A method for hydrological prediction of a reservoir with insufficient data based on remote sensing data

Abstract

The application provides a kind of based on remote sensing data's lack of information reservoir hydrology prediction method, comprising: model construction, lack of information reservoir water level-area-storage variable relationship identification, reservoir regulation rule training and parameterization modeling, and basin storage variable process simulation and reservoir regulation coupling etc. steps.This method combines remote sensing inversion technology, reservoir regulation behavior analysis and hydrological model construction, realizes the integration of remote sensing information and hydrological regulation simulation.This application solves the problem of prediction deviation caused by lack of regulation data in traditional methods, has the advantages of low data dependence, strong adaptability, high precision, good scalability, especially suitable for hydrological simulation and basin management in typical lack of information area, provides reliable technical support for water resources regulation and flood warning.

Description

A Data-Deficient Reservoir Hydrological Forecasting Method Based on Remote Sensing Data Technical Field

[0001] This invention belongs to the technical field of remote sensing technology and hydrological forecasting, and specifically relates to a method for hydrological forecasting of reservoirs with insufficient data based on remote sensing data. It is specifically applied to hydrological simulation, watershed runoff prediction and reservoir scheduling decision-making scenarios in areas with no or insufficient data. Background Technology

[0002] In the field of hydrological forecasting for watersheds regulated by reservoirs, existing technologies consistently face the core technical bottleneck of "lack of data making it difficult to quantify reservoir operation behavior," resulting in the inability to meet practical needs in simulating and forecasting watershed hydrological processes. Existing forecasting schemes and their shortcomings are as follows:

[0003] The scheduling method based on historical operation records: This method uses continuously measured operational data such as water level, outflow, and storage capacity of the reservoir as its core. By constructing a scheduling module coupled with a hydrological model, it achieves direct simulation of the reservoir's regulation behavior. Its core reliance is "complete and systematic measured scheduling data." However, due to the lack of long-term continuous measured data for some reservoirs, this method cannot be implemented in real-world scenarios.

[0004] Fixed parameter method: As an alternative to scenarios where no measured data is available, this method simulates reservoir operation by pre-setting a constant outflow coefficient or empirical scheduling rules. Its essential drawback lies in "ignoring the time-varying nature of hydrological processes"—it does not consider the impact of dynamic changes in meteorological and hydrological factors such as rainfall intensity, inflow, and evaporation on scheduling decisions. For example, after heavy rainfall during the flood season, fixed parameters cannot reflect the actual demand for emergency flood discharge from the reservoir, and during the dry season, they will excessively underestimate the reservoir's water storage and supply capacity, resulting in generally large errors in downstream runoff simulation.

[0005] Physical model method: This method constructs a mechanism-based scheduling model based on physical parameters such as reservoir capacity-water level curves and spillway dimensions obtained during the reservoir design phase, which theoretically possesses strong interpretability. However, two key problems exist in practical applications: first, some reservoir design data are missing or deviate significantly from the current situation; second, actual scheduling methods often deviate from design rules, causing the model to fail to reflect the true control mechanism and limiting forecast accuracy.

[0006] The analogy transplantation method involves selecting existing reservoirs with similar storage capacity and regulation types, and then transferring their scheduling experience or parameters to the target reservoir. While simple to operate, this method's core flaw is that it "ignores the differences in watershed characteristics"—failing to consider the fundamental differences between the target and analog reservoirs in terms of catchment area, climate conditions, and water demand. For example, transplanting scheduling parameters from reservoirs in humid regions to those in arid regions results in significant errors in dry season flow simulations, making it unsuitable for unconventional hydrological conditions.

[0007] The common defects of the aforementioned existing technologies ultimately lead to two major technical problems that cannot be solved: First, the quantitative analysis of the regulation effect fails—because the dynamic scheduling behavior of reservoirs lacking data cannot be captured, hydrological models cannot simulate the reservoir's role in intercepting and reducing flood peaks, resulting in amplified errors in downstream flood peak flow and a lag in peak appearance time; Second, the adaptability to scenarios without data is poor—existing solutions all rely on "measured data" or "subjective analogy assumptions," making it impossible to achieve reliable hydrological forecasting in typical areas such as plateaus, border regions, and arid areas where basic data is lacking, and thus failing to meet the actual decision-making needs of real-time reservoir scheduling and flood warning. Summary of the Invention

[0008] This invention aims to address the inaccuracy of existing hydrological simulation and forecasting methods in situations lacking reservoir operation data. It provides a data-scarce reservoir hydrological forecasting method based on remote sensing data. This method introduces remote sensing data to invert water area and water level, and combines this with empirical volume calculation methods to construct a water storage change sequence. Based on the random forest method, it establishes the relationship between meteorological driving factors and historical water storage change characteristics, training a parameterized reservoir operation behavior model. By embedding this model into a traditional hydrological model as its operation module, the accuracy of water storage variable simulation and forecasting is effectively improved, enabling the reconstruction and prediction of data-scarce reservoir operation behavior, and expanding the application boundaries of remote sensing technology in the field of hydrological simulation.

[0009] The objective of this invention is achieved as follows:

[0010] This invention provides a method for forecasting hydrological data of reservoirs with insufficient data based on remote sensing data, comprising the following steps:

[0011] Step 1, Model Building:

[0012] Based on the actual conditions of the study area, a distributed or semi-distributed hydrological model suitable for the watershed regulated by the reservoir is selected. Meteorological data of the study area are collected. The watershed boundary and river network structure are delineated by combining a high-precision digital elevation model. The location of the control section and the reservoir is clarified. The model is calibrated and verified by measured data, thereby constructing a basic framework for hydrological simulation.

[0013] Step 2, Identify the water level-area-storage variable characteristics of reservoirs with no data:

[0014] Identify reservoirs within the study area lacking data and simultaneously acquire their water level-area-storage variation characteristics, specifically including:

[0015] S21, Reservoir water body extraction and area sequence construction: Water body extraction and temporal analysis were performed on the identified reservoirs without data using multi-source remote sensing image data. The normalized difference water index (NDWI) was used to extract the reservoir boundaries. An NDWI threshold was set to determine water body pixels and a minimum area threshold was set. Isolated patches were removed by combining morphological operations. The daily reservoir area sequence was obtained by sequence processing of multi-temporal images.

[0016] S22, Reservoir Regulation Characteristics Analysis: By comprehensively utilizing digital elevation models, geographic information systems, and daily reservoir area sequences, the reservoir regulation characteristics are analyzed, the regulation type is determined, and the typical regulation range of the reservoir is obtained.

[0017] S23, Reservoir Water Level Inversion and Unified Correction: Reservoir water level is inverted using altimeter satellite data, effective trajectory points are extracted and outliers are removed, water level data from different satellite sources are converted to a unified benchmark, and a reservoir water level-area relationship is established.

[0018] S24, Calculation of reservoir water storage sequence: Based on the reservoir area sequence and water level changes obtained above, calculate the reservoir water storage change, and then construct the water storage change time series of reservoirs without data.

[0019] Step 3, Training and parametric modeling of reservoir scheduling rules:

[0020] Based on remote sensing image data, the water level-area function relationship of the reservoir is constructed, and the water level and storage volume sequence for each 16-day period is obtained by inversion. The meteorological dataset of the study area is introduced. The random forest regression method is used, with meteorological factors as input variables, and the storage variable of the reservoir under the scheduling action is output. The optimal model structure and parameter combination are screened through residual test and cross-validation. The storage variable for each 16-day period is used as a correction factor and superimposed on the flow forecast results of the original hydrological model, thereby constructing the parameterization module of the scheduling rules.

[0021] Step 4, Coupling the watershed storage process simulation with reservoir scheduling:

[0022] The scheduling rule parameterization module established in step 3 is embedded into the hydrological model structure. During the model operation, the scheduling module is called in real time according to the meteorological forcing data at the current moment, and the corresponding storage variables are output to realize the dynamic simulation of the watershed hydrological process by the reservoir regulation behavior.

[0023] Furthermore, in step 1, the distributed or semi-distributed hydrological model includes the Xin'anjiang model and the HBV model.

[0024] Furthermore, in step 2, the multi-source remote sensing image data includes Jason1-3 and Landsat5-8 remote sensing image data, and the altimeter satellite data includes ICESat, Sentinel-3, or CryoSat-2 satellite data.

[0025] Furthermore, in step 2, the formula for calculating the Normalized Difference Water Index (NDWI) is as follows:

[0026]

[0027] In the formula: ρ Green and ρ NIR These represent the green band and mid-infrared band in remote sensing images, respectively.

[0028] Furthermore, in step 2, the change in reservoir water storage is calculated using the following formula:

[0029]

[0030] In the formula: For the change in reservoir water volume over a specified period of time, and Let be the area of ​​the reservoir on day t and day t-1, respectively. and The elevations of the reservoir are on day t and day t-1, respectively.

[0031] Furthermore, in step 2, when performing sequence processing on multi-temporal images, images from the relatively stable hydrological period within the same year are given priority. If images for that period are lacking, images from the entire year are used to supplement the data. The influence of cloud shadows is reduced by median synthesis, and finally, the daily area sequence S(t) of the reservoir without data is obtained.

[0032] Furthermore, in step 3, the 16-day storage variable is the difference between the actual water storage of the reservoir and the water storage under natural conditions. Its calculation and output time scale is consistent with the remote sensing inversion sequence and meteorological input data, and is set to a 16-day scale.

[0033] Furthermore, in step 3, the meteorological dataset includes key meteorological factors such as 16-day precipitation, temperature, shortwave radiation, longwave radiation, wind speed, and specific humidity.

[0034] Furthermore, in step 4, the meteorological forcing data includes air temperature, air pressure, specific humidity, wind speed, downward shortwave radiation flux, downward longwave radiation flux, and precipitation.

[0035] The "Remote Sensing Data-Based Reservoir Hydrological Forecasting Method" proposed in this invention addresses the core pain points of existing technologies in data-scarce reservoir watershed hydrological forecasting: "difficulty in quantifying the control effect and poor adaptability to scenarios without data." It achieves a technological breakthrough through deep coupling of remote sensing technology and hydrological models, with the following specific beneficial effects:

[0036] 1. Significantly improves the accuracy of hydrological simulation in data-scarce scenarios, solving the problem of "failure in quantifying regulation effects": Existing technologies often lead to large errors in downstream runoff simulation and delayed flood peaks due to the inability to obtain dynamic scheduling data of reservoirs with limited data. This invention uses multi-source remote sensing data to invert the reservoir's water level-area-storage variable sequence and combines it with the random forest algorithm to construct a parameterized scheduling model. This accurately embeds the actual regulation behavior of the reservoir into the hydrological model, effectively capturing the reservoir's role in flood control and peak reduction, as well as the water storage and supply patterns during the dry season. It completely solves the technical bottleneck of the difficulty in quantifying the scheduling effects of reservoirs with limited data, providing accurate support for basin flood early warning and runoff forecasting.

[0037] 2. Overcoming the limitation of "data dependence" and improving the method's adaptability and universality: Existing technologies either rely on measured scheduling data or subjective analogy assumptions, making them difficult to apply in data-scarce areas such as plateaus, arid regions, and border areas. This invention only requires publicly available remote sensing data and general meteorological data, without relying on actual reservoir operation records or design data. On the one hand, remote sensing data has wide coverage and stable updates, adaptable to reservoirs of different sizes and climate types; on the other hand, the parametric design of the scheduling model supports flexible replacement of hydrological models, allowing adjustment of input variables and parameters according to watershed characteristics. This feature fills the gap in hydrological forecasting technology in data-scarce areas and meets the needs of real-time scheduling decision-making.

[0038] 3. Constructing an innovative technological path integrating remote sensing and hydrology, possessing both scientific research and industrial value: This invention breaks away from the traditional technological inertia of hydrological forecasting relying on measured data, innovatively integrating remote sensing inversion technology, machine learning, and hydrological simulation to form a complete technological chain of "data inversion - rule training - model coupling." At the scientific research level, it provides a new paradigm for studying hydrological processes in data-scarce watersheds, expanding the application boundaries of remote sensing technology in the field of hydrological simulation. At the industrial application level, it can directly serve scenarios such as water resource management, flood control, and ecological protection, without requiring additional investment in measured equipment, reducing the cost of technology implementation, and possessing broad prospects for promotion.

[0039] In summary, this invention achieves significant breakthroughs in forecast accuracy, scenario adaptability, and technological innovation. It not only solves the core problem of reservoir hydrological forecasting with insufficient data, but also provides an innovative "data-driven + model-coupled" solution for the field of hydrological forecasting, and has important technical and application value. Attached Figure Description

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0041] Figure 1 is a schematic diagram showing the location of the Mangyou River Basin in the application example;

[0042] Figure 2 shows the water storage volume change in the application example. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments disclosed herein will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0044] Example 1:

[0045] This embodiment provides a method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data, including the following steps:

[0046] Step 1, Model Building:

[0047] Based on the actual conditions of the study area, a distributed or semi-distributed hydrological model suitable for watersheds regulated by reservoirs (such as the Xin'anjiang model and the HBV model) is selected. Meteorological data of the study area are collected, and the watershed boundary and river network structure are delineated in combination with a high-precision digital elevation model. The locations of control sections and reservoirs are clarified, and the model parameters are calibrated and verified through measured data, thereby constructing a basic framework for hydrological simulation.

[0048] Step 2, Identify the water level-area-storage variable characteristics of reservoirs with no data:

[0049] Identify reservoirs within the study area lacking data and simultaneously acquire their water level-area-storage variation characteristics, specifically including:

[0050] S21, Reservoir water body extraction and area sequence construction: Water body extraction and temporal analysis are performed on identified reservoirs without data using multi-source remote sensing image data. The Normalized Differential Water Index (NDWI) is used to extract reservoir boundaries. An NDWI threshold is set to determine water body pixels and a minimum area threshold is set. Isolated patches are removed by combining morphological operations. The multi-temporal images are processed sequentially to obtain a daily reservoir area sequence. The multi-source remote sensing image data includes Jason1-3 and Landsat5-8 remote sensing image data.

[0051] The formula for calculating the Normalized Differential Water Index (NDWI) is as follows:

[0052]

[0053] In the formula: ρ Green and ρ NIRThese represent the green band and mid-infrared band in remote sensing images, respectively.

[0054] When the NDWI value is greater than the set threshold of 0.28, it is identified as a water body pixel. To avoid misjudgment due to noise and small water spots, a minimum area threshold of 30 meters is set, and isolated patches are removed by combining morphological opening and closing operations. When performing sequence processing on multi-temporal images, images from the relatively stable hydrological period within the same year (such as September to November) are preferred. If images for this period are lacking, images from the whole year are used to supplement the data. The influence of cloud shadows is reduced by median composite, and finally, the daily area sequence S(t) of the reservoir without data is obtained.

[0055] S22, Reservoir Regulation Characteristics Analysis: By comprehensively utilizing digital elevation model (DEM), geographic information system (GIS) and long-term hydrological observation data (flow rate, water level), the reservoir regulation characteristics are analyzed and the regulation type is determined, including whether it is annual regulation, seasonal regulation or daily regulation, and the typical regulation range of the reservoir is obtained.

[0056] S23, Reservoir Water Level Inversion and Unified Correction: For identified reservoirs without data, the reservoir water level is inverted using altimeter satellite data. The intersection segment between the altimeter satellite and the reservoir water body boundary is extracted, and effective trajectory points on the water surface are screened out. Waveform retracing is used, and outliers are removed through sliding window regression. The median value of the water surface points on the same day or passing through the same area is obtained after linear fitting and is taken as the representative water level H(t) of the reservoir at that time. At the same time, a unified benchmark conversion is performed on water level data from different satellite sources to ensure the consistency of time-series water level data and establish the reservoir water level-area relationship. The altimeter satellite data includes ICESat, Sentinel-3, or CryoSat-2 satellite data.

[0057] S24, Calculation of reservoir water storage sequence: Based on the reservoir area sequence and water level changes obtained above, calculate the reservoir water storage change, and then construct the water storage change time series of reservoirs without data.

[0058] To obtain the change in the relative water volume of a reservoir, we can first simplify the reservoir to an approximation of a trapezoid. When the water level rises, its surface area increases, and the water volume increases accordingly; when the water level falls, its surface area decreases, and the water volume decreases accordingly. Therefore, the change in the reservoir's water volume can be obtained from the changes in the reservoir's area and water level. Regardless of whether the water volume increases or decreases, the change in the reservoir's storage capacity is calculated using the following formula:

[0059]

[0060] In the formula: For the change in reservoir water volume over a specified period of time, and Let be the area of ​​the reservoir on day t and day t-1, respectively. and The elevations of the reservoir are on day t and day t-1, respectively.

[0061] Step 3, Training and parametric modeling of reservoir scheduling rules:

[0062] Based on remote sensing image data, a water level-area function relationship of the reservoir was constructed, and the water level and storage volume sequence for each 16-day period was obtained by inversion. The China Meteorological Forcing Data Set (CMFD) for the study area was introduced. This dataset contains key meteorological factors including precipitation, temperature, shortwave radiation, longwave radiation, wind speed, and specific humidity for each 16-day period. The data were processed into 16-day averages to characterize the reservoir replenishment and consumption process. The random forest regression method was used, with the above meteorological factors as input variables, to output the 16-day storage variable of the reservoir under the action of scheduling, that is, the difference between the actual storage volume of the reservoir and the storage volume under natural conditions. This 16-day time scale is consistent with the remote sensing inversion and meteorological driving data to ensure that the model accurately captures the periodic response characteristics of reservoir scheduling. During training, the model's fitting effect and generalization ability are evaluated using methods such as residual testing and cross-validation to select the optimal model structure and parameter combination, ensuring that the model can stably reproduce the reservoir's operating patterns. Finally, the 16-day storage variable is superimposed as a correction factor onto the original hydrological model's flow forecast results to construct a parameterized module for scheduling rules. This module corrects and reproduces the downstream cross-sectional runoff process under reservoir scheduling, thus accurately simulating the impact of reservoir operation on downstream runoff even in the absence of scheduling data or with incomplete scheduling rules, thereby improving the accuracy and reliability of hydrological forecasts.

[0063] Step 4, Coupling the watershed storage process simulation with reservoir scheduling:

[0064] The scheduling rule parameterization module established in step 3 is embedded into the hydrological model structure. During the model operation, the scheduling module is called in real time according to the meteorological forcing data (CMFD) at the current moment, and the corresponding storage variables are output to realize the dynamic simulation of the watershed hydrological process by the reservoir regulation behavior.

[0065] The specific implementation method is as follows: Input the temperature, air pressure, specific humidity, wind speed, downward shortwave radiation flux, downward longwave radiation flux, and precipitation from the CMFD into the hydrological model, and then call the scheduling module to calculate the reservoir's storage variable. Since this scheduling rule is derived from the actual water storage state retrieved through remote sensing inversion and driven by historical meteorological forcing, it can significantly improve the accuracy and real-world adaptability of the simulation results in the absence of measured scheduling data. It is particularly suitable for hydrological forecasting applications in typical data-scarce areas such as plateaus, arid regions, and border areas.

[0066] Application examples:

[0067] Taking the Mangyou River Reservoir in Lincang, Yunnan Province as an example, the implementation process of the data-deficient reservoir hydrological forecasting method described in this invention is explained below:

[0068] Step 1: Model Building

[0069] Using data from the China Regional Surface Meteorological Element Driven Dataset (CMFD) from 1979 to 2018, air temperature, air pressure, specific humidity, wind speed, downward shortwave radiation flux, downward longwave radiation flux, and precipitation were processed into 16-day averages. Based on the 30-meter resolution ASTER GDEM V3 digital elevation model, ArcGIS was used to divide the Mangyou River basin into sub-basins (Figure 1) and determine reservoir coordinates. The Xin'anjiang Three-Source Model was selected, and parameters were calibrated in SWAT-CUP. Key parameters included the evapotranspiration conversion factor K=0.82 and the average soil water storage capacity WM=120mm. The calibration period was 1985-2000, and the validation period was 2001-2010. The objective function NSE>0.75.

[0070] Step 2: Identify the characteristics of water level-area-storage changes in reservoirs with no available data:

[0071] Using Landsat remote sensing imagery from 2003 to 2024, the reservoir water surface boundary was extracted for 16-day intervals based on the NDWI index and the Otsu thresholding method, obtaining a sequence of water surface area changes. Simultaneously, lake water levels were extracted using orbital data provided by the ICESat altimeter satellite from 2003 to 2009. Specifically, ICESat elevation data was converted from a T / P ellipsoid to a WGS84 ellipsoid, geoid correction was performed, and lake surface elevation points on the remote sensing water body boundary extraction trajectory were overlaid. After removing outliers, the median and mean methods were used to obtain the lake water level. The water level sequence for the entire study period was further calculated using a water level-area relationship fitting function. Finally, based on the geometric relationship between area and water level, the 16-day reservoir water storage change process was estimated, providing a target variable for scheduling rule modeling.

[0072] Step 3: Reservoir scheduling rule training and parametric modeling

[0073] Based on the 16-day water storage variation sequence of the reservoir obtained through remote sensing inversion and CMFD meteorological data processed with a 16-day average, a reservoir scheduling module is established using the random forest regression method. The model input includes meteorological driving factors (such as precipitation, temperature, wind speed, humidity, etc.) and variables such as the water storage status before and after the current period; the output is the storage variable. During the training process, the model's performance is evaluated through cross-validation and residual analysis to ensure that the goodness of fit is generally not less than 0.8 and that the assumptions of residual normality and independence are met. Finally, the model training results are solidified into a callable parameterized scheduling function, realizing the function of predicting the storage variable from meteorological conditions. This model does not require actual measured scheduling records, relying only on publicly available remote sensing and meteorological data, and possesses good adaptability and transferability.

[0074] Step 4: Coupling of watershed storage simulation and scheduling:

[0075] The scheduling module was embedded in the Xin'anjiang model and invoked every 16 days: inputting meteorological forcing data and outputting storage variables. Figure 2 shows the changes in storage variables, verifying the coupling mechanism: the sharp drop in storage in December 2023 corresponds to flood control scheduling behavior, while the steady decline in February 2024 reflects dry season regulation. Experiments show that the coupled model significantly improves accuracy: peak flow error is reduced by 21%, dry season flow stability is improved by 14%, and the Nash coefficient for fitting storage changes reaches 0.79, proving the effectiveness of this method in data-scarce reservoir basins.

[0076] Finally, it should be noted that the above is only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention (such as the application of various formulas, the order of steps, etc.) without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for forecasting hydrological conditions in reservoirs with limited data based on remote sensing data, characterized in that, The method includes the following steps: Step 1, Model Construction: Based on the actual situation of the study area, a distributed or semi-distributed hydrological model suitable for the reservoir-regulated watershed is selected. Meteorological data of the study area is collected. The watershed boundary and river network structure are delineated using a high-precision digital elevation model. The location of control sections and reservoirs is clarified. The model is calibrated and verified using measured data, thereby constructing a basic framework for hydrological simulation. Step 2, Identifying Water Level-Area-Storage Variable Characteristics of Reservoirs Without Data: Reservoirs without data are identified within the study area, and their water level-area-storage variable change characteristics are obtained simultaneously. Specifically, this includes: S21, Reservoir Water Body Extraction and Area Sequence Construction: Water body extraction and time series analysis are performed on the identified reservoirs without data using multi-source remote sensing image data. The reservoir boundary is extracted using the Normalized Difference Water Index (NDWI), and the NDWI is set. Thresholding is used to determine water body pixels and set a minimum area threshold. Isolated patches are removed using morphological operations. Multi-temporal images are processed sequentially to obtain a daily reservoir area sequence; S22, Reservoir regulation characteristics analysis: Utilizing digital elevation models, geographic information systems, and daily reservoir area sequences, reservoir regulation characteristics are analyzed, regulation types are determined, and typical regulation ranges of the reservoir are obtained; S23, Reservoir water level inversion and unified correction: Reservoir water levels are inverted using altimeter satellite data, effective trajectory points are extracted and outliers are removed, and a unified benchmark conversion is performed on water level data from different satellite sources to establish a reservoir water level-area relationship; S24, Reservoir water storage volume sequence calculation: Based on the obtained reservoir area sequence and water level changes, reservoir water storage volume changes are calculated, and a time series of water storage volume changes for reservoirs without data is constructed; Step 3, Training and parameterized modeling of reservoir scheduling rules: A water level-area function relationship for the reservoir was constructed using remote sensing image data, and the water level and storage volume sequences for each of the 16 days during the study period were obtained by inversion. A meteorological dataset of the study area was introduced. The random forest regression method was used, with meteorological factors as input variables, to output the storage variables of the reservoir under the action of scheduling. The optimal model structure and parameter combination were screened through residual test and cross-validation. The storage variables of each of the 16 days were used as correction factors and superimposed on the flow forecast results of the original hydrological model, thereby constructing the scheduling rule parameterization module. Step 4, watershed storage process simulation and reservoir scheduling coupling: The scheduling rule parameterization module established in step 3 was embedded as a parameterization module into the hydrological model structure. During the model operation, the scheduling module was called in real time according to the meteorological forcing data at the current moment, and the corresponding storage variables were output, realizing the dynamic simulation of the watershed hydrological process by the reservoir regulation behavior.

2. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 1, the distributed or semi-distributed hydrological model includes the Xin'anjiang model and the HBV model.

3. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 2, the multi-source remote sensing image data includes Jason1-3 and Landsat5-8 remote sensing image data, and the altimeter satellite data includes ICESat, Sentinel-3, or CryoSat-2 satellite data.

4. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 2, the formula for calculating the Normalized Difference Water Index (NDWI) is as follows: In the formula: ρ Green and ρ NIR These represent the green band and mid-infrared band in remote sensing images, respectively.

5. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 2, the change in reservoir water storage is calculated using the following formula: In the formula: For the change in reservoir water volume over a specified period of time, and Let be the area of ​​the reservoir on day t and day t-1, respectively. and The elevations of the reservoir are on day t and day t-1, respectively.

6. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 2, when performing sequence processing on multi-temporal images, images from the relatively stable hydrological period within the same year are preferred. If images for that period are lacking, images from the entire year are used to supplement the data. The influence of cloud shadows is reduced by median synthesis, and finally, the daily area sequence S(t) of the reservoir without data is obtained.

7. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 3, the 16-day storage variable is the difference between the actual water storage of the reservoir and the water storage under natural conditions, and the time scale is set to a 16-day scale.

8. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 3, the meteorological dataset includes key meteorological factors such as 16-day precipitation, temperature, shortwave radiation, longwave radiation, wind speed, and specific humidity.

9. The method for forecasting hydrological conditions in reservoirs with insufficient data based on remote sensing data according to claim 1, characterized in that, In step 4, the meteorological forcing data includes air temperature, air pressure, specific humidity, wind speed, downward shortwave radiation flux, downward longwave radiation flux, and precipitation.

Images