Physical Information Neural Network PINN in this embodiment) is constructed, driven by long-series meteorological and hydrological elements and with a short-series gridded soil moisture reference field as the learning objective. The input driving factors of this model include long-series (1950-2025) gridded data provided by ERA5-Land, such as temperature, precipitation, evaporation, shortwave radiation, longwave radiation, surface runoff depth, and large-scale circulation anomaly indices (including ONI, SOI, IOD, and PDO indices).

[0076] The training objective of the model is to make the simulated soil moisture as close as possible to the gridded soil moisture baseline field during the period of 2000-2025. In the loss function of the Physical Information Neural Network (PINN), the following physical constraints based on the soil moisture balance equation are introduced to ensure that the simulated soil moisture changes are physically and logically consistent with hydrological processes such as precipitation, evaporation, and runoff, thereby improving the reliability of the model's extrapolation over historical periods (1950-2000).

[0077] The total loss function consists of a data fitting term and a physical constraint term, and its expression is:

[0078] ×

[0079] in: It is the total loss that needs to be minimized during model training; It is the data fitting loss, which calculates the root mean square error between the soil moisture simulated by the model and the generated gridded soil moisture reference field. This is a hyperparameter used to balance the weights of data and physical terms in the total loss; in this embodiment, it is set to 0.45; physical constraint term. L physics The core is a soil moisture balance equation based on a single layer. Is the physical information neural network in the first... i Simulated soil moisture at individual time points; The physical information neural network represents the first... i Simulated soil moisture at each time point Regarding time t The partial derivatives, which characterize the rate of change of soil moisture, can be automatically calculated through backpropagation of a neural network. P i It is the first i Rainfall data at each grid point (from ERA5-Land); E i It is the first i Evaporation at each grid point (from ERA5-Land); R i It is the first iSurface runoff depth at each grid point (from ERA5-Land); N This is the total number of spatiotemporal samples used in training. This physical constraint forces the model to learn dynamic changes in soil moisture that must be consistent with the main hydrological inputs (precipitation) and outputs (evaporation, runoff), thereby ensuring that even during the period of 1950–2000 when direct observations were lacking, the soil moisture series extrapolated by the model follows basic hydrophysical laws, significantly improving the physical rationality and reliability of long-series data reconstruction.

[0080] Furthermore, during the period 2000–2025, a 0.1-degree resolution gridded hourly soil moisture dataset reconstructed using the aforementioned PINN model, an ERA5-Land hourly soil moisture dataset, and discrete station observation data spatially interpolated to a 0.1-degree grid consistent with the ERA5-Land data were compared. For each grid point, the total error between the two datasets was calculated. :

[0081]

[0082] In the formula: , These represent the years 2000-2025, specifically the years 1-2025. k The grid point at the th grid point j Soil moisture reconstructed from PINN at a given time point, soil moisture spatially interpolated from site observation data, and soil moisture data from the ERA5-Land reanalysis dataset; NN represents the total number of grid points in the study area, and M represents the total number of hours from 2000 to 2025.

[0083] Furthermore, calculate the weights of the two sets of data:

[0084]

[0085]

[0086] In the formula: and These represent the weights of the PINN reconstructed dataset and the ERA5-Land dataset, respectively.

[0087] Furthermore, the following formula was used to obtain the final complete gridded hourly soil moisture dataset of the study area from 1950 to 2025 with a resolution of 0.1 degrees.

[0088]

[0089] In the formula: SMrec is the soil moisture reconstruction dataset finally used in this embodiment.

[0090] Finally, using this calibrated model, and driving a long series of meteorological and hydrological data and circulation indices from 1950 to 2000, a complete, 0.1-degree resolution gridded hourly soil moisture dataset for the study area from 1950 to 2025 can be reconstructed. This dataset, together with the reconstructed rainfall data, constitutes a comprehensive hydro-climatic foundation for analyzing and predicting rapid shifts between drought and flood.

[0091] In step 3 of this embodiment, for each microwave link in the study area, different rain attenuation and rainfall intensity models are used to establish the relationship between rain attenuation and rainfall intensity, and the corresponding link rainfall intensity is derived; and based on the long series of gridded rainfall data reconstructed in step 2, the rain attenuation and rainfall intensity model suitable for the study area is selected.

[0092] like Figure 2 As shown, base stations are generally distributed in a honeycomb pattern, with distances between them typically within 10km. Since the spatial distribution of the central base station and surrounding base stations is relatively uniform, microwave links are generally established between the central base station and surrounding base stations. Microwave links can also be established between adjacent surrounding base stations as needed. The average rain attenuation rate of the link is calculated based on the received and transmitted power of adjacent base stations. , length is L Received power on the link With transmission power With rain attenuation rate The relationship is:

[0093]

[0094] in, Represents the integral over length.

[0095] From the above formula, the average rain attenuation rate of the link can be derived. for:

[0096]

[0097] Furthermore, a rainfall attenuation and rainfall intensity model was established, and the link-average rainfall intensity was derived. R。

[0098] In existing technologies, different rain attenuation and rainfall intensity models are applicable to different frequency bands and different systems of ground-to-ground and ground-to-air links. It is difficult to determine a suitable rain attenuation and rainfall intensity model before evaluation using large-scale satellite remote sensing data. Therefore, different models are used to establish the relationship between rain attenuation and rainfall intensity, i.e., the link's average rain attenuation rate. and average rainfall intensity R The relationship between them was determined, and the average rainfall intensity of each link was calculated. RModels include, but are not limited to, Crane Global model, Assis Einloft Improved model, SAM model, DAH model, ITU-R P.618 model, China model, ITU-R P.838 model, Moupfouma model, EXCELL model, and Garcia-Lopez model.

[0099] For example, the relationship between rainfall attenuation and rainfall intensity can be established using the ITU-R P.618 model, as shown in the following equation:

[0100]

[0101] In the formula, k and The parameter can be estimated using existing methods recommended in ITU reports. According to the above formula, the link average rain attenuation rate... The average rainfall intensity of the link was obtained by reverse calculation. R This means that the average rainfall intensity of the link can be derived using the ITU-R P.618 model.

[0102] In practice, the relationship between rainfall attenuation and rainfall intensity is established, and can be implemented according to the specific model selected. This invention will not elaborate on this point.

[0103] like Figure 3 As shown, considering the geographical conditions of the area, the density of base station deployment, and other availability information, the observation area is reasonably divided into grids of appropriate scale. In this embodiment, it is assumed that the area is divided into G identical grids. Figure 3 Taking a microwave link as an example, point F is the signal transmitting base station and point Z is the signal receiving base station, which constitutes a single microwave link.

[0104] For each of the above-established rainfall attenuation and rainfall intensity models, the rainfall corresponding to each grid under different models can be obtained: such as Figure 4 As shown, for each grid, if it contains microwave links, under different rain attenuation and rainfall intensity models, the historical microwave link information (microwave signals) included in the grid can be extracted based on the average rainfall intensity of the corresponding links; for example, a certain grid Gr contains M There are 10 microwave links, and the length of each microwave link in the grid is denoted as . L j ( j =1, ... , M The average rainfall intensity of each link, calculated in step 2, is expressed as follows: R j ( j =1, ... , M The weighted link method was used to obtain the rainfall for this grid. Prep for:

[0105]

[0106] In the formula, sumL The length of each microwave link in the grid within that grid. L j sum.

[0107] like Figure 5 As shown, some grids in the region do not contain microwave links. For such grids, a reasonable radius threshold is selected, and microwave links within the radius threshold are searched. Then, the reciprocal distance weighting method is used to calculate the rainfall for that grid. For example, M microwave links are found within the radius threshold of a certain grid, and the vertical distance from the grid center point to each link is... S j ( j =1, ... , M If the rainfall in the grid is such that the reciprocal distance weighting method is used, then the rainfall in that grid can be obtained. Prep for:

[0108]

[0109] In the formula, sumS For the grid sum.

[0110] Figure 5 In a certain grid G Three microwave links were obtained by searching within a radius threshold of 0. The perpendicular distance from the grid center point to each link is... S 1. S 2. S 3. Rainfall in this grid .

[0111] For each established rainfall attenuation and intensity model, the corresponding historical regional rainfall for different models can be calculated. Let the number of rainfall attenuation and intensity models be... N The number of grids is G Let the first g The grid passes through the first i ( i =1,..., N The grid rainfall obtained from the rainfall attenuation and rainfall intensity models is: Then the first one can be obtained. i Historical regional rainfall corresponding to the rainfall attenuation and intensity model. AR i for:

[0112]

[0113] Furthermore, the hourly rainfall at each grid point in the study area since 1950, reconstructed in step 2, is used as the true observed value. The regional rainfall corresponding to different rainfall attenuation and intensity models derived using rainfall attenuation data is used as the simulated value. The combined index of Nash (Nash efficiency coefficient), RMSE (root mean square error), and Pearson correlation coefficient is used as the evaluation index to select the best rainfall attenuation and intensity model.

[0114]

[0115] In the formula: Index is an evaluation index used to select the optimal rain attenuation and rainfall intensity model. The higher the Index, the better the model performance.

[0116] In step 4 of this embodiment, based on the preferred rainfall attenuation and intensity model, a correction module for the rainfall attenuation and intensity model is constructed by further considering the rainfall data retrieved from meteorological satellites, and a correction module for the intensity of drought-flood transition events is constructed by extracting drought-flood transition events based on the reconstructed dataset.

[0117] Using the optimized rain attenuation and intensity model obtained in step 3, and inputting the microwave link information of the study area, the average regional rainfall over a historical period (in this embodiment, the historical period refers to the period from the establishment of the mobile base station to 2025) can be obtained, denoted as . AAR t ,in t For a moment.

[0118] Furthermore, based on the hourly rainfall dataset from 2000 to 2025 provided by the GPM IMERG-Final version collected in step 1, the average hourly rainfall series of the study area is derived. Furthermore, using the average hourly rainfall series of the study basin retrieved by the meteorological satellite as a benchmark, the microwave signals observed by each mobile communication station in the study area and the regional average rainfall AAR during historical periods are used as inputs. Considering the time lag effect (such as 2 hours), a calibration module for the rainfall attenuation and rainfall intensity model is established through a calibrated convolutional neural network model.

[0119] Based on the data obtained from the above steps, a convolutional neural network model is constructed. Specifically, this method utilizes the average rainfall in the study area derived from an optimized rainfall attenuation and intensity model and microwave signals observed by various mobile communication stations as explanatory variables, with the average hourly rainfall series of the study basin retrieved from meteorological satellites as the target variable. The method first preprocesses the rainfall data and microwave signals, including data partitioning, normalization, and reshaping, to adapt to the input format of the convolutional neural network. Then, a three-layer neural network architecture containing convolutional layers, pooling layers, and fully connected layers is constructed. The convolutional layers extract spatial features from the data, the pooling layers reduce data dimensionality, and the fully connected layers achieve feature integration and rainfall simulation. Finally, a temperature scaling method is used to calibrate the model's output probability to improve the accuracy of the simulated probability. The convolutional neural network constructed in this implementation is used to correct the errors of the rainfall attenuation and intensity model and improve the accuracy of capturing the average rainfall in the study area.

[0120] The convolutional neural network model constructed using the method described above can be represented as follows:

[0121]

[0122] In the formula: Indicates the period from 2000 to 2025. t The hourly average rainfall in the study area after time correction. These represent the periods from 2000 to 2025. t The regional average rainfall at time t-1 was calculated using the rain attenuation and rainfall intensity model optimized in step 4. During the period from 2000 to 2024 respectively t The microwave signals observed by all mobile communication stations in the study area at time t-1; F DL The diagram shows the calibrated convolutional neural network model. To optimize the model's parameters, the minimum batch gradient descent method, a conventional technique in this field, is used to train the convolutional neural network model.

[0123] Furthermore, based on the long-series, high-resolution gridded rainfall (Prep_rec) and soil moisture (SM_rec) datasets reconstructed in step 2 from 1950 to 2025, and the aforementioned corrected regional average rainfall series, This module is used to accurately identify historical drought-flood transition events and determine their intensity.

[0124] First, we define the drought-flood transition index. For a given time window Δt, the drought-flood transition index FFAI(t) at time t is defined as a comprehensive representation of the intensity of the preceding drought and the intensity of the subsequent flood:

[0125] FFAI(t) = [ (SPI_{-τ1} - SPI_{-τ2}) ]×[ (SSI_{0} - SSI_{-τ1}) ]×W

[0126] In the formula: SPI is the standard precipitation index, used to characterize meteorological drought / flooding. SPI_{-τ2} and SPI_{-τ1} represent the accumulated SPI at time t-τ2 and t-τ1 respectively (τ2>τ1), and the difference (SPI_{-τ1} - SPI_{-τ2}) reflects the intensity of the meteorological transition from drought to flooding; SSI is the standard soil moisture index, used to characterize agricultural drought / waterlogging; in this embodiment, the soil dataset reconstructed in step 2 is used for calculation; SSI_{0} and SSI_{-τ1} represent the SSI at time t and t-τ1 respectively, and the difference (SSI_{0} - SSI_{-τ1}) reflects the intensity of soil moisture response to the rainfall process; W is a weighting factor, which can be calibrated according to the underlying surface characteristics of the study area, used to balance the influence of meteorological and agricultural hydrology, and is taken as 1.0 in this embodiment.

[0127] Furthermore, historical drought-flood transition events are extracted. When FFAI(t) exceeds a preset threshold (determined by percentile method, in this embodiment, the 95th percentile of the historical sequence is used), it is considered that a drought-flood transition event has occurred at time t. This embodiment uses hourly scale data, which can realize the identification of drought-flood transition events at the hourly scale.

[0128] Next, an event intensity correction module is constructed. Since microwave links differ from satellite remote sensing in their ability to capture rainfall, especially under severe convective weather, the two are complementary. This embodiment utilizes the corrected CorP sequence to perform secondary calibration of the intensity of the FFAI index calculated based on reconstructed data at the time of the event, establishing an intensity correction relationship:

[0129] FFAI_final(t_event)=α×FFAI_recon(t_event)+β×(CorP(t_event) / AAR(t_event))

[0130] In the formula: FFAI_final(t_event) is the final drought-flood transition intensity after correction at the event time t_event; FFAI_recon(t_event) is the original drought-flood transition intensity calculated based on the grid data reconstructed in step 2; CorP(t_event) and AAR(t_event) are the regional average rainfall corrected by this module and the regional average rainfall calculated from the reconstructed rainfall data at time t_event, respectively; α and β are regression coefficients obtained by fitting the historical event dataset.

[0131] This formula essentially uses real-time rainfall information fused and corrected from microwave links and satellite data as key evidence to correct and enhance event intensity judgments based solely on historical reconstruction data. This makes the calibration of historical events closer to the "truth," providing a more reliable intensity benchmark for the early warning module. Ultimately, this step outputs a calibrated, high-precision database of historical drought-flood transition events, where each event includes its occurrence time, duration, and calibrated FFAI_final intensity, laying a solid foundation for the next step of building a dynamic early warning model.

[0132] In step 5 of this embodiment, a dynamic early warning module for abrupt shifts between drought and flood is constructed by fusing circulation signals. An attention mechanism network is introduced to construct the "dynamic early warning module for abrupt shifts between drought and flood". The inputs of this module include: real-time rainfall inverted from the optimal rainfall intensity model, calculated real-time drought-flood abrupt shift index, satellite-inverted rainfall, real-time microwave signals, and real-time large-scale circulation anomaly signals as forward-looking indicator factors.

[0133] Step 5 further includes the following sub-steps:

[0134] This step aims to overcome the limitations of traditional hydrological and meteorological early warning systems that rely solely on current and past conditions. By introducing large-scale circulation signals with physical predictability and utilizing deep learning models with attention mechanisms, it achieves forward-looking dynamic early warning of the probability and intensity of sudden shifts between drought and flood events.

[0135] First, a multidimensional, spatiotemporally aligned input feature sequence is constructed for the early warning model. For the current early warning time T, a time window of length L (e.g., L=72 hours) [T-L+1, T] is constructed, and the following multi-source early warning factors are extracted within this window:

[0136] Real-time rainfall sequence: including R_mw_corrected(t): the corrected regional average hourly rainfall sequence output by the rainfall attenuation model correction module in step 4 (t belongs to [T-L+1, T]). This sequence represents the rainfall information closest to reality; R_satellite(t): the regional average hourly rainfall sequence retrieved from meteorological satellites (such as GPM), serving as an independent source of rainfall observations.

[0137] Real-time underlying surface response sequence: Based on the long-series soil moisture data SM_rec reconstructed in step 2 and the corrected rainfall data in step 4, a real-time (quasi-real-time) drought-flood transition index FFAI(t) sequence was calculated. This index comprehensively reflects the intensity of the transition between previous meteorological drought and current soil flooding, and serves as a direct indicator signal for early warning.

[0138] Real-time microwave signal feature sequences: Extracting the dynamic characteristics of microwave signals from all mobile communication base stations within the study area over a time window, such as the regional average received signal level RSL(t) and bit error rate BER(t). These signals contain comprehensive information on atmospheric stratification, turbulence, and precipitation particle concentration, and can serve as supplementary probes for rainfall and atmospheric instability.

[0139] Prospective circulation forcing signals: Real-time large-scale circulation anomalies are acquired as prospective indicators. Although these signals do not directly affect the study area, their evolution has predictive significance for regional weather and climate for 1-4 weeks. Specifically, they include the ONI, SOI, IOD, and PDO indices. To match the real-time series, the values ​​of these indices within the time window [T-L+1, T] are used as input, and the model will learn their teleconnection patterns with subsequent abrupt shifts in regional drought and flood.

[0140] After standardizing all the above factors, they are concatenated to form a multidimensional feature tensor X(T), which is related to the total dimension of the features. Furthermore, an attention mechanism network is introduced to construct a "dynamic early warning module for sudden shifts between drought and flood". The core of this module is an encoder-decoder architecture incorporating a multi-head self-attention mechanism, whose advantage lies in its ability to automatically weigh the relative importance of different early warning factors at different times.

[0141] Encoder: The encoder receives the input feature tensor X(T). It first passes through a multi-head self-attention layer. This mechanism allows the model to interact within the sequence; for example, it can learn that "when a specific circulation signal (such as a strong El Niño) occurs, the continuous light rain signal of the past 24 hours is more dangerous than the signal itself," thus dynamically assigning different attention weights to different times and features in the sequence. The output of the self-attention layer is then nonlinearly transformed by a feedforward neural network, ultimately outputting a context code containing all input information and its complex dependencies.

[0142] Decoder and Early Warning Output: The decoder's task is to predict the risk of a sudden shift from drought to flood within the next τ hours (e.g., τ=48 hours) based on the encoder's output; the decoder's query is the location code for the future time, while the key and value come from the encoder's output; the decoder's final output is two key early warning products: 1) the probability of a sudden shift from drought to flood occurring hourly in the future: P(FFAI>Threshold | t), t = T+1, T+2, ..., T+τ; 2) the intensity index of a sudden shift from drought to flood occurring hourly in the future: FFAI_forecast(t), t = T+1, T+2, ..., T+τ.

[0143] Wherein, FFAI represents the content abrupt change index; Threshold | t represents the content abrupt change threshold selected in this embodiment; t represents time; FFAI_forecast(t) represents the drought-flood abrupt change intensity index at time t; the mathematical expression of the model can be simplified as follows:

[0144] {P_forecast, FFAI_forecast} = F_Attn(R_mw_corrected, R_satellite,FFAI, MS, Climate_Indices )

[0145] In the formula: FFAI_forecast represents the predicted drought-flood transition intensity index at different future times; P_forecast represents the predicted precipitation level; F_Attn represents the trained attention mechanism network model; R_mw_corrected represents the corrected regional average hourly precipitation sequence output by the rain attenuation model correction module in step 4; R_satellite represents the regional average hourly precipitation sequence retrieved from meteorological satellites (such as GPM) as an independent source of precipitation observation; MS represents microwave signal characteristics; and Climate_Indices refers to the forward-looking circulation forcing signal.

[0146] A large number of training samples are constructed using historical data, with real historical drought-flood transition events and their intensity used as training objectives. The model parameters are calibrated by minimizing prediction errors using gradient descent (e.g., binary cross-entropy loss for probability prediction and mean squared error for intensity prediction). After the model is trained, it can be put into operational use, receiving real-time multi-source data streams and automatically generating dynamic early warning products for future periods.

[0147] The innovation of this module lies in its organic integration of real-time surface conditions from ground base stations and satellite remote sensing with macroscopically predictive climate-driven signals within a deep learning framework. It achieves intelligent information filtering through an attention mechanism, thus realizing a leap from "current status monitoring" to "mechanism-driven forward-looking early warning".

[0148] In step 6 of this embodiment, a real-time drought-flood transition risk warning is generated and issued. Based on the output of the dynamic warning module in step 5, a comprehensive drought-flood transition risk map is generated, including risk level, transition intensity, potential impact range, and warning timeliness. The final visualized warning information is then released to relevant departments and the public through an early warning platform or interface, providing decision support for disaster prevention and mitigation.

[0149] Step 6 specifically includes the following sub-steps:

[0150] This step is the final output of the system. Its core function is to transform the numerical forecast results of the dynamic early warning module in step 5, which contain the probability of a sudden shift from drought to flood (P_forecast) and the intensity index (FFAI_forecast) for future periods, into a comprehensive drought-flood sudden shift risk map that integrates four key elements: risk level, intensity of the sudden shift, potential impact range, and early warning timeliness. The generation of this map is an automated and structured process, and the specific conversion rules are shown in the table below:

[0151]

[0152] Based on this, structured early warning reports for key areas, such as cities and key agricultural areas identified in the map, are automatically generated. These reports not only include an early warning summary but also incorporate model attention weight analysis of the dominant risk factors (such as the combined effect of localized severe convection and the El Niño background field), and provide specific disaster prevention and mitigation recommendations. For example, municipal departments should guard against urban flooding, natural resources departments should pay attention to flash floods and geological disasters, and agricultural departments should guide field drainage and be wary of damage to crop roots caused by drought-to-flood transitions.

[0153] To achieve efficient information transmission, the aforementioned early warning products are integrated and published through a visualization and decision support platform. The risk map will be overlaid on the WebGIS platform with electronic maps, administrative divisions, and disaster-bearing body data (such as population and infrastructure) to form a unified risk map, providing interactive query functionality. Simultaneously, the system will push early warning data to the business systems of departments such as water resources and emergency management in real time through standardized API interfaces.

[0154] Ultimately, through a tiered and categorized release mechanism, customized early warning information is precisely delivered to target users. For professional decision-making departments: complete technical products and decision reports are released through government platforms and data interfaces. For the general public: concise, clear, and specific early warning information is released through official new media, mobile apps, radio, and television channels, thus constructing a complete decision support chain from accurate forecasting to effective disaster prevention.

[0155] The invention of this embodiment is as follows: Real-time monitoring data, rainfall data retrieved from meteorological satellites, real-time microwave signal data from mobile communication base stations in the study area, historical and background field data, and large-scale climate-driven signals are collected to form a unified multi-source data cube. Then, historical long-series, high-resolution gridded rainfall and soil moisture data of the study area are reconstructed to accurately characterize the region's historical drought and flood background. Different rainfall attenuation and intensity models are used to establish the relationship between rainfall attenuation and intensity, and the corresponding link rainfall intensity is derived. A rainfall attenuation and intensity model suitable for the study area is selected. Based on the selected rainfall attenuation and intensity model, a correction module for the rainfall attenuation and intensity model is constructed, taking into account rainfall data retrieved from meteorological satellites. A correction module for the intensity of drought-flood abrupt transition events is constructed based on the reconstructed dataset. Finally, a dynamic early warning module for drought-flood abrupt transition events, incorporating circulation signals, is constructed to generate and issue real-time drought-flood abrupt transition risk warnings.

[0156] Example 2

[0157] This embodiment provides a drought and flood emergency warning system that couples satellite and ground base station signals, including:

[0158] Multi-source data acquisition module: It is used to acquire multi-source data of the study area and perform fusion preprocessing;

[0159] Data reconstruction module: It is used to load the hydro-climate reconstruction model and reconstruct long-series gridded rainfall data and soil moisture data of the study area based on the multi-source data.

[0160] The rain attenuation and intensity model optimization module is used to optimize various rain attenuation and intensity models retrieved from microwave link signals based on reconstructed long-series gridded rainfall data. The calibration module is used to construct a calibration module for the rain attenuation and intensity models based on the optimized models and rainfall data retrieved from meteorological satellites, calibrating the real-time regional rainfall sequence retrieved from microwave links. It also calculates the real-time drought-flood abrupt change index based on the reconstructed long-series gridded rainfall data and soil moisture data, constructs a drought-flood abrupt change event intensity calibration module, and generates a calibrated historical drought-flood abrupt change event database. The drought-flood abrupt change dynamic early warning module is used to construct a drought-flood abrupt change dynamic early warning module that integrates circulation signals. It takes the calibrated real-time regional rainfall sequence, real-time drought-flood abrupt change index, real-time microwave signal characteristics, and large-scale circulation anomaly signals as inputs, and outputs early warning results for future drought-flood abrupt changes.

[0161] The results output module is used to generate and publish drought and flood rapid change risk warning information based on the warning results output by the drought and flood rapid change dynamic warning module. This information includes the risk level, rapid change intensity, impact range and warning time.

[0162] It should be understood that any parts not described in detail in this specification belong to the prior art.

[0163] It should be understood that the above description of the preferred embodiments is quite detailed, but this should not be construed as limiting the scope of protection of this invention. It is neither necessary nor possible to exhaustively describe all possible implementations. Those skilled in the art, guided by this invention, can make substitutions or modifications without departing from the scope of the claims, all of which fall within the scope of protection of this invention. The scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals, characterized in that, include: Acquire multi-source data of the study area and perform fusion preprocessing; Load the hydro-climate reconstruction model, and reconstruct long-series gridded rainfall data and soil moisture data of the study area for historical periods based on the multi-source data; Based on reconstructed long-series gridded rainfall data, various rainfall attenuation and intensity models derived from microwave link signals are optimized to obtain the best model. Using the regional average rainfall and corresponding microwave signals retrieved from the optimal model as explanatory variables, and the regional average hourly rainfall sequence retrieved from meteorological satellites as the target variable, a convolutional neural network model is constructed and trained to correct the historical regional average rainfall sequence retrieved from the optimal model. Based on the reconstructed long-series gridded rainfall data and soil moisture data of the study area, a historical drought-flood transition index is defined and calculated, and historical drought-flood transition events are extracted based on a preset threshold. Using the corrected historical regional average rainfall sequence, the intensity of the calculated drought-flood transition event is recalibrated, an intensity correction relationship is established, and a calibrated historical drought-flood transition event database is generated. A dynamic early warning module for drought-flood transition fused with circulation signals is constructed, which takes the corrected real-time regional rainfall sequence, real-time drought-flood transition index, real-time microwave signal characteristics, and large-scale circulation anomaly signal as input, and outputs the early warning result of drought-flood transition in the future period. Based on the early warning results output by the dynamic early warning module for rapid shifts between drought and flood, early warning information on the risk of rapid shifts between drought and flood, including risk level, intensity of the rapid shift, scope of impact, and warning time, is generated and released.

2. The method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals according to claim 1, characterized in that, The multi-source data includes real-time monitoring data, historical and background field data, and large-scale climate driving signals; the real-time monitoring data includes rainfall data retrieved from meteorological satellites and real-time microwave signal correlation data from mobile communication base stations in the study area; the historical and background field data includes meteorological and hydrological data provided by reanalysis datasets and meteorological and hydrological forecast products output by Earth system models; the large-scale climate driving signals include sea surface temperature anomaly data and large-scale circulation anomaly indices in key sea areas.

3. The method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals according to claim 1, characterized in that, The fusion preprocessing includes a spatiotemporally aligned dataset formed by performing spatiotemporal matching, standardization, and quality control on multi-source data.

4. The method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals according to claim 2, characterized in that, The reconstructed long-term gridded rainfall and soil moisture data for the historical period of the study area include: Using the reanalysis dataset, meteorological and hydrological data output from the Earth system model, and large-scale circulation index as inputs, the first deep learning model is used to reconstruct long-series gridded rainfall data of the historical period in the study area; Driven by meteorological and hydrological data and large-scale circulation indices provided by long-series reanalysis datasets, and combined with short-series soil moisture reference fields generated by site observation interpolation and their physical constraints, a second deep learning model was used to reconstruct long-series gridded soil moisture data for historical periods in the study area.

5. The method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals according to claim 4, characterized in that, The optimization of various rainfall attenuation and intensity models derived from microwave link signals includes: For each microwave link within the study area, the relationship between rainfall attenuation and rainfall intensity is established using multiple candidate rainfall attenuation and rainfall intensity models, and the rainfall intensity of the corresponding microwave link is estimated. Based on the rainfall intensity of the microwave link, the regional rainfall for the corresponding historical period is calculated for each candidate rainfall attenuation and rainfall intensity model. Using long-series gridded rainfall data from historical periods in the reconstructed study area as the observed true values, and the regional rainfall corresponding to each candidate rainfall attenuation and intensity model as the simulated values, the Nash efficiency coefficient, root mean square error, and / or Pearson correlation coefficient were used as evaluation indicators to obtain the optimal rainfall attenuation and intensity model.

6. The method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals according to claim 1, characterized in that, The dynamic early warning module for rapid drought and flood transition adopts an encoder-decoder network architecture that incorporates an attention mechanism; the encoder is used to encode the multidimensional feature sequence of the input, including corrected real-time rainfall, satellite-retrieved rainfall, real-time drought and flood transition index, microwave signal features, and large-scale circulation anomaly signals. The decoder is used to predict the probability and intensity index of sudden shifts between drought and flood in future periods based on the encoding results.

7. The method for early warning of sudden shifts between drought and flood based on coupled satellite and ground base station signals according to claim 6, characterized in that, The specific rules for generating the aforementioned drought-flood transition risk warning information include: Risk levels are dynamically classified based on the predicted probability of a sudden shift from drought to flood. Map the predicted drought-flood transition intensity index to transition intensity; Spatial analysis is used to identify and delineate continuous areas that reach a predetermined risk level, which serve as the scope of impact. Clearly indicate the effective period of the warning results.

8. A drought and flood rapid shift early warning system that couples satellite and ground base station signals, characterized in that, include: Multi-source data acquisition module: It is used to acquire multi-source data of the study area and perform fusion preprocessing; Data reconstruction module: It is used to load the hydro-climate reconstruction model and reconstruct long-series gridded rainfall data and soil moisture data of the study area based on the multi-source data. The rain attenuation and intensity model optimization module is used to optimize various rain attenuation and intensity models derived from microwave link signals based on reconstructed long-series gridded rainfall data to obtain the optimal model. The correction module uses the regional average rainfall for historical periods derived from the optimal rain attenuation and intensity model and the corresponding microwave signal as explanatory variables, with the regional average hourly rainfall sequence derived from meteorological satellites as the target variable. It constructs and trains a convolutional neural network model to correct the regional average rainfall sequence for historical periods derived from the optimal rain attenuation and intensity model. Based on the reconstructed long-series gridded rainfall data and soil moisture data for the study area, it defines and calculates the historical drought-flood abrupt reversal index and extracts historical drought-flood abrupt reversal events based on a preset threshold. Using the corrected historical regional average rainfall sequence, the calculated intensity of drought-flood transition events is recalibrated, an intensity correction relationship is established, and a calibrated historical drought-flood transition event database is generated. Drought-flood transition dynamic early warning module: It is used to construct a drought-flood transition dynamic early warning module that integrates circulation signals. It takes the corrected real-time regional rainfall sequence, real-time drought-flood transition index, real-time microwave signal characteristics, and large-scale circulation anomaly signals as inputs, and outputs early warning results of drought-flood transitions in future periods. Results output module: It is used to generate and publish drought and flood rapid change risk warning information based on the warning results output by the drought and flood rapid change dynamic warning module, including risk level, rapid change intensity, impact range and warning time. The drought and flood abrupt change early warning system that couples satellite and ground base station signals is used to perform the steps in the drought and flood abrupt change early warning method that couples satellite and ground base station signals according to any one of claims 1-7.

Images