A basin local drought and flood early warning method and device

Abstract

A basin local drought and flood early warning method and device, the method comprising: predicting future land use spatial distribution based on future climate data, natural environment and social economic driving factor data of a target basin and land use data in a historical period; the historical period includes at least two different time nodes; predicting a future leaf area index dynamic sequence according to the future climate data and a pre-trained leaf area index prediction model; inputting the land use spatial distribution, the future leaf area index dynamic sequence and the future climate data into a distributed hydrological model to obtain actual runoff and intermediate layer soil terminal water content, and combining climate observation data, soil available water capacity and a drought index model to calculate a drought index; and warning according to a preset risk level corresponding to the drought index. The method realizes the collaborative coupling of climate, land use and leaf area index, and can improve the simulation accuracy of basin local drought and flood response and the prediction accuracy of drought and flood risk.

Description

Technical Field

[0001] This application relates to the field of hydrological prediction technology, specifically to a method and device for early warning of local droughts and floods in a watershed. Background Technology

[0002] The combined impacts of climate change and human activities have profoundly altered the water cycle processes in river basins. Global warming has exacerbated the spatial and temporal differentiation of the water cycle, leading to a simultaneous increase in precipitation in humid regions and increased evaporation in arid regions, resulting in a rise in the frequency and intensity of localized droughts and floods in river basins. Simultaneously, land use / cover change (LUCC) caused by human activities further alters underlying surface conditions, interacting with climate change to jointly influence runoff formation and evolution.

[0003] In the complex water cycle system driven by both climate and human activities, vegetation serves as a dynamic interface connecting the atmosphere, soil, and hydrology. Among these, the leaf area index (LAI), a key parameter characterizing vegetation canopy structure, directly influences surface evapotranspiration, interception, and runoff processes. However, traditional hydrological models typically perform simple parameterization of climate, land use, and vegetation elements, failing to fully consider the dynamic interactions and co-evolutionary mechanisms among them. This results in insufficient accuracy in simulating local drought and flood responses within watersheds, particularly in capturing the evolutionary patterns of extreme drought and wet events under future scenarios. Existing drought assessment indices often employ fixed parameters, lacking mechanisms for dynamic calibration using historical climate data from different regions. This leads to systematic biases in drought assessment results across different climatic zones, resulting in poor spatial comparability. Summary of the Invention

[0004] This application provides a method and apparatus for early warning of local drought and flood in watersheds, which can solve the technical problem that the existing technology cannot combine climate, land use and vegetation leaf area index, resulting in insufficient simulation accuracy of local drought and flood response in watersheds, leading to inaccurate prediction of drought and flood risks under future scenarios.

[0005] To achieve the above objectives, in a first aspect, this application provides a method for early warning of local droughts and floods in a watershed, the method comprising: Based on future climate data, natural environment and socio-economic driving factor data, and historical land use data for the target watershed, predict the future spatial distribution of land use; the historical period includes at least two different time points.

[0006] Based on the aforementioned future climate data and a pre-trained leaf area index prediction model, predict the future dynamic sequence of leaf area index.

[0007] The spatial distribution of land use, the dynamic sequence of future leaf area index, and the future climate data are input into a distributed hydrological model to obtain the actual runoff and the terminal water content of the intermediate soil layer. The drought index is then calculated by combining climate observation data, soil effective water holding capacity, and the drought index model.

[0008] Early warnings are issued based on the preset risk level corresponding to the drought index.

[0009] Furthermore, in one embodiment, the future climate data is derived from the International Climate Modeling Project, including temperature, precipitation, wind speed, and relative humidity.

[0010] Furthermore, in one embodiment, the prediction of future land use spatial distribution includes: The expansion areas of each land use type are extracted from the land use data, and the contribution of the natural environment and socio-economic driving factors to the expansion of each land use type is calculated to generate a development probability distribution map of each land use type; the land use types include cultivated land, forest land and grassland.

[0011] Based on the aforementioned future climate data and the socio-economic development assumptions corresponding to the shared socio-economic path, the total demand for each land use type at multiple preset time nodes in the future is generated through a system dynamics model.

[0012] Based on the development probability distribution map, neighborhood influence coefficient, land use type transfer matrix, and total demand, a patch-based land use simulation model is used to simulate spatial layout and generate the future spatial distribution of land use.

[0013] Furthermore, in one embodiment, the land use data is a historical raster data layer containing the land use type.

[0014] The natural environment and socio-economic driving factors include topographic factors, soil properties, climate factors, and socio-economic factors.

[0015] The topographic factors include elevation, slope, and aspect; the soil properties include soil type and organic matter content; the climate factors include distance from the river, average annual precipitation, and average annual temperature; and the socioeconomic factors include population density, spatial distribution of GDP, transportation network, and distribution of settlements.

[0016] Furthermore, in one embodiment, the predicted future leaf area index dynamic sequence includes: Multiple sets of future climate data are input into the leaf area index prediction model to obtain multiple preliminary leaf area index prediction sequences.

[0017] The multiple preliminary leaf area index prediction sequences are integrated using a multi-model ensemble averaging method to obtain a dynamic sequence of future leaf area index.

[0018] Furthermore, in one embodiment, the method for calculating the actual runoff includes: The direct runoff is calculated based on the water content of the intermediate soil layer at the previous moment, the precipitation in the future climate data, and the saturated water content and maximum infiltration capacity of the intermediate soil layer in the spatial distribution of land use.

[0019] The infiltration rate is calculated based on the water content of the intermediate soil layer at the previous moment, as well as the residual soil moisture, saturated permeability coefficient, soil pore size distribution index, and saturated water content of the intermediate soil layer in the land use spatial distribution.

[0020] The sum of the direct runoff and the infiltration is taken as the actual runoff.

[0021] Furthermore, in one embodiment, the method for calculating the terminal moisture content of the intermediate soil layer includes: Based on the future climate data, the future leaf area index dynamic sequence, and the future land use spatial distribution, calculate canopy evaporation, transpiration, and bare soil evaporation.

[0022] The actual evapotranspiration is calculated based on the canopy evaporation, the transpiration, the bare soil evaporation, and the vegetation cover ratio and bare soil cover ratio in the future land use spatial distribution.

[0023] The terminal moisture content of the intermediate soil layer is calculated based on the moisture content of the intermediate soil layer at the previous moment, the precipitation, the direct runoff, the infiltration, and the actual evapotranspiration.

[0024] Furthermore, in one embodiment, the calculation of the drought index includes: Based on the climate observation data and soil effective water holding capacity, potential evapotranspiration, potential recharge, potential runoff, and potential loss are calculated using a two-layer soil model.

[0025] The climate-appropriate precipitation is calculated based on the potential evapotranspiration, the potential replenishment, the potential runoff, and the potential loss.

[0026] The moisture anomaly index for the current month is calculated based on the precipitation, the climate-suitable precipitation, the actual runoff, the actual evapotranspiration, and the terminal moisture content of the intermediate soil layer.

[0027] The historical water anomaly index sequence was divided into wet and dry periods according to positive and negative values, respectively, and linear regression models were established for each period to obtain the corresponding duration factor.

[0028] The drought index for the current month is calculated using a recursive formula based on the current month's moisture anomaly index, the previous month's drought index, and the duration factor.

[0029] The climate observation data includes temperature, precipitation, wind speed, and relative humidity.

[0030] Furthermore, in one embodiment, the step of issuing an early warning based on a preset risk level corresponding to the drought index includes: Determine the risk level corresponding to the drought index and generate corresponding early warning information.

[0031] The risk levels are set based on the drought index and include mild drought, moderate drought, severe drought, extreme drought, mild humidity, moderate humidity, severe humidity, and extreme humidity.

[0032] Secondly, this application provides a localized drought and flood early warning device for a watershed, the device comprising: The land use module is used to predict the future spatial distribution of land use based on future climate data, natural environment and socio-economic driving factors data, and historical land use data of the target watershed.

[0033] The leaf area index module is used to predict the future dynamic sequence of leaf area index based on the future climate data and a pre-trained leaf area index prediction model.

[0034] The drought index module is used to input the land use spatial distribution, the future leaf area index dynamic sequence, and the future climate data into the distributed hydrological model to obtain the actual runoff and the terminal water content of the intermediate soil layer, and to calculate the drought index by combining climate observation data, soil effective water holding capacity, and the drought index model.

[0035] The early warning module is used to issue early warnings based on the preset risk level corresponding to the drought index.

[0036] The beneficial effects of the technical solutions provided in this application include: This application predicts the future spatial distribution of land use based on future climate data, natural environment and socio-economic driving factor data, and historical land use data of the target watershed, achieving a synergistic representation of climate change and human activities at the land use evolution level. It also predicts the future dynamic sequence of leaf area index (LAI) based on the future climate data and a pre-trained LAI prediction model, coupling vegetation dynamic response with future climate conditions. Furthermore, by inputting the land use spatial distribution, the future LAI dynamic sequence, and the future climate data into a distributed hydrological model, it obtains actual runoff and intermediate soil terminal moisture content, and calculates the drought index using climate observation data and a drought index model, constructing a synergistic driving mechanism of climate-land use-vegetation-hydrology-drought and flood. This allows for early warning based on the drought index and a preset risk level.

[0037] This method achieves synergistic coupling of climate, land use, and leaf area index, which can effectively improve the simulation accuracy of local drought and flood responses in watersheds, thereby improving the accuracy of drought and flood risk prediction under future scenarios. Attached Figure Description

[0038] Figure 1 This is a flowchart of a watershed local drought and flood early warning method according to an embodiment of this application.

[0039] Figure 2 This is a detailed flowchart of step S1 in an embodiment of this application.

[0040] Figure 3 This is a detailed flowchart of step S2 in an embodiment of this application.

[0041] Figure 4 This is a block diagram of a localized drought and flood early warning device for a watershed, as described in an embodiment of this application. Detailed Implementation

[0042] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0043] The report indicates that since the pre-industrial period (1850-1900), the rise in global average land surface temperature has been significantly higher than the rise in global average surface temperature. Climate observation data shows that the significant warming of land surface temperature by 1.53°C has exceeded the global average warming by 0.87°C, with this difference peaking between 2006 and 2015. Compared to the pre-industrial climate baseline, the thermal inertia of the ocean system has buffered the effects of the greenhouse effect, and the difference in thermal response sensitivity between land and ocean directly leads to the spatial differentiation of global warming rates. Climate warming accelerates the terrestrial water cycle, increasing global water vapor transport intensity while altering the relationship between evaporation and precipitation. This results in increased precipitation intensity in humid regions and intensified evaporation in arid regions, causing uneven spatial and temporal distribution of regional water resources. On the one hand, this induces persistent regional droughts and associated soil erosion; on the other hand, it leads to frequent heavy rainfall events and floods. This synergistic enhancement of dry and wet extremes ultimately leads to a chain reaction of imbalances in regional ecosystems by altering surface runoff patterns and disrupting the hydrological regulation function of vegetation. The increased frequency and intensity of compound risks and extreme events caused by climate change, making disaster risks more complex, cross-regional, and cross-systemic, pose new challenges to future disaster prevention and mitigation efforts.

[0044] Compared to the impacts of climate change on the watershed water cycle, the impacts of human activities on the watershed water cycle under changing environments cannot be ignored. Changes in CO2 concentration caused by human activities are a significant cause of global warming, severely impacting climate change. Simultaneously, land use changes and water conservancy projects resulting from human activities have caused complex changes to the underlying surface of the watershed. The occurrence of extreme weather events under global warming has a severe impact on land use, while changes in land use further exacerbate the occurrence of extreme weather events. Climate change and land use change interact, thus affecting the watershed hydrological cycle and leading to significant changes in watershed runoff. As a crucial factor in the regional water resource cycle, changes in runoff inevitably have a certain impact on the regional socio-economic situation. Considering the combined effects of climate change and human activities, scientifically and accurately predicting runoff changes has significant scientific significance and application value for watershed water resource planning and management, and for the sustainable development of the regional socio-economic situation.

[0045] In the complex water cycle system driven by both climate and human activities, vegetation plays a crucial role as a dynamic interface connecting the atmosphere, soil, and hydrology. Climate change directly affects the physiological and ecological processes of vegetation by altering temperature, precipitation patterns, and CO2 concentration. The CO2 fertilization effect may improve the water use efficiency of some plants in the short term, while warming and drought stress can induce stomatal closure, inhibit photosynthesis, and even lead to vegetation degradation. On the other hand, human-driven land use / cover changes (such as deforestation, agricultural expansion, or urbanization) directly alter the distribution, structure, and function of vegetation. Changes in LAI significantly alter surface albedo, roughness, and evapotranspiration, thereby regulating energy and water vapor exchange between the surface and the atmosphere. When LAI decreases due to drought or deforestation, surface runoff increases while evapotranspiration decreases, potentially exacerbating local hydrological drought and inducing soil erosion. Conversely, increased LAI due to vegetation restoration enhances evapotranspiration and soil moisture consumption, potentially intensifying water resource pressures in arid regions. Therefore, incorporating the dynamic response of vegetation (especially LAI) into the "climate-soil-land use" coupled system for comprehensive analysis is an indispensable part of analyzing and accurately simulating the local drought and flood response mechanism of watersheds.

[0046] Changes in watershed water cycle processes caused by climate change, soil, land use, and vegetation alterations have led to an increase in the frequency, intensity, and scope of natural disasters in watersheds, with more complex attributions. On May 22, 2023, the World Meteorological Organization (WMO) released the latest data showing that between 1970 and 2021, extreme weather events caused 11,778 disasters, resulting in over two million deaths and economic losses of US$4.3 trillion. Drought and floods are the most frequent natural disasters under changing environments. For drought disasters, hydrological drought has a more severe impact on socio-economic development. Furthermore, due to the cumulative and lagging effects of meteorological changes, it is difficult to provide accurate early warnings, leading to more severe losses, such as the severe drought event in the Panama Canal in Central America in 2023. The increasingly frequent extreme weather events pose serious challenges to human society, making it imperative to predict changes in regional hydrological drought characteristics and assess flood risks under changing environments. Coupled multi-model forecasting of drought and flood changes under different future scenarios can not only provide a basic guarantee for the stable development of regional socio-economic conditions, but also enhance regional disaster prevention and mitigation capabilities and provide a scientific basis for watershed water resources planning and management.

[0047] This application focuses on early warning of local drought and flood response in watersheds under changing environments. It constructs a synergistic driving mechanism of climate, land use, vegetation, hydrology, and drought and flood, which can deeply analyze the physical causes of the synergistic enhancement of extreme dry and wet events. The aim is to achieve prediction and forward-looking early warning of future drought and flood risks in watersheds, and to provide core scientific tools for effectively responding to increasingly complex and cross-regional compound disaster risks.

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0049] In one aspect, embodiments of this application provide a method for early warning of local droughts and floods in a watershed.

[0050] In one embodiment, see Figure 1 As shown, the above-mentioned local drought and flood early warning methods for watersheds include: S1. Based on future climate data, natural environment and socio-economic driving factor data, and historical land use data for the target watershed, predict the future spatial distribution of land use. The historical period includes at least two different time points.

[0051] S2. Based on the aforementioned future climate data and the pre-trained leaf area index prediction model, predict the future dynamic sequence of leaf area index.

[0052] S3. Input the above-mentioned land use spatial distribution, the above-mentioned future leaf area index dynamic sequence, and the above-mentioned future climate data into the distributed hydrological model to obtain the actual runoff and the terminal water content of the intermediate soil layer, and calculate the drought index by combining climate observation data, soil effective water holding capacity, and drought index model.

[0053] S4. Issue an early warning based on the preset risk level corresponding to the above drought index.

[0054] In this embodiment, the spatial distribution of future land use is predicted based on future climate data, natural environment and socio-economic driving factors data, and historical land use data of the target watershed, thus explicitly incorporating the alteration of the underlying surface by human activities into the prediction framework in a spatiotemporal manner. By predicting the dynamic sequence of future leaf area index (LAI) based on future climate data and a pre-trained LAI prediction model, the seasonal rhythms and interannual evolution of vegetation can dynamically respond to future climate scenarios. By inputting the spatial distribution of land use, the dynamic sequence of future LAI, and future climate data into a distributed hydrological model, actual runoff and intermediate soil terminal moisture content are obtained. Combined with climate observation data and a drought index model, a drought index is calculated, establishing a complete transmission chain from climate drivers and underlying surface conditions to hydrological response. Finally, early warning is issued based on a preset risk level corresponding to the drought index.

[0055] This method simulates land use evolution, vegetation dynamics, and hydrological processes in a unified framework, making the simulation of runoff formation and drought / flood evolution closer to the real physical mechanisms, thus providing a more reliable quantitative basis for local drought and flood early warning in watersheds.

[0056] Furthermore, in one embodiment, the future climate data in step S1 above comes from the International Climate Model Project, including key meteorological elements such as temperature, precipitation, wind speed, relative humidity, and solar radiation.

[0057] In this embodiment, climate data acquisition adopts a multi-source data fusion acquisition method based on the coupled shared socioeconomic paths - representative concentration paths (SSPs-RCPs). By integrating global climate model outputs, reanalysis data and ground observation data, a climate data system oriented towards the watershed scale is constructed. Specifically, multi-model ensemble data were first obtained from international climate model projects such as CMIP5 (Coupled Model Intercomparison Project Phase 5) and CMIP6 (Coupled Model Intercomparison Project Phase 6). Simulation results of key meteorological elements such as temperature, precipitation, wind speed, relative humidity, and solar radiation under four typical emission scenarios were selected: SSP1-2.6 (Shared Socioeconomic Pathway 1-2.6), SSP2-4.5 (Shared Socioeconomic Pathway 2-4.5), SSP3-7.0 (Shared Socioeconomic Pathway 3-7.0), and SSP5-8.5 (Shared Socioeconomic Pathway 5-8.5).

[0058] To ensure the quality of the acquired data, the quantile mapping method is used to correct the bias in the output of the climate model. By establishing the statistical relationship between the model simulation values ​​and the climate observation data (actual observation values) at different quantiles, the model bias is systematically eliminated.

[0059] To address the application requirements at the watershed scale, a combination of dynamic and statistical downscaling methods is employed. The coarse-resolution output of the aforementioned climate model is downscaled to the spatial resolution required for the watershed using the Weather Research and Forecasting Model (WRF) and statistical transformation models. Regarding precipitation elements, the probability distribution characteristics of precipitation under different seasons and weather conditions are considered to ensure the reliability of extreme precipitation event simulations.

[0060] The processed climate data is stored in a regular raster format, covering the aforementioned historical periods and extending to future multi-scenario periods. It provides multi-model ensemble averaging and individual model outputs to quantify the uncertainty of climate model output data and provide reliable climate driving conditions for local drought and flood early warning in the watershed.

[0061] Furthermore, in one embodiment, see [reference needed]. Figure 2 As shown, in step S1 above, the prediction of future land use spatial distribution involves the following steps: S101. Extract the expansion areas of each land use type from the land use data, calculate the contribution of natural environment and socio-economic driving factors to the expansion of each land use type, and generate a development probability distribution map for each land use type. Land use types include cultivated land, forest land, and grassland.

[0062] S102. Based on future climate data and socioeconomic development assumptions corresponding to shared socioeconomic paths, the total demand for each land use type at multiple preset time nodes is generated through a system dynamics model.

[0063] S103. Based on the development probability distribution map, neighborhood influence coefficient, land use type transfer matrix and total demand, a patch-generated land use simulation model is used to simulate spatial layout and generate the future spatial distribution of land use.

[0064] In this embodiment, a multivariate comprehensive prediction method based on the PLUS (Patch-generating Land Use Simulation) model is used to predict the future spatial distribution of land use. This model integrates historical land use change patterns, natural environmental factors, and socio-economic driving factors to construct a watershed-scale land use change simulation and prediction system. Combining the advantages of traditional cellular automata with a patch-based land use simulation strategy, this model effectively addresses the limitations of traditional models in simulating complex land use change processes, demonstrating significant advantages in handling the competitive relationships between different land use types and simulating the spatial morphological evolution of patches.

[0065] Furthermore, a Land-Expansion Analysis Strategy (LEAS) is employed to uncover historical patterns in land use change. Expansion areas for each land use type are extracted from historical land use data. A random forest algorithm is used to calculate the contribution of various natural environmental and socio-economic driving factors to the expansion of each land use type, generating development probability distribution maps for each land use type. This process not only identifies key driving factors influencing land use change but also quantifies the impact of each driving factor on land use type transformation, providing a reliable data foundation for subsequent predictions of future land use spatial distribution. Further, a CARS (Ca-Markov based on RandomSeeds) model is used for land use change prediction. This model comprehensively considers the development probability, neighborhood influence coefficient, and transfer costs of various land uses, dynamically adjusting the rate of change for each type of land use through an adaptive inertial competition mechanism. The model also incorporates regional constraints, designating permanent basic farmland and ecological red line areas as restricted development zones to ensure that the prediction results meet actual management needs.

[0066] Furthermore, a multi-index comprehensive evaluation method was adopted to verify the consistency between the model output results and real land use data by calculating indices such as FoM (Figure of Merit), Kappa (Kappa Coefficient), and overall accuracy. The validated model parameters were then used to predict the spatial distribution of future land use.

[0067] The neighborhood influence coefficient is determined by analyzing the spatial characteristics of historical land use patterns and is used to quantify the spatial interactions between different land use types. The transition matrix is ​​obtained by analyzing the conversion patterns between land use types in historical periods and is used to set reasonable conversion rules (e.g., arable land cannot be directly converted into water bodies).

[0068] The aforementioned land use data is a historical raster data layer containing the described land use types. The aforementioned natural environment and socioeconomic drivers include topographic factors, soil properties, climate factors, and socioeconomic factors. Topographic factors include elevation, slope, and aspect; soil properties include soil type and organic matter content; climate factors include distance from rivers, average annual precipitation, and average annual temperature; and socioeconomic factors include population density, spatial distribution of GDP, transportation network, and settlement distribution. All data are standardized to the same spatial reference and resolution.

[0069] Furthermore, by combining SSPs-RCPs climate scenarios and socio-economic development paths, different land use demand constraints are set. Based on future climate data and socio-economic development assumptions, the total demand for various land use types in different future periods is generated through a system dynamics model (SD). These demands serve as the overall control conditions of the aforementioned PLUS model and are combined with spatial allocation rules to generate the future spatial distribution of land use.

[0070] Furthermore, in one embodiment, see [reference needed]. Figure 3 As shown, in step S2 above, the dynamic sequence of future leaf area index is predicted as follows: S201. Input multiple sets of future climate data into the leaf area index prediction model to obtain multiple preliminary leaf area index prediction sequences.

[0071] S202. The multiple preliminary leaf area index prediction sequences mentioned above are integrated using the multi-model ensemble averaging method to obtain the future dynamic leaf area index sequence.

[0072] This embodiment employs a deep learning ensemble prediction method based on the CMIP6 climate model, comprehensively utilizing three independent satellite remote sensing LAI datasets: GLOB MAP (GLOBal MAPping project), GIMMS LAI3g (Global Inventory Modeling and Mapping Studies LAI 3rd generation), and GLASS LAI (Global LAnd Surface Satellite). Cross-validation and consistency analysis of the datasets ensure the reliability and continuity of historical LAI data. To more accurately predict future leaf area indices, 19 different Earth System Models (ESMs) are selected, and a multi-model ensemble averaging method is used to integrate the advantages of each model, effectively reducing the uncertainty and random error of single-model predictions.

[0073] In this embodiment, an Attention-Enhanced Long Short-Term Memory Network (AELSTM) model is used as the core prediction algorithm. This model can effectively capture the long-term dependencies of LAI time series and uses an attention mechanism to weight and focus on the impact of key climate factors. The AELSTM model is trained using historical climate observation data and corresponding LAI observation data. Compared with the traditional LSTM model, the AELSTM model adds an attention enhancement layer. The introduction of the attention mechanism is achieved through the following formula: (1), (2), (3), in, This represents the eigenvectors calculated in the first weighted calculation. The weight matrix representing the attention mechanism. This represents the input to the attention mechanism. This indicates a deviation in the attention mechanism; Indicates the first t The attention weights assigned to the hidden states at each time step Indicates the current time step t The eigenvectors obtained after transforming the hidden states using formula (1) are: Indicates time step k The corresponding feature vector, This represents the learnable global query vector in the attention mechanism. T This represents the total number of time steps in the sequence; Represents the context vector. Indicates that in generating the first t When the output is the nth, the nth in the input sequence i Attention weights for hidden states at each time step.

[0074] The CMIP6 data above were integrated using a Bayesian model averaging method, as shown in the following formula: (4), in, This represents the final posterior prediction distribution. express LAI The observed values, express LAI The predicted value, D Represents the training dataset. Indicates the first k A CMIP6 model Indicating in the training data D Next, the k The posterior weights of a CMIP6 model.

[0075] In the training process of the above AELSTM model, a composite loss function combining mean square error and phenological constraints is used, as shown in the following formula: (5), in, L This represents the total loss value. N This represents the total number of training samples. Indicates the first iLAI prediction values ​​for each sample Indicates the first i The actual observed LAI values ​​for each sample. The trade-off parameter indicating the start of the growing season constraint, All of these represent trade-off parameters for the constraint of the end of the growing season. S This represents the total number of samples during the growing season. Indicates the first j Predicted growing season start date for each sample Indicates the first j The actual start date of the growing season for each sample. Indicates the first j Predicted end date of the growing season for each sample. Indicates the first j The predicted end date of the growing season for each sample.

[0076] The aforementioned AELSTM model incorporates an attention mechanism, enabling it to dynamically identify and enhance meteorological factors during the prediction process. This allows for a more accurate capture of the complex nonlinear temporal relationship between meteorological variables and LAI (Local Area Index). Furthermore, the AELSTM model employs a unique pixel-by-pixel modeling strategy, rather than holistic biome modeling. This effectively avoids interference caused by heterogeneity within vegetation types and the common mixed-pixel problem in low-to-medium resolution remote sensing images, significantly improving the model's prediction accuracy and applicability in spatially heterogeneous regions. Simultaneously, the AELSTM model can seamlessly integrate with land surface process models to accurately predict downstream variables such as total primary productivity (TPMP), and maintain high-precision LAI predictions under various future climate scenarios. This provides a powerful data-driven solution to address the bottleneck of traditional diagnostic models' inability to predict future vegetation dynamics.

[0077] Furthermore, in one embodiment, the method for calculating the actual runoff in step S3 above is as follows: Direct runoff is calculated based on the soil moisture content of the intermediate layer at the previous moment, the precipitation data from the aforementioned future climate data, and the saturated soil moisture content and maximum infiltration capacity of the intermediate layer in the aforementioned land use spatial distribution. The soil moisture content of the intermediate layer calculated at the previous moment is the terminal soil moisture content calculated at the previous moment, and this value is dynamically transferred during model iteration.

[0078] The infiltration rate is calculated based on the soil moisture content of the intermediate layer at the previous moment, as well as the soil residual moisture, saturated permeability coefficient, soil pore size distribution index, and saturated water content of the intermediate layer in the above-mentioned land use spatial distribution. Among them, the currently calculated soil moisture content of the intermediate layer is the real-time soil moisture content calculated at the current moment, which is dynamically updated during the water balance calculation process.

[0079] The sum of the above-mentioned direct runoff and the above-mentioned infiltration is taken as the actual runoff.

[0080] The method for calculating the terminal moisture content of the intermediate soil layer is as follows: Based on the aforementioned future climate data, the aforementioned future leaf area index dynamic sequence, and the aforementioned future land use spatial distribution, calculate canopy evaporation, transpiration, and bare soil evaporation.

[0081] The actual evapotranspiration is calculated based on the above-mentioned canopy evaporation, transpiration, bare soil evaporation, and the vegetation cover ratio and bare soil cover ratio in the spatial distribution of future land use.

[0082] The terminal moisture content of the intermediate soil layer is calculated based on the moisture content of the intermediate soil layer at the previous moment, the aforementioned precipitation, the aforementioned direct runoff, the aforementioned infiltration, and the aforementioned actual evapotranspiration.

[0083] In this embodiment, the distributed hydrological model uses the VIC (Variable Infiltration Capacity) model to calculate the actual runoff and the terminal moisture content of the intermediate soil layer. The distributed hydrological model is a spatially explicit model based on physical processes. Its core advantage lies in its ability to simulate the spatiotemporal changes of the hydrological cycle within a watershed. It emphasizes that actual hydrological processes are not determined by a single factor, but are driven by multiple factors such as climate, vegetation, and land use. The VIC model integrates climate data, soil data, land use data, and LAI data to construct a collaborative driving framework, enabling more accurate simulation of key hydrological processes such as runoff, runoff, and evapotranspiration. This model is suitable for applications such as water resource management and drought / flood risk assessment under the influence of climate change and human activities. The collaborative driving mechanism emphasizes that climate provides energy and water input, soil data reflects the underlying surface condition, land use data defines surface characteristics, and LAI reflects vegetation dynamics. These three factors interact and influence each other through parameterization and process coupling.

[0084] The water balance in the VIC model takes into account three types of evaporation: canopy evaporation, transpiration, and bare soil evaporation.

[0085] The calculation steps for the above canopy evaporation are as follows: Based on the precipitation and future leaf area index dynamic series in future climate data, the canopy water retention capacity is determined as the upper limit of canopy evaporation. Based on the temperature, solar radiation, wind speed and relative humidity in future climate data, the Penman-Monteith equation is used and the canopy surface impedance is set as the wet canopy surface impedance to calculate the potential canopy evaporation. The smaller value between the canopy water retention capacity and the potential canopy evaporation is taken as the actual canopy evaporation.

[0086] The calculation steps for the above transpiration are as follows: Based on the dynamic sequences of future climate data (temperature, solar radiation, wind speed, relative humidity, and future leaf area index), and combined with the dynamically updated intermediate soil moisture content from the VIC model, stomatal impedance was determined. A smaller stomatal impedance value was used when the intermediate soil moisture content was sufficient, and a larger value was used when the intermediate soil moisture content was insufficient, to characterize the inhibitory effect of soil moisture stress on vegetation transpiration. Based on the future climate data (temperature, solar radiation, wind speed, and relative humidity), the Penman-Monteith equation was used, and the stomatal impedance divided by the leaf area index was substituted into the equation as the canopy surface impedance to calculate potential transpiration. The potential transpiration was then reduced according to the intermediate soil moisture content to determine the actual transpiration. When the intermediate soil moisture content was sufficient, the actual transpiration equaled the potential transpiration; when the intermediate soil moisture content was insufficient to support the potential transpiration, the actual transpiration was reduced proportionally based on soil moisture availability.

[0087] The calculation steps for the above bare soil evaporation are as follows: Based on future climate data including temperature, solar radiation, wind speed, and relative humidity, and combined with dynamically updated surface soil moisture content from the VIC model, soil impedance is determined. A smaller soil impedance value is used when surface soil moisture content is sufficient, and a larger value is used when surface soil moisture content is insufficient, to characterize the inhibitory effect of soil drying on bare soil evaporation. Based on the aforementioned future climate data (temperature, solar radiation, wind speed, and relative humidity), the Penman-Monteith equation is used, and the soil impedance is substituted as the surface impedance into the equation to calculate potential bare soil evaporation. The potential bare soil evaporation is then reduced by the surface soil moisture content to determine the actual bare soil evaporation. When surface soil moisture content is sufficient, the actual bare soil evaporation equals the potential bare soil evaporation; when surface soil moisture content is insufficient to support the potential bare soil evaporation, the actual bare soil evaporation is reduced proportionally based on the availability of surface soil moisture.

[0088] The actual evapotranspiration mentioned above is the sum of three evapotranspiration types: canopy evaporation, transpiration, and bare soil evaporation. The weights are the percentage of the area corresponding to each land use type, calculated using the following formula: (6), in, represents evapotranspiration, Indicates the first n Vegetation coverage ratio of different vegetation types Indicates the first n Canopy evaporation rate of vegetation type Indicates the first n Transpiration rate of vegetation type Indicates the percentage of bare soil coverage. This indicates the amount of bare soil evaporation.

[0089] The formula for calculating the terminal moisture content of the intermediate soil layer is as follows: (7), in, This indicates the terminal moisture content of the intermediate soil layer. This represents the final moisture content of the intermediate soil layer calculated at the previous moment. Indicates precipitation. Indicates direct runoff. This indicates the amount of soil that infiltrates from the second to the third layer due to gravity within a given time period. This indicates the evaporation rate of the intermediate soil layer. Indicates the length of the time period.

[0090] The above The calculation formula is: (8), in, Represents the saturated permeability coefficient. This indicates the current moisture content of the intermediate soil layer. This indicates the saturated water content of the intermediate soil layer. Indicates residual soil moisture. This represents the soil pore size distribution index.

[0091] The above The calculation formula is: (9).

[0092] The formula for calculating the above-mentioned direct runoff is as follows: (10) in, Indicates initial infiltration capacity. This indicates the maximum infiltration capacity.

[0093] Furthermore, in one embodiment, the step of calculating the drought index in step S3 above is as follows: Based on the above climate observation data and soil effective water holding capacity, potential evapotranspiration, potential recharge, potential runoff, and potential loss are calculated using a two-layer soil model.

[0094] Based on the aforementioned potential evapotranspiration, potential replenishment, potential runoff, and potential loss, calculate the climate-appropriate precipitation.

[0095] Based on the above precipitation, the above suitable precipitation, the above actual runoff, the above actual evapotranspiration, and the above intermediate soil terminal moisture content, calculate the current month's moisture anomaly index.

[0096] The water anomaly index sequences of the above historical periods were divided into wet periods and dry periods according to positive and negative numbers, respectively. Linear regression models were established for each period, and the corresponding duration factors were obtained by fitting the data.

[0097] The drought index for the current month is calculated using a recursive formula based on the current month's moisture anomaly index, the previous month's drought index, and the aforementioned duration factor.

[0098] The above climate observation data includes temperature, precipitation, wind speed, and relative humidity.

[0099] In this embodiment, the scPDSI (self-calibrating Palmer Drought Severity Index) is used to calculate the drought index. Drought is one of the most destructive natural disasters globally, posing a persistent threat to agriculture, water resources, ecosystems, and socio-economic development. To quantify the intensity, duration, and spatial extent of drought, various drought index calculation methods have been developed, among which the PDSI is one of the most widely used and far-reaching drought indices in history. The PDSI is a soil moisture algorithm index based on the water balance principle, comprehensively considering factors such as precipitation, temperature, and local available water holding capacity. Through a simple two-layer soil model, it simulates the cycle of water between evaporation, runoff, soil infiltration, and recharge. Its output is a standardized numerical value that can describe the dryness and wetness conditions at a regional scale. scPDSI does not completely abandon the traditional PDSI framework; rather, it retains the classic water balance calculation module of PDSI, but reforms the process of climate coefficient calibration and probability distribution standardization.

[0100] The potential evapotranspiration, potential recharge, potential runoff, and potential loss are the water balance components in the two-layer soil model. The potential evapotranspiration is calculated based on climate observation data, while the potential recharge, potential runoff, and potential loss are derived from the two-layer soil model in combination with the effective water holding capacity of the soil.

[0101] The formula for calculating the suitable precipitation for the above climate is as follows: (11), in, This indicates the amount of precipitation suitable for the climate. This represents the evapotranspiration weighting coefficient. represents potential evapotranspiration, Indicates the supply weighting coefficient. Indicates potential supply volume. This represents the runoff weighting coefficient. Indicates potential runoff. This represents the loss weighting coefficient. This indicates the potential amount of loss.

[0102] Calculate the water deficit based on the above precipitation and the above-mentioned suitable precipitation for climate. d The formula is as follows: (12).

[0103] Further standardization processes are introduced to correct the d-sequences in order to improve the correlation across time and space comparability.

[0104] Based on the historical statistical characteristics of the above-mentioned actual runoff, actual evapotranspiration, and terminal moisture content of the intermediate soil layer, the climate characteristic coefficient is calculated using the following formula: (13) in, Indicates the climate characteristic coefficient. Indicates time correction weight. This represents the spatial correction weights.

[0105] The calculation formula is: (14) in, represents the average potential evapotranspiration, Indicates the average supply amount. Indicates average runoff. This represents the average precipitation. Indicates the average loss amount. Indicates the first j Average monthly water deficit. and Historical statistical values ​​of actual evapotranspiration and actual runoff output from VIC can be used as a substitute, making It is more in line with the actual characteristics of the watershed.

[0106] Considering that the main drawback of PDSI is its poor spatial performance, this embodiment focuses on... Make corrections. The calculation formula is: (15) in, This indicates the PDSI value under extremely dry conditions. This indicates the PDSI value under extremely humid conditions.

[0107] The above-mentioned moisture anomaly index for the current month Z The calculation formula is: (16).

[0108] Introducing improved spatial correction weights This solves the core flaw of traditional PDSI models, which have poor applicability in different climate zones, making... K Instead of relying solely on constants that consider climate characteristics, this mechanism dynamically calibrates based on historical climate data for each location, thus customizing a benchmark for different regions. By eliminating systematic biases in regional climate backgrounds, this mechanism effectively avoids misjudgments of drought severity in arid areas while ensuring accurate identification of drought signals in humid areas. This results in a drought index with uniform physical meaning across all regions. It not only provides a reliable and consistent universal template for drought monitoring, precise comparison, and coordinated emergency response across watersheds, regions, and even nationwide, but also significantly enhances the objectivity of drought early warning and the effectiveness of decision support.

[0109] In the original PDSI, the duration factors p and q were set to 0.897 and 1 / 3, respectively. p controls the decay rate of the previous drought index, and q controls the contribution of the current moisture anomaly to the drought index. This method is not suitable for all climate zones. Therefore, to identify the most suitable p and q values ​​for each watershed from historical climate data, the historical data... Z The sequence is divided into dry and wet periods according to its positive or negative sign. Z<0 It is a dry period. Z>0 For the wet period), linear regression models were established respectively.

[0110] The linear regression model equation for the above-mentioned drought period is as follows: (17) in, This indicates the current month's moisture abnormality index. and Both represent the duration factor of the drought period. This indicates the moisture abnormality index for the previous month. This indicates the water deficit for the current month. This represents the regression residual.

[0111] The linear regression model equation for the above-mentioned wet period is: (18) in, and Both represent the duration factor of the wet period.

[0112] The least squares method was used to fit the above linear regression model equations (17) and (18) to obtain the corresponding duration factor.p and q .

[0113] Traditional PDSI models use fixed parameters p and q applicable to specific regions, leading to systematic biases when applied to other climate zones. This embodiment utilizes a self-calibration mechanism, taking advantage of the data from each station... Z The data from dry and wet periods were separated and linear regression models were established for each period. The least squares method was used to fit the optimal p and q values. This method enables the model to accurately reflect the persistence and mitigation characteristics of drought events in different climatic zones, significantly improving the regional adaptability and temporal accuracy of drought index calculation, and providing a more reliable decision-making basis for water resource management under different climatic conditions.

[0114] The formula for calculating the drought index for the current month is as follows: (19) in, This indicates the drought index for the previous month. This indicates the current month's moisture abnormality index.

[0115] In this embodiment, the VIC model simulates the watershed hydrological cycle process, including evaporation, runoff and soil moisture dynamics, while the scPDSI model calculates the drought index based on the soil moisture and climate data output by VIC, and assesses drought and wet conditions in real time.

[0116] Furthermore, in one embodiment, in step S4 above, an early warning is issued based on the preset risk level corresponding to the drought index, as follows: Determine the risk level corresponding to the above drought index and generate the corresponding early warning information.

[0117] The risk levels mentioned above are set based on the drought index, including mild drought, moderate drought, severe drought, extreme drought, mild humidity, moderate humidity, severe humidity, and extreme humidity.

[0118] In this embodiment, the early warning risk level is set according to the drought index, as shown in Table 1 below. The drought index has the same range as the traditional PDSI value, mainly varying between -4.0 and 4.0. The threshold setting in Table 1 takes into account the watershed specificity.

[0119] Table 1. PDSI Index Dryness / Wetness Levels

[0120] The drought index is used to determine the risk level, generate corresponding early warning information (including risk level, scope of impact and response recommendations), and support data sharing to achieve collaborative decision-making.

[0121] Secondly, embodiments of this application also provide a localized drought and flood early warning device for a watershed.

[0122] In one embodiment, see Figure 4 As shown, the aforementioned localized drought and flood early warning device for the watershed includes a land use module, a leaf area index module, a drought index module, and an early warning module, specifically: The land use module is used to predict the future spatial distribution of land use based on future climate data, natural environment and socio-economic driving factors data, and historical land use data of the target watershed.

[0123] The leaf area index module is used to predict the future dynamic sequence of leaf area index based on the aforementioned future climate data and a pre-trained leaf area index prediction model.

[0124] The drought index module is used to input the above-mentioned land use spatial distribution, the above-mentioned future leaf area index dynamic sequence, and the above-mentioned future climate data into the distributed hydrological model to obtain the actual runoff and the terminal water content of the intermediate soil layer, and to calculate the drought index by combining climate observation data, soil effective water holding capacity, and the drought index model.

[0125] The early warning module is used to issue early warnings based on the preset risk level corresponding to the aforementioned drought index.

[0126] This application provides a watershed local drought and flood early warning technology based on the coupling of multiple factors such as climate, land use and leaf area index. It constructs a synergistic driving mechanism of climate-land use-vegetation-hydrology-drought and flood, realizes dynamic interaction and synergistic simulation of multiple factors, and can effectively improve the accuracy of watershed local drought and flood early warning.

[0127] Secondly, it integrates advanced data processing and machine learning algorithms, including climate data downscaling based on SSPs-RCPs, land use spatial distribution prediction using the PLUS model, and LAI prediction using the AELSTM network, forming a complete solution that ensures reliability throughout the entire process from data input to result output. Simultaneously, it supports parallel computing, significantly improving computational efficiency and providing technical support for operational applications.

[0128] Furthermore, it achieves high spatiotemporal resolution simulation capabilities, covering multi-scale requirements from daily to annual time scales. This refined simulation capability enables it to accurately capture local drought and flood response characteristics within watersheds, providing strong support for refined water resource management. It also offers multi-scenario analysis functions, supporting simulation and prediction of various future scenarios from SSP1-2.6 to SSP5-8.5, and can quantify uncertainties at each stage, providing a scientific basis for decision-making.

[0129] Furthermore, this application employs a VIC-scPDSI coupled model, overcoming the inherent shortcomings of traditional PDSI indexes, such as fixed parameters and poor spatial comparability. Coupling leverages the watershed's own historical climate data and novel spatial correction weighting coefficients to calibrate key parameters, making drought assessment results more adaptable to unique local climate conditions. This allows the method to be flexibly applied to watersheds with different climate characteristics and ensures comparability of assessment results across different regions, significantly expanding the system's application scope.

[0130] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0131] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0132] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0133] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0134] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0136] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for early warning of local drought and flood in a watershed, characterized in that, The method includes: Based on future climate data, natural environment and socio-economic driving factor data of the target watershed, and land use data from historical periods, predict the spatial distribution of future land use; the historical periods include at least two different time points. Based on the aforementioned future climate data and a pre-trained leaf area index prediction model, predict the future dynamic sequence of leaf area index; The land use spatial distribution, the future leaf area index dynamic sequence, and the future climate data are input into a distributed hydrological model to obtain the actual runoff and the terminal water content of the intermediate soil layer. The drought index is then calculated by combining climate observation data, soil effective water holding capacity, and the drought index model. Early warnings are issued based on the preset risk level corresponding to the drought index.

2. The method for early warning of local drought and flood in a watershed as described in claim 1, characterized in that, The future climate data mentioned comes from the International Climate Modeling Project and includes temperature, precipitation, wind speed, and relative humidity.

3. The method for early warning of local drought and flood in a watershed as described in claim 1, characterized in that, The prediction of future land use spatial distribution includes: The expansion areas of each land use type are extracted from the land use data, and the contribution of the natural environment and socio-economic driving factors to the expansion of each land use type is calculated to generate a development probability distribution map of each land use type; the land use types include cultivated land, forest land and grassland. Based on the future climate data and the socio-economic development assumptions corresponding to the shared socio-economic path, the total demand for each land use type at multiple preset time nodes in the future is generated through a system dynamics model. Based on the development probability distribution map, neighborhood influence coefficient, land use type transfer matrix, and total demand, a patch-based land use simulation model is used to simulate spatial layout and generate the future spatial distribution of land use.

4. The method for early warning of local drought and flood in a watershed as described in claim 1 or 3, characterized in that, The land use data is a historical raster data layer containing the land use type; The natural environment and socio-economic driving factors include topographic factors, soil properties, climate factors, and socio-economic factors. The topographic factors include elevation, slope, and aspect; the soil properties include soil type and organic matter content; the climate factors include distance from the river, average annual precipitation, and average annual temperature; and the socioeconomic factors include population density, spatial distribution of GDP, transportation network, and distribution of settlements.

5. The method for early warning of local drought and flood in a watershed as described in claim 1, characterized in that, The predicted future leaf area index dynamic sequence includes: Multiple sets of future climate data are input into the leaf area index prediction model to obtain multiple preliminary leaf area index prediction sequences. The multiple preliminary leaf area index prediction sequences are integrated using a multi-model ensemble averaging method to obtain a dynamic sequence of future leaf area index.

6. The method for early warning of local drought and flood in a watershed as described in claim 1, characterized in that, The method for calculating the actual runoff includes: The direct runoff is calculated based on the water content of the intermediate soil layer at the previous moment, the precipitation in the future climate data, and the saturated water content and maximum infiltration capacity of the intermediate soil layer in the spatial distribution of land use. The infiltration rate is calculated based on the water content of the intermediate soil layer at the previous moment, as well as the residual soil moisture, saturated permeability coefficient, soil pore size distribution index, and saturated water content of the intermediate soil layer in the land use spatial distribution. The sum of the direct runoff and the infiltration is taken as the actual runoff.

7. The method for early warning of local drought and flood in a watershed as described in claim 6, characterized in that, The method for calculating the terminal moisture content of the intermediate soil layer includes: Based on the future climate data, the future leaf area index dynamic sequence, and the future land use spatial distribution, calculate canopy evaporation, transpiration, and bare soil evaporation. The actual evapotranspiration is calculated based on the canopy evaporation, the transpiration, the bare soil evaporation, and the vegetation cover ratio and bare soil cover ratio in the future land use spatial distribution. The terminal moisture content of the intermediate soil layer is calculated based on the moisture content of the intermediate soil layer at the previous moment, the precipitation, the direct runoff, the infiltration, and the actual evapotranspiration.

8. The method for early warning of local drought and flood in a watershed as described in claim 7, characterized in that, The calculation of the drought index includes: Based on the climate observation data and soil effective water holding capacity, potential evapotranspiration, potential recharge, potential runoff, and potential loss are calculated using a two-layer soil model. The climate-suitable precipitation is calculated based on the potential evapotranspiration, the potential recharge, the potential runoff, and the potential loss. The moisture anomaly index for the current month is calculated based on the precipitation, the climate-suitable precipitation, the actual runoff, the actual evapotranspiration, and the terminal moisture content of the intermediate soil layer. The historical water anomaly index sequence was divided into wet and dry periods according to positive and negative numbers, respectively, and linear regression models were established for each period to obtain the corresponding duration factor. The drought index for the current month is calculated using a recursive formula based on the current month's moisture anomaly index, the previous month's drought index, and the duration factor. The climate observation data includes temperature, precipitation, wind speed, and relative humidity.

9. The method for early warning of local drought and flood in a watershed as described in claim 1, characterized in that, The step of issuing an early warning based on a preset risk level corresponding to the drought index includes: Determine the risk level corresponding to the drought index and generate corresponding early warning information; The risk levels are set based on the drought index and include mild drought, moderate drought, severe drought, extreme drought, mild humidity, moderate humidity, severe humidity, and extreme humidity.

10. A localized drought and flood early warning device for a watershed, characterized in that, The device includes: The land use module is used to predict the future spatial distribution of land use based on future climate data, natural environment and socio-economic driving factor data, and historical land use data of the target watershed. The leaf area index module is used to predict the future dynamic sequence of leaf area index based on the future climate data and a pre-trained leaf area index prediction model. The drought index module is used to input the land use spatial distribution, the future leaf area index dynamic sequence, and the future climate data into the distributed hydrological model to obtain the actual runoff and the terminal water content of the intermediate soil layer, and to calculate the drought index by combining climate observation data, soil effective water holding capacity, and the drought index model. The early warning module is used to issue early warnings based on the preset risk level corresponding to the drought index.

Images