Drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning

Abstract

The application provides a drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning, acquires NDVI time series data corresponding to a research area, and extracts feature information of a disaster-bearing body corresponding to the research area based on the NDVI time series data; acquires water resource supply data corresponding to the research area, and extracts water resource supply features corresponding to the water resource supply data; through a machine learning model obtained by pre-training, a disaster-bearing body NDVI prediction value of the research area at a future time is predicted based on the disaster-bearing body feature information and the water resource supply features; the disaster-bearing body NDVI prediction value is used for identifying the drought risk of the research area in combination with precipitation forecast data, or dynamically adjusting the anti-drought irrigation strategy data. The application can significantly improve the accuracy of drought risk prediction and effectively improve the dynamic adjustment effect of the anti-drought irrigation strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of drought risk identification and early warning, and particularly relates to a drought risk identification and early warning method based on multi-source satellite fused NDVI and machine learning. BACKGROUND

[0002] At present, existing drought risk identification and early warning means usually include means based on meteorological station data, means based on satellite remote sensing technology, and soil moisture sensor network, etc., but the above means still have certain technical defects, resulting in problems such as insufficient drought risk early warning accuracy and lagging drought resistance strategy formulation. SUMMARY

[0003] Therefore, the present application aims to provide a drought risk identification and early warning method based on multi-source satellite fused NDVI and machine learning, which can significantly improve the accuracy of drought risk prediction and effectively improve the dynamic adjustment effect of drought resistance irrigation strategy.

[0004] In a first aspect, the present application provides a drought risk identification and early warning method based on multi-source satellite fused NDVI and machine learning, comprising:

[0005] obtaining NDVI time series data corresponding to a research area, and extracting hazard-bearing body feature information corresponding to the research area based on the NDVI time series data, the hazard-bearing body feature information including one or more of NDVI time series features, NDVI fluctuation features and NDVI change rate features;

[0006] obtaining water resource supply data corresponding to the research area, and extracting water resource supply features corresponding to the water resource supply data, the water resource supply data including precipitation forecast data, or including precipitation forecast data and drought resistance irrigation strategy data;

[0007] predicting a hazard-bearing body NDVI prediction value of the research area at a future time through a machine learning model obtained by pre-training, based on the hazard-bearing body feature information and the water resource supply features;

[0008] The hazard-bearing body NDVI prediction value is used to identify the drought risk of the research area in combination with the precipitation forecast data in the case where the water resource supply data only includes the precipitation forecast data, and to dynamically adjust the drought resistance irrigation strategy data in the case where the water resource supply data includes the precipitation forecast data and the drought resistance irrigation strategy data.

[0009] In an implementation, the hazard-bearing body feature information corresponding to the research area is extracted based on the NDVI time series data, comprising:

[0010] The NDVI time series data is subjected to a local weighted regression processing to extract NDVI time series features of the NDVI time series data, the NDVI time series features including a long-term change trend feature, a seasonal feature and a residual feature;

[0011] In addition, a short-term abnormal fluctuation threshold is determined based on the residual feature, and an NDVI fluctuation feature of the NDVI time series data is extracted according to the residual feature and the short-term abnormal fluctuation threshold, the fluctuation feature being used to describe a short-term abnormal fluctuation of the NDVI time series data;

[0012] In addition, a sliding window is set for the NDVI time series data, a change rate corresponding to the sliding window is determined based on an NDVI value corresponding to each data point located in the sliding window, the sliding window is continuously moved until the sliding window moves to the end of the NDVI time series data, and an NDVI change rate feature corresponding to the NDVI time series data is extracted based on the change rates corresponding to all the sliding windows, the NDVI change rate feature being used to reflect a disaster-bearing body vulnerability and a disaster-bearing body resilience.

[0013] In an embodiment, water resource supply features corresponding to the water resource supply data are extracted, including:

[0014] Cumulative precipitation features, precipitation anomaly features and precipitation fluctuation features corresponding to the precipitation forecast data are extracted.

[0015] The cumulative precipitation features are used to represent a water resource supply capacity of the research region in a short term, the precipitation anomaly features are used to reflect an abnormal degree of precipitation in the research region, and the precipitation fluctuation features are used to describe a uniformity of precipitation distribution in the research region.

[0016] In an embodiment, the method further includes:

[0017] The seasonal feature contained in the NDVI time series features is used to eliminate an influence of seasonal factors on the disaster-bearing body NDVI predicted value, so as to obtain a new disaster-bearing body NDVI predicted value.

[0018] Based on the new disaster-bearing body NDVI predicted value, an NDVI prediction rate feature corresponding to the research region is extracted.

[0019] According to a change trend and a positive or negative nature of the NDVI prediction rate feature, an initial drought risk level corresponding to the research region is determined.

[0020] The precipitation forecast data is used as an auxiliary factor to correct the initial drought risk level, so as to obtain a target drought risk level corresponding to the research region.

[0021] In an embodiment, the method further includes:

[0022] The machine learning model obtained by pre-training is used to predict a disaster-bearing body NDVI prediction value and a target drought risk level of the research region at a future moment based on the disaster-bearing body feature information and the water resource supply feature.

[0023] In an embodiment, the anti-drought irrigation strategy data is used to describe an irrigation point deployed in the research region and a corresponding irrigation amount and irrigation time thereof; the method further comprises:

[0024] An evaluation value corresponding to the anti-drought irrigation strategy data is determined based on the disaster-bearing body NDVI prediction value through an irrigation strategy evaluation function.

[0025] The anti-drought irrigation strategy is adjusted based on the evaluation value.

[0026] The machine learning model is used to predict a new disaster-bearing body NDVI prediction value of the research region at a future moment based on the disaster-bearing body feature information, the water resource supply feature and a new anti-drought irrigation strategy feature until a preset stop condition is met, so as to obtain optimal anti-drought irrigation strategy data.

[0027] In an embodiment, the irrigation strategy evaluation function is a function aiming at minimizing irrigation cost and maximizing drought risk reduction.

[0028] In a second aspect, the present application further provides a drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning, comprising:

[0029] A disaster-bearing body feature extraction module is configured to acquire NDVI time series data corresponding to the research region, and extract disaster-bearing body feature information corresponding to the research region based on the NDVI time series data, wherein the disaster-bearing body feature information comprises one or more of NDVI time series features, NDVI fluctuation features and NDVI change rate features.

[0030] A water resource supply feature extraction module is configured to acquire water resource supply data corresponding to the research region, and extract water resource supply features corresponding to the water resource supply data, wherein the water resource supply data comprises precipitation forecast data, or comprises both the precipitation forecast data and anti-drought irrigation strategy data.

[0031] A prediction module is configured to use a machine learning model obtained by pre-training to predict a disaster-bearing body NDVI prediction value of the research region at a future moment based on the disaster-bearing body feature information and the water resource supply feature.

[0032] The disaster-bearing body NDVI prediction value is used to identify the drought risk of the research region in combination with the precipitation forecast data in the case where the water resource supply data only comprises the precipitation forecast data, and is used to dynamically adjust the anti-drought irrigation strategy data in the case where the water resource supply data comprises both the precipitation forecast data and the anti-drought irrigation strategy data.

[0033] In a third aspect, the present application also provides an electronic device, comprising a processor and a memory, the memory storing computer executable instructions capable of being executed by the processor, and the processor executes the computer executable instructions to implement the method of any one of the first aspect.

[0034] In a fourth aspect, the present application also provides a computer readable storage medium, the computer readable storage medium storing computer executable instructions, and the computer executable instructions, when invoked and executed by a processor, cause the processor to implement the method of any one of the first aspect.

[0035] The present application provides a drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning, which first acquires NDVI time series data corresponding to a research region, and extracts disaster-bearing body feature information corresponding to the research region based on the NDVI time series data, the disaster-bearing body feature information including one or more of NDVI time series features, NDVI fluctuation features and NDVI change rate features; then acquires water resource supply data corresponding to the research region, and extracts water resource supply features corresponding to the water resource supply data, the water resource supply data including precipitation forecast data, or including precipitation forecast data and anti-drought irrigation strategy data; finally, through a machine learning model obtained by pre-training, the disaster-bearing body NDVI prediction value of the research region at a future time is predicted based on the disaster-bearing body feature information and the water resource supply feature; wherein the disaster-bearing body NDVI prediction value is used to: in the case that the water resource supply data only contains the precipitation forecast data, the drought risk of the research region is identified in combination with the precipitation forecast data; in the case that the water resource supply data includes the precipitation forecast data and the anti-drought irrigation strategy data, the anti-drought irrigation strategy data is dynamically adjusted. The above method automatically identifies and evaluates the drought risk through specific NDVI time series features and machine learning models, and realizes real-time identification and grading early warning of drought risks of different climate zones and ecological systems based on the growth and health conditions of vegetation, and can also realize dynamic adjustment of anti-drought irrigation strategy data. Compared with the traditional drought risk identification and early warning method, the method has significant advantages in NDVI time series feature extraction, drought risk monitoring and identification, risk level division, prediction accuracy and strategy adjustment effect, and is particularly suitable for real-time dynamic identification and early warning of drought risks of regions containing various different ecological systems (such as forests, grasslands, farmlands, deserts and wetlands).

[0036] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the description, claims and drawings.

[0037] In order to make the above objectives, characteristics and advantages of the present application more apparent, clear and easy to understand, the following preferred embodiments are specifically described below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings described below are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0039] Figure 1 A process schematic diagram of a drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning provided for an embodiment of the present application;

[0040] Figure 2 A specific processing flow schematic diagram of model building provided for an embodiment of the present application;

[0041] Figure 3 A drought risk early warning business process diagram based on fusion NDVI provided for an embodiment of the present application;

[0042] Figure 4 A structure schematic diagram of a drought risk identification and early warning device based on multi-source satellite fusion NDVI and machine learning provided for an embodiment of the present application;

[0043] Figure 5 A structure schematic diagram of an electronic device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the present application will be described below in connection with the embodiments, obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0045] At present, the existing drought risk identification and early warning means has problems of insufficient drought risk prediction accuracy and lagging drought resistance strategy making, based on which, the present application provides a drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning, which can significantly improve the accuracy of drought risk prediction and effectively improve the dynamic adjustment effect of drought resistance irrigation strategy.

[0046] For the convenience of understanding the present embodiment, first, a kind of drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning disclosed in the embodiment of the application is introduced in detail, referring to Figure 1 The flowchart of the drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning is shown in Fig.

[0047] Step S102, the NDVI time series data corresponding to the study area is obtained, and the disaster-bearing body feature information corresponding to the study area is extracted based on the NDVI time series data.

[0048] Among them, the disaster-bearing body feature information includes one or more of NDVI time series feature, NDVI fluctuation feature and NDVI change rate feature, the NDVI time series feature includes long-term change trend feature, seasonal feature and residual feature, the NDVI fluctuation feature is used to describe the short-term abnormal fluctuation of NDVI time series data, the NDVI change rate feature is used to reflect the vulnerability of disaster-bearing body and the resilience of disaster-bearing body, and the NDVI change rate feature includes maximum change rate, average change rate and standard deviation of change rate.

[0049] In an example, the NDVI time series data is decomposed by local weighted regression to obtain the NDVI time series feature; the short-term abnormal fluctuation of the NDVI time series data is further identified based on the residual feature in the NDVI time series feature to obtain the NDVI fluctuation feature; the change rate of the NDVI time series data is determined, and the NDVI change rate feature is extracted on this basis.

[0050] Step S104, the water resource supply data corresponding to the study area is obtained, and the water resource supply feature corresponding to the water resource supply data is extracted.

[0051] Among them, the water resource supply data includes precipitation forecast data, or includes precipitation forecast data and anti-drought irrigation strategy data, and the water resource supply feature includes cumulative precipitation feature, precipitation anomaly feature and precipitation fluctuation feature of the precipitation forecast data, wherein the cumulative precipitation feature is used to represent the water resource supply capacity of the study area in the short term, the precipitation anomaly feature is used to reflect the abnormal degree of precipitation in the study area, and the precipitation fluctuation feature is used to describe the uniformity of precipitation distribution in the study area.

[0052] In step S106, the machine learning model obtained by pre-training is used to predict the NDVI prediction value of the hazard body in the research area at the future time based on the hazard body feature information and the water resource supply feature; wherein, the NDVI prediction value of the hazard body is used to identify the drought risk of the research area in combination with the precipitation forecast data in the case that the water resource supply data only includes the precipitation forecast data; and the drought-resistant irrigation strategy data is dynamically adjusted in the case that the water resource supply data includes the precipitation forecast data and the drought-resistant irrigation strategy data.

[0053] In one example, the input of the machine learning model is the hazard body feature information and the water resource supply feature, and the output is the NDVI prediction value of the hazard body; or, the input of the machine learning model is the hazard body feature information and the water resource supply feature, and the output is the NDVI prediction value of the hazard body and the target drought risk level, that is, the machine learning model obtained by pre-training is used to predict the NDVI prediction value of the hazard body in the research area at the future time and the target drought risk level based on the hazard body feature information and the water resource supply feature; or, the input of the machine learning model is the hazard body feature information, the water resource supply feature and the drought-resistant irrigation strategy data, and the output is the NDVI prediction value of the hazard body.

[0054] The drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning provided by the embodiments of the present application realizes automatic identification and evaluation of drought risk through specific NDVI time series features and machine learning models, and realizes real-time identification and hierarchical early warning of drought risk of different climate zones and ecological systems based on the growth and health conditions of vegetation, and can also dynamically adjust the drought-resistant irrigation strategy data. Compared with the traditional drought risk identification and early warning method, the method has significant advantages in NDVI time series feature extraction, drought risk monitoring and identification, risk level division, prediction accuracy and strategy adjustment effect, and is particularly suitable for real-time dynamic identification and early warning of drought risk of regions containing various different ecological systems (such as forests, grasslands, farmlands, deserts and wetlands).

[0055] For the convenience of understanding, the embodiments of the present application provide a specific implementation of the drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning.

[0056] For the foregoing step S102, the embodiments of the present application provide a specific implementation of extracting the hazard body feature information corresponding to the research area based on the NDVI time series data.

[0057] NDVI (Normalized Difference Vegetation Index) is a key indicator of vegetation health and coverage, closely related to exposure, vulnerability, and resilience. Areas with low NDVI values have sparse or severely degraded vegetation, often with insufficient coverage, typically have higher exposure and vulnerability, and are extremely sensitive to external disturbances, making them vulnerable to environmental pressures such as drought. Conversely, areas with high NDVI values and small changes in amplitude generally have stronger resilience, i.e., they can quickly recover vegetation health in adverse conditions such as drought.

[0058] The preprocessing of NDVI time series data mainly removes interference data (such as clouds) and outliers from satellite fusion data, ensuring the smoothness of time series data. It mainly includes the following aspects, (1) Time series interpolation: Interpolate the NDVI values with cloud coverage using the data at the previous and next time points, commonly used methods include linear interpolation and spline interpolation. (2) Spatial neighborhood averaging: When the time series method is not applicable (such as missing key monitoring dates), the spatial neighborhood averaging method can be used to fill in the missing values with the average NDVI of neighboring pixels. (3) Outlier detection and processing: Based on experience or historical data, set a reasonable range for NDVI, and remove outliers exceeding the threshold. For example, NDVI values in vegetation areas are usually between 0 and 1, and abnormally high or low values may be noise. Use a sliding window to calculate the average NDVI, smooth short-term fluctuations, and thus reduce noise. Calculate the average and standard deviation of NDVI values to identify and remove abnormal data points exceeding a certain Z-score threshold.

[0059] Specifically, the process of extracting hazard-bearing body feature information is as follows:

[0060] (1.1) Local weighted regression processing is performed on the NDVI time series data to extract the NDVI time series features of the NDVI time series data, including long-term trend features, seasonal features, and residual features.

[0061] The time series features of NDVI can directly reflect the vulnerability and resilience of the hazard-bearing body when facing environmental pressures (such as drought, climate change, etc.). By analyzing the time series changes of NDVI, the long-term average state and short-term dynamic trend of vegetation health can be evaluated, and it can be identified which areas are more vulnerable to disasters and the recovery ability of these areas after disasters. Stable and less volatile time series features usually indicate that the hazard-bearing body has strong resilience, while sharp declines or long-term low values mean higher vulnerability. Therefore, how to accurately extract and analyze the time series features of NDVI is a key step in technical applications, which directly affects the accuracy and efficiency of drought risk identification, ecological impact assessment, and disaster warning.

[0062] In a specific implementation, a seasonal-trend decomposition method is used to extract the long-term trend in the data. Seasonal-Trend Decomposition Using LOESS (STL) is a time series decomposition method based on the locally weighted regression method (LOESS), commonly used to extract the long-term trend, seasonal component, and residual in time series data.

[0063] The core idea of STL is to decompose time series data into three parts: (1) Trend Component: reflects the long-term trend of data, is the smoothest part, removes seasonal fluctuations and short-term anomalies. (2) Seasonal Component: reflects the seasonal fluctuations in the data, usually the periodic part. (3) Residual Component: the noise part after removing the trend and seasonality, that is, the irregular component in the data.

[0064] STL uses local weighted regression (LOESS) to smooth time series data to obtain the decomposed components: first, smooth the time series data by local regression to obtain the long-term trend part, LOESS regression is a smoothing method based on local neighborhood weighting, suitable for processing complex nonlinear data. Secondly, in the data after removing the trend, extract the periodic fluctuations (i.e. seasonal component), LOESS regression will smooth the data according to the local situation of each time point, helping to identify seasonal patterns. Finally, by removing the trend and seasonal components, the residual part is obtained, representing the part of the data not explained by the trend and seasonal components.

[0065] The specific decomposition process can be completed through the following steps: first, use the LOESS method to perform local weighted regression on the original time series data to extract a preliminary trend component. The key to the LOESS method is to use weighted least squares regression fitting based on the data around each data point, so it can effectively remove local noise. After removing the trend component, the remaining part is considered to be seasonal fluctuations. Secondly, STL further applies LOESS smoothing to the de-trended data to identify periodic seasonal changes. The seasonal component reflects the periodic fluctuations in the data, such as changes in different seasons within a year. Finally, the data after removing the trend and seasonal components is called residual, which represents the random fluctuations or noise part in the data. The STL method extracts the residual component, and the remaining part is the trend and seasonal components.

[0066] For specific parameter settings: STL method needs to set a seasonal period (e.g. NDVI data with annual period needs 365 days), which determines the accuracy of the decomposition of the seasonal component of the data. Trend smoothing window: the smoothness of the trend extraction is determined by the size of the window. The larger the window, the smoother the extracted trend, which can effectively remove short-term fluctuations, but may ignore some small trend changes. Seasonal smoothing window: the smoothness of the seasonal fluctuations determines the accuracy of the extraction of the seasonal component. Too large a window will cause the seasonal fluctuations to be blurred, while too small a window may not accurately capture the seasonal changes.

[0067] STL method can adaptively extract trend and seasonal components according to the changes of data through local weighted regression, which is suitable for complex and nonlinear time series data, can effectively remove seasonal fluctuations, help identify long-term trends in data, and is particularly suitable for drought risk identification and early warning, which needs to pay attention to long-term changes. It can handle irregular data: due to the local weighted characteristics of LOESS regression method, STL method has good robustness for data containing noise and missing values.

[0068] In drought risk identification, STL method can clearly observe the annual change trend of NDVI by removing seasonal fluctuations, such as the long-term downward trend of NDVI, which reflects the change of vegetation condition. Seasonal component helps to reveal the seasonal change of vegetation growth in a year, identifies the change of vegetation coverage in different seasons, and provides reference for seasonal characteristics analysis of drought. The residual part can reveal the abnormal fluctuations in the data, such as the degree of influence of extreme weather events on vegetation, which helps to identify the size of the influence of drought and other extreme events.

[0069] STL method can effectively extract long-term trend, seasonal fluctuation and random noise by decomposing time series data through local weighted regression. In drought identification, using STL to decompose NDVI data can clearly identify the long-term downward trend of vegetation coverage, while removing the influence of seasonal fluctuations, providing reliable analysis basis for early warning of drought risk.

[0070] (1.2) Based on the residual characteristics and the short-term abnormal fluctuation threshold, the NDVI fluctuation characteristics of the NDVI time series data are extracted, and the fluctuation characteristics are used to describe the short-term abnormal fluctuation of the NDVI time series data.

[0071] Abnormal fluctuations in NDVI in the short term are important indicators of vulnerability and resilience of hazard-affected bodies, and early signals of possible drought events. To identify abnormal fluctuations in NDVI in the short term and use them as characteristic parameters for predicting drought risk, a residual-based anomaly detection method can be used. This method identifies abnormal fluctuations by analyzing the residuals of the decomposed data.

[0072] After using seasonal-trend decomposition (STL), we can obtain three parts: the trend component, the seasonal component, and the residuals. Abnormal fluctuations are usually represented by extreme values or abrupt changes in the residuals part. We identify abnormal fluctuations in the short term based on the residuals part as one of the early signals for drought risk warning.

[0073] The residuals part after STL decomposition represents irregular fluctuations after removing the trend and seasonal components. Drought events usually cause a sharp decline in plant growth, which is reflected as abnormal fluctuations in NDVI data. Therefore, extreme changes or abrupt changes in the residuals part can be considered as early signals of drought. Then, by calculating the mean and standard deviation of the residuals part, an anomaly threshold is defined. For example, set the threshold to be the mean of the residuals ± 1.5 or 2 times the standard deviation, and any residual point exceeding this threshold is considered an abnormal fluctuation. When the residuals exceed the set threshold, it is considered that an abnormal fluctuation has occurred, indicating a possible drought event. By examining the fluctuations in the residuals part, identify those outliers that change rapidly and are outliers, as early signals of potential drought events. If the residual value at a certain point in time is significantly higher or lower than the normal range and lasts for a period of time, it may indicate the beginning of a drought.

[0074] If the residuals part of NDVI has a large abnormal change in the short term (such as a few days or weeks), and this change exceeds the normal fluctuation range, it can be considered as an early signal of drought. For example, the NDVI of a certain area should remain stable within the normal seasonal fluctuation range, but a sudden and significant decline that lasts for a period of time may indicate that the area is experiencing drought.

[0075] When such abnormal fluctuations are identified, a drought warning signal can be issued, indicating that a drought may occur or that drought prevention measures need to be taken.

[0076] The residual-based anomaly detection method can effectively identify short-term abnormal fluctuations in NDVI data, which are usually early signals of drought. By analyzing the residuals part after STL decomposition and setting a reasonable anomaly threshold, we can detect abnormal changes in vegetation cover in a timely manner before drought occurs, providing a basis for identifying and early warning of drought risk.

[0077] (1.3) Set a sliding window for the NDVI time series data, and determine the change rate of the sliding window based on the NDVI value of each data point within the sliding window. Continue to move the sliding window until it reaches the end of the NDVI time series data, and based on the change rate of all sliding windows, extract the NDVI change rate feature corresponding to the NDVI time series data. The NDVI change rate feature is used to reflect the vulnerability and resilience of the hazard-affected body.

[0078] The change rate of NDVI is used to detect the speed of drought intensification or alleviation, providing feature support for machine learning models. Change rate refers to the speed of NDVI change within a certain period of time, which reflects the degree and rate of vegetation cover change. In drought risk identification and early warning, change rate can be an important feature of drought intensification or alleviation, especially when drought continues to intensify or alleviate, the change rate of NDVI will increase significantly. By calculating the change rate of NDVI, we provide effective feature support for machine learning models to help predict and evaluate the trend and severity of drought.

[0079] Change rate can usually be achieved by calculating the change amount of NDVI time series. The specific steps are as follows:

[0080] (1.31) Calculate the change rate of NDVI: The change rate of NDVI can be calculated by first-order difference of NDVI time series data. First-order difference is the difference between the current time point and the previous time point, representing the change amount of NDVI. The formula is:

[0081] Change rate t = NDVI t - NDVI t-1 ;

[0082] Where NDVI t is the NDVI value of the current time point, and NDVI t-1 is the NDVI value of the previous time point.

[0083] To eliminate the influence of seasonal fluctuations, we standardize the change rate of NDVI to reduce the interference of seasonal changes and focus on vegetation changes caused by drought. For example, we can divide the change rate by the standard deviation of NDVI to make it have zero mean and unit variance:

[0084]

[0085] To capture the long-term development trend of drought, we calculate the NDVI change rate within a longer time window (such as one month, one quarter, one year). The method is as follows:

[0086] A fixed window (e.g., 30 days) is slid over the time series, and the mean or median of the NDVI change rate within each window is calculated. This effectively smooths out short-term fluctuations and highlights long-term trends.

[0087]

[0088] where N is the length of the window.

[0089] (1.32) Statistical features of vulnerability and resilience of the bearing body: The change rate of vegetation index is an important indicator to reflect the vulnerability and resilience of the bearing body. As an important feature in the machine learning model, it helps the model to capture the timeliness and severity of drought risk. For example, areas with larger change rates may require more urgent drought warning. The change rate includes short-term changes, which are directly used as inputs to the drought warning system, and the change rate of long-term trends, which further identify the speed of drought intensification.

[0090] In addition to directly calculating the change rate, its statistical features are extracted to enhance the prediction ability of the model: (1) maximum change rate: captures the most dramatic moment of vegetation change, often used to identify sudden changes in drought. (2) average change rate: reflects the long-term change speed, used to detect the gradual aggravation or mitigation of drought. (3) standard deviation of change rate: used to measure the degree of fluctuation of change, high standard deviation may represent a drought event with large fluctuations.

[0091] For the foregoing step S104, the embodiment of the present application provides a specific implementation of extracting water supply features corresponding to water supply data, including extracting cumulative precipitation features, precipitation anomaly features, and precipitation fluctuation features corresponding to precipitation forecast data.

[0092] The preprocessing of model grid data such as GRAPES (Global / Regional Assimilation and Prediction System), EC (European Centre for Medium-Range Weather Forecasts), and CFS (Climate Forecast System) is an important step to ensure efficient combination with NDVI data. First, the daily grid precipitation data provided by the model for the next 30 days needs to be time-aligned. Since the time resolution of NDVI data may be weekly or monthly, the daily grid precipitation provided by the model must be adjusted to the same time step as NDVI through aggregation or interpolation methods. For example, daily precipitation is aggregated into weekly cumulative precipitation to better describe the impact of precipitation on vegetation change.

[0093] Secondly, the spatial resolution of GRAPES, EC, CFS and other model grid data is usually different from that of NDVI data (for example, GRAPES is 0.25°x0.25°, while NDVI is 1km). It is necessary to adjust the resolution of the model grid data to the level matching NDVI through spatial resampling technology to ensure that both can be analyzed on the same spatial scale. In addition, precipitation data usually contains short-term violent fluctuations, which may interfere with the training of the model. Therefore, using methods such as moving average or exponential weighted smoothing to smooth the precipitation data can effectively reduce noise and highlight long-term trends and regional characteristics.

[0094] In order to further tap the information of precipitation data, a series of features can also be calculated for GRAPES, EC, CFS and other model grid data, such as the cumulative precipitation in the next 30 days, to represent the water supply capacity in the short term. In addition, by analyzing the deviation of GRAPES, EC, CFS and other model grid data from historical precipitation data, the standardized precipitation index (SPI) can be calculated to reflect the degree of precipitation anomaly. Such features are particularly important for identifying abnormal precipitation patterns that may trigger drought.

[0095] In the model training process, it is crucial to integrate GRAPES, EC, CFS and other model grid data with NDVI data into a multi-dimensional time series input. Precipitation data can be directly input into the model at a daily time step to capture the dynamic impact of precipitation on vegetation change. At the same time, constructing precipitation fluctuation features (such as standard deviation or coefficient of variation) can help the model identify the potential impact of precipitation distribution uniformity on drought development. Through this series of feature extraction and preprocessing methods, model grid data can work together with NDVI data to provide rich and accurate input support for drought identification and early warning.

[0096] For the foregoing step S106, the embodiments of the present application first explain the machine learning model.

[0097] The Long Short-Term Memory (LSTM) model is used to machine learn the time series of NDVI to capture the dynamic changes of drought risk. The LSTM (Long Short-Term Memory) model is a special type of recurrent neural network (RNN) that is particularly good at processing and predicting time series-based data, and can effectively capture long-term dependencies and nonlinear changes. In drought risk identification and early warning, the LSTM model can use NDVI time series data and precipitation forecasts to predict the development, change and potential severity of drought and the exposure of vegetation index, and provide the vulnerability and resilience of vegetation index, to achieve dynamic identification and early warning of drought risk.

[0098] LSTM is an extension of Recurrent Neural Networks (RNNs) designed to address the vanishing and exploding gradient problems encountered by traditional RNNs when processing long-term data. LSTM introduces memory units, allowing the model to selectively retain or forget information, thus effectively capturing long-term dependencies. The core components of LSTM include forget gates, input gates, and output gates, which determine which information to discard, which new information to add, and which information will affect the final output. These gating mechanisms, by flexibly controlling the flow of information, enable LSTM to better capture long-term dependencies in sequential data processing, effectively overcoming the limitations of traditional RNNs in handling long-range dependencies.

[0099] In drought risk identification and early warning, NDVI (Normalized Difference Vegetation Index) is a key indicator reflecting vegetation health. As drought occurs, NDVI values ​​gradually decrease. LSTM can learn the dynamic changes in drought risk based on time-series NDVI data and model gridded precipitation forecast data, predicting future drought risk trends and changes in vegetation status.

[0100] In drought risk identification and early warning, LSTM models can be used to predict NDVI value changes over the next few days to a month. The construction of an LSTM model typically includes: an input layer: receiving gridded precipitation forecast data and historical time-series NDVI data, using a sliding window to apply data from several past time steps as input features; an LSTM layer: the core of the model, learning long-term dependencies in the time-series data through a gating mechanism to capture drought trends and seasonal variations; a fully connected layer (Dense Layer): usually following the LSTM layer, a fully connected layer is used to output prediction results, such as future NDVI values ​​and the probability of drought occurrence; and an output layer: the output layer generates the model's final prediction, which may be continuous NDVI values ​​(for regression tasks) or drought prediction classification results (such as whether drought occurs, drought level, etc.).

[0101] The embodiments of the present invention provide, as follows Figure 2 The diagram illustrates a specific processing flow for model building, including: (a) reading the NDVI data required for model building; (b) preprocessing the NDVI data to remove interfering data (such as clouds) and outliers to ensure the smoothness of time series data; (c) extracting and analyzing the temporal characteristics of NDVI; (d) preprocessing precipitation data, including spatial and temporal interpolation; (e) extracting features from precipitation forecast data; (f) training and optimizing the model; and (g) outputting relevant data and parameter information of the early warning model.

[0102] In model training, the choice of loss function depends on the type of task: for regression tasks, mean squared error (MSE) is usually used as the loss function, while for classification tasks, cross-entropy loss function is used. In terms of optimizers, Adam optimizer is often used in machine learning models such as LSTM due to its adaptive learning rate characteristics. During training, the dataset is divided into training and validation sets. The training set is used to optimize model parameters, while the validation set is used to evaluate the generalization ability of the model to ensure the model's performance on unseen data.

[0103] The embodiment of the present application provides a specific process of machine learning network training: using historical precipitation data and NDVI time series data for model training, and repeatedly optimizing the model to prepare for predicting drought risk. In drought risk identification and early warning, the core goal of training and optimizing the model is to use historical precipitation data and NDVI time series data to build a model that can accurately identify and predict drought risk. The training and optimization process includes data preparation, model selection, training strategy, and continuous adjustment and optimization of model parameters to enable the model to identify and accurately classify different levels of drought risk. Specifically:

[0104] (I) Data preparation and preprocessing:

[0105] 1) Data annotation: Training an efficient model first relies on high-quality training data. For drought prediction tasks, historical NDVI time series data and annotated drought event data (such as drought time, severity, etc.) are usually required. For each time point or time period, the corresponding drought level needs to be marked. Drought levels can be defined according to standard drought classification indicators (such as standardized precipitation index (SPI) or precipitation evapotranspiration ratio (PET / P)) and divided into different levels (such as mild drought, severe drought, etc.).

[0106] 2) Training strategy: To train a model that can effectively identify different levels of drought risk, the following steps are usually taken: divide the training set and the validation set. Training set: used for model training, usually accounting for 80% of the dataset. Validation set: used to evaluate model performance during training, usually accounting for 20% of the dataset. Test set: an independent test set used after training to evaluate the generalization ability of the final model.

[0107] 3) Feature engineering and input design: Time series features: models such as LSTM need to learn through time series data, so a reasonable input format needs to be designed, usually taking the past several time steps of data as input features. The sliding window method can be used to input the NDVI values in each time window, combined with the precipitation forecast data of the grid point to predict the drought level at a certain time in the future. In order to increase the robustness and generalization ability of the algorithm, historical precipitation forecast data is used for training as much as possible.

[0108] (II) Model Training: Adjust the weights of the model through backpropagation and gradient descent algorithms. 2) Prevent overfitting by techniques such as cross-validation or Early Stopping, ensuring that its performance on the validation set does not decrease. 3) For LSTM models, multiple hyperparameters may need to be adjusted during training, such as the number of LSTM layers, the number of hidden units, the learning rate, the batch size, etc.

[0109] (III) Optimization and Adjustment: The optimization phase aims to improve the performance of the model through a series of methods to accurately identify different levels of drought risk. The main optimization methods include:

[0110] (III-1) Hyperparameter Adjustment: Learning Rate: Learning rate controls the update amplitude of each step, too large will lead to unstable convergence, too small will lead to slow convergence. Learning rate decay or adaptive learning rate (such as Adam optimizer) can be used to dynamically adjust the learning rate. Batch Size: Batch size affects the frequency of parameter updates, larger batch size may lead to slower convergence, but higher computational efficiency; smaller batch size can find better solutions faster. LSTM layers and hidden units: Increasing the number of LSTM layers and hidden units can enhance the expression ability of the model, but may also increase the training time and complexity, which needs to be balanced.

[0111] (III-2) Selection of Loss Function: For drought level prediction (classification problem), cross-entropy loss function can be used. For regression problems (for example, predicting the specific value of NDVI), mean square error (MSE) or mean absolute error (MAE) can be used as loss function.

[0112] (III-3) Model Regularization: Dropout: Prevent overfitting by randomly dropping some neurons during training. L2 regularization: Add a penalty term to the model parameters to prevent the model from overfitting the training data.

[0113] (III-4) Model Ensemble: Through ensemble learning methods such as Bagging or Boosting, combine the prediction results of multiple models to improve accuracy and robustness.

[0114] (IV) Evaluation and Validation: Evaluating the performance of the model is an important part of the optimization process. Common evaluation metrics include: Accuracy: In classification problems, the ratio of correctly predicted drought levels. F1-score: Combines precision and recall, especially important in cases of class imbalance. Confusion matrix: Used to view the distribution of predicted results for each class, to understand where the model performs poorly. AUC-ROC (Area Under the Receiver Operating Characteristic Curve) curve: used to evaluate the classification effect of the model.

[0115] After training and optimization, the final model can be deployed in real-time data systems to receive real-time NDVI data and predict future drought risks through the model. For example, in the agricultural field, drought warnings can be provided based on the model's prediction results, and relevant drought response measures can be suggested. Anti-drought strategies such as increasing irrigation plans can be added to the prediction model to predict the trend of vegetation index.

[0116] Through historical precipitation data and NDVI time series data, combined with the training and optimization of machine learning models such as LSTM, an accurate drought risk identification and prediction system can be established. During the training process of the model, through data preprocessing, feature engineering, hyperparameter adjustment, regularization and model evaluation, etc., the accuracy and generalization ability of the model are continuously optimized. Finally, the model can effectively identify different degrees of drought risk, providing support for drought warning and response decision-making. According to the exposure, vulnerability and resilience of the bearing body output by the model, drought risk prediction is formed.

[0117] Based on the foregoing embodiments, the trained LSTM model is used to input historical NDVI data and gridded precipitation forecast data to predict NDVI changes in the future period, analyze the vulnerability and resilience of the disaster-bearing body, and realize drought risk identification and dynamic adjustment of irrigation strategy. For example, see Figure 3 The specific processing process is as follows: (a) read the NDVI data required for drought risk identification and early warning; (b) preprocess the NDVI data to remove interference data (such as clouds) and outliers, and ensure the smoothness of the time series data; (c) precipitation data preprocessing, including spatial and temporal interpolation processing, etc.; (d) input into the early warning model for drought risk prediction; (e) anti-drought assessment based on irrigation strategy, etc.; (f) output drought risk warning and anti-drought strategy related products and information.

[0118] In one application scenario, the drought risk level of the study area is evaluated by using the hazard body NDVI prediction value, which specifically includes the following processes: the seasonal characteristics contained in the NDVI time series characteristics are used to eliminate the influence of seasonal factors on the hazard body NDVI prediction value, and a new hazard body NDVI prediction value is obtained; based on the new hazard body NDVI prediction value, the NDVI prediction rate characteristics corresponding to the study area are extracted; according to the change trend and positive and negative of the NDVI prediction rate characteristics, the initial drought risk level corresponding to the study area is determined; the precipitation forecast data is used as an auxiliary factor to correct the initial drought risk level, and the target drought risk level corresponding to the study area is obtained.

[0119] For example, if the prediction result shows that the NDVI value continues to decrease, it may indicate that the drought risk is increasing. Based on the NDVI prediction value output by the model, drought classification can also be performed by setting a threshold. For example, when the NDVI prediction value is lower than a certain set threshold, it can be determined that the region is in a drought state. In order to evaluate the performance of the model, indicators such as mean square error, mean absolute error or accuracy can be used for verification.

[0120] In one specific embodiment, the NDVI change rate is an important dynamic indicator of drought risk, directly reflecting the change trend of the vegetation health condition. If the predicted NDVI change rate is negative (positive), it indicates that the vegetation condition may deteriorate (improve), and the drought risk increases (decreases); the greater the negative (positive) value of the rate, the higher (lower) the risk. If the NDVI change rate is positive but lower than the historical average, it may indicate a potential drought risk, because the improvement speed of the vegetation is not enough to offset the adverse environmental impact. The precipitation in the next 1-30 days is an important driving factor affecting the NDVI change. The less the precipitation, the more restricted the vegetation growth, and the higher the possibility of NDVI decrease. In the case of insufficient precipitation, even if the NDVI change rate is currently positive, it may turn negative due to continuous water shortage, and the drought risk further increases. If the precipitation is abundant, it may have a positive regulating effect on the NDVI change, thereby reducing the drought risk. For the comprehensive prediction of drought risk, the NDVI change rate in the next 30 days is the core indicator for judging the drought risk, directly reflecting the future health condition of the vegetation. The precipitation in the next 30 days is used as an auxiliary factor, and together with the NDVI change, it constitutes the prediction system of the drought risk. When the NDVI prediction rate is negative and the future precipitation is insufficient, the drought risk significantly increases; when the NDVI prediction rate is close to stable or positive, but the precipitation is insufficient, potential risks still need to be vigilant; and when the precipitation is sufficient and the NDVI rate is positive, the drought risk is low.

[0121] The LSTM model has multiple advantages. It can capture long-term trends and seasonal changes in NDVI data, making it well-suited for handling long-term natural phenomena such as drought. At the same time, LSTM is good at learning non-linear relationships in data, allowing it to accurately model the dynamic changes in drought risk. In addition, it can automatically learn features from time series data without the need for manual selection or design of features, simplifying the complexity of feature engineering. By analyzing the change patterns of historical NDVI data, the LSTM model can not only predict future trends in drought development, but also learn the non-linear relationships between vegetation vulnerability, exposure, and resilience, directly providing future trends in vegetation indices for analyzing vegetation recovery after drought and predicting the speed of vegetation recovery, further providing scientific basis for agricultural and ecological management.

[0122] In another application scenario, the anti-drought irrigation strategy data is dynamically adjusted using the predicted NDVI values of the hazard-affected body. The anti-drought irrigation strategy data is used to describe the irrigation points deployed within the study area and their corresponding irrigation volume and irrigation time. The specific process includes the following steps: through the irrigation strategy evaluation function, based on the predicted NDVI values of the hazard-affected body, the evaluation value corresponding to the anti-drought irrigation strategy data is determined, wherein the irrigation strategy evaluation function is a function that aims to minimize irrigation cost and maximize drought risk reduction. Based on the evaluation value, the anti-drought irrigation strategy is adjusted; through the machine learning model, based on the hazard-affected body feature information, water resource supply characteristics and new anti-drought irrigation strategy characteristics, the new predicted NDVI values of the hazard-affected body in the future time of the study area are predicted until the preset stopping condition is met, and the optimal anti-drought irrigation strategy data is obtained.

[0123] The drought risk assessment model based on NDVI and future precipitation is combined with anti-drought strategies such as irrigation to optimize the results, thereby improving the practicality and decision support capability of the model. In the prediction model, irrigation strategy is introduced as an adjustment factor to simulate the positive impact of irrigation on NDVI. Irrigation can directly improve the health status of vegetation and alleviate the rate of NDVI decline, thereby reducing drought risk. Specifically, the relationship between irrigation volume and the improvement range of NDVI can be quantified through the model, for example, by using historical data to fit the influence function of irrigation on NDVI. By combining precipitation prediction, NDVI time series features and anti-drought strategies, an integrated prediction and optimization drought risk assessment system can be constructed. This not only improves the accuracy of risk warning, but also provides scientific basis for agricultural management, effectively reducing the adverse effects of drought.

[0124] In developing irrigation strategies to combat drought risk, based on existing irrigation facilities, combined with the regional daily water supply capacity as the initial scheme for modeling. The core of the irrigation strategy is to establish a functional relationship between the irrigation scheme and the reduction of risk, that is, to analyze the interaction between irrigation water, time distribution and drought risk and its influence mechanism. Specifically, the distribution of irrigation facilities determines the spatial allocation of water resources, and the daily water supply as a constraint condition provides a preliminary optimization range for the model. On this basis, through historical data and experimental observation, the marginal benefit of unit irrigation water on the improvement of NDVI is quantified, for example, for different types of vegetation, the NDVI response may have a nonlinear relationship. In order to establish this functional relationship, mathematical modeling tools are introduced to simulate the dynamic changes of irrigation amount (x) and vegetation state (y) as a set of control equations, which considers the coupling effect of precipitation, evapotranspiration, soil moisture and other environmental factors. The change trend of NDVI under certain irrigation conditions can be deduced, so as to quantify the mitigation effect of irrigation on drought risk. Further, in order to achieve optimal decision-making, a target function can be used to balance the trade-off between irrigation investment and risk reduction, for example, to achieve the goal of minimizing drought risk with the least irrigation cost. Finally, through dynamic adjustment and feedback mechanism, real-time monitoring data is integrated into the model, which continuously corrects the irrigation scheme to make it more in line with actual demand, thereby providing scientific and effective decision support for drought risk management.

[0125] In summary, the core of the embodiment of the present application is: drought risk assessment based on NDVI time series dynamic characteristics and grid point precipitation forecast data: using the trend, fluctuation and change rate of NDVI and precipitation forecast data, the time series feature set for drought risk identification is directly constructed; application of machine learning model: combined with LSTM neural network model, focusing on dynamic feature analysis of time series data, capturing the time series change of NDVI, accurately identifying the occurrence and future trend of drought risk.

[0126] The embodiment of the present application focuses on the basic evolution and abnormal fluctuation characteristics of NDVI (normalized vegetation index) time series data based on multi-source satellite fusion, combined with 30-day grid point precipitation forecast data, and uses machine learning model for drought risk identification and early warning. Using NDVI and grid point precipitation time series data, combined with seasonal-trend decomposition method (STL) and long short-term memory network (LSTM) model, NDVI time series features and short-term abnormal fluctuation features, NDVI change rate, 30-day future precipitation and drought risk level are calculated and the corresponding relationship is obtained. A model specially used for early drought risk identification and early warning is constructed, which is used to produce high spatio-temporal resolution drought risk early warning products for 1-30 days, providing early warning services for agricultural production, drought resistance water resource scheduling, drought disaster emergency, ecological environment protection and repair.

[0127] On the basis of the foregoing embodiments, the embodiment of the present application provides a drought risk identification and early warning device based on multi-source satellite fused NDVI and machine learning, referring to Figure 4 As shown in a structural schematic diagram of a drought risk identification and early warning device based on multi-source satellite fused NDVI and machine learning, the device mainly comprises the following parts:

[0128] The disaster-bearing body feature extraction module 402 is configured to acquire NDVI time series data corresponding to the research area, and extract disaster-bearing body feature information corresponding to the research area based on the NDVI time series data, wherein the disaster-bearing body feature information comprises one or more of NDVI time series features, NDVI fluctuation features and NDVI change rate features;

[0129] The water resource supply feature extraction module 404 is configured to acquire water resource supply data corresponding to the research area, and extract water resource supply features corresponding to the water resource supply data, wherein the water resource supply data comprises precipitation forecast data, or comprises precipitation forecast data and anti-drought irrigation strategy data;

[0130] The prediction module 406 is configured to predict a disaster-bearing body NDVI prediction value of the research area at a future time based on the disaster-bearing body feature information and the water resource supply features through a machine learning model obtained by pre-training.

[0131] The disaster-bearing body NDVI prediction value is used to identify the drought risk of the research area in combination with the precipitation forecast data in the case where the water resource supply data only comprises the precipitation forecast data, and is used to dynamically adjust the anti-drought irrigation strategy data in the case where the water resource supply data comprises the precipitation forecast data and the anti-drought irrigation strategy data.

[0132] The drought risk identification and early warning device based on multi-source satellite fused NDVI and machine learning provided by the embodiment of the present application automatically identifies and evaluates the drought risk through specific NDVI time series features and a machine learning model, and realizes real-time identification and hierarchical early warning of drought risks of different climate zones and ecological systems based on the growth and health conditions of vegetation, and can also dynamically adjust the anti-drought irrigation strategy data. Compared with the traditional drought risk identification and early warning method, the method has significant advantages in NDVI time series feature extraction, drought risk monitoring and identification, risk level division, prediction accuracy and strategy adjustment effect, and is particularly suitable for real-time dynamic identification and early warning of drought risks of regions containing various different ecological systems (such as forests, grasslands, farmlands, deserts and wetlands).

[0133] In one embodiment, the disaster-bearing body feature extraction module 402 is specifically configured to:

[0134] The NDVI time series data is subjected to a local weighted regression processing to extract NDVI time series features of the NDVI time series data, the NDVI time series features including a long-term change trend feature, a seasonal feature and a residual feature;

[0135] In addition, a short-term abnormal fluctuation threshold is determined based on the residual feature, and an NDVI fluctuation feature of the NDVI time series data is extracted according to the residual feature and the short-term abnormal fluctuation threshold, the fluctuation feature being used to describe a short-term abnormal fluctuation of the NDVI time series data;

[0136] In addition, a sliding window is set for the NDVI time series data, and a change rate corresponding to the sliding window is determined based on an NDVI value corresponding to each data point located in the sliding window, the sliding window is continuously moved until the sliding window moves to the end of the NDVI time series data, and an NDVI change rate feature corresponding to the NDVI time series data is extracted based on the change rates corresponding to all the sliding windows, the NDVI change rate feature being used to reflect a disaster-bearing body vulnerability and a disaster-bearing body resilience.

[0137] In an embodiment, the water resource supply feature extraction module 404 is specifically configured to:

[0138] extract a cumulative precipitation feature, a precipitation anomaly feature and a precipitation fluctuation feature corresponding to the precipitation forecast data;

[0139] The cumulative precipitation feature is used to represent a water resource supply capacity of the research region in a short term, the precipitation anomaly feature is used to reflect an abnormal degree of precipitation in the research region, and the precipitation fluctuation feature is used to describe a uniformity of precipitation distribution in the research region.

[0140] In an embodiment, the drought grade evaluation module is further configured to:

[0141] The seasonal feature contained in the NDVI time series feature is used to eliminate the influence of seasonal factors on the disaster-bearing body NDVI prediction value, so as to obtain a new disaster-bearing body NDVI prediction value;

[0142] Based on the new disaster-bearing body NDVI prediction value, an NDVI prediction rate feature corresponding to the research region is extracted;

[0143] According to a change trend and a positive or negative nature of the NDVI prediction rate feature, an initial drought risk grade corresponding to the research region is determined;

[0144] The precipitation forecast data is used as an auxiliary factor to correct the initial drought risk grade, so as to obtain a target drought risk grade corresponding to the research region.

[0145] In an embodiment, the drought grade evaluation module is further configured to:

[0146] The machine learning model obtained by pre-training is used to predict the NDVI prediction value of the disaster-bearing body and the target drought risk level of the research area at a future time based on the disaster-bearing body feature information and the water resource supply feature.

[0147] In an embodiment, the anti-drought irrigation strategy data is used to describe the irrigation points deployed in the research area and the corresponding irrigation amount and irrigation time thereof;The irrigation strategy adjustment module is further included, which is used to:

[0148] An evaluation value corresponding to the anti-drought irrigation strategy data is determined by the irrigation strategy evaluation function based on the NDVI prediction value of the disaster-bearing body;

[0149] The anti-drought irrigation strategy is adjusted based on the evaluation value.

[0150] The new NDVI prediction value of the disaster-bearing body in the research area at a future time is predicted by the machine learning model based on the disaster-bearing body feature information, the water resource supply feature and the new anti-drought irrigation strategy feature until the preset stopping condition is met, so as to obtain the optimal anti-drought irrigation strategy data.

[0151] In an embodiment, the irrigation strategy evaluation function is a function with the minimum irrigation cost and the maximum reduction of drought risk as the target.

[0152] The device provided by the embodiment of the present application has the same implementation principle and technical effects as the foregoing method embodiments, and for the sake of brevity, the part not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments.

[0153] The electronic device provided by the embodiment of the present application comprises a processor and a storage device;The storage device stores a computer program, and the computer program executes the method of any one of the foregoing embodiments when being run by the processor.

[0154] Figure 5 The structure schematic diagram of the electronic device provided by the embodiment of the present application, the electronic device 100 comprises a processor 50, a memory 51, a bus 52 and a communication interface 53, the processor 50, the communication interface 53 and the memory 51 are connected through the bus 52;The processor 50 is used to execute the executable module stored in the memory 51, such as a computer program.

[0155] The memory 51 can include a high-speed random access memory (RAM, Random Access Memory) and can also include a non-volatile memory such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used.

[0156] The bus 52 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the figure to represent the bus, but it does not mean that there is only one bus or only one type of bus.

[0157] The memory 51 is used to store programs, and the processor 50 executes the programs after receiving execution instructions. The method executed by the device defined by the flow process disclosed in any of the embodiments of the application can be applied to the processor 50 or implemented by the processor 50.

[0158] The processor 50 can be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method can be completed by the integrated logic circuit of hardware in the processor 50 or the instruction in the form of software. The processor 50 described above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. It can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the application. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is mature in the art. The storage medium is located in the memory 51, and the processor 50 reads the information in the memory 51 and combines the hardware to complete the steps of the above method.

[0159] The computer program product of the readable storage medium provided by the embodiment of the present application comprises a computer readable storage medium storing program codes, and the program codes comprise instructions for executing the method described in the foregoing method embodiments. The specific implementation can be referred to the foregoing method embodiments, and will not be described here.

[0160] When the functions are realized in the form of software function units and sold or used as independent products, the functions can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the technical solutions that essentially contribute to the prior art can be embodied in the form of software products. The computer software product is stored in a storage medium and includes instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0161] Finally, it should be noted that: the above-described embodiments are merely specific implementations of the present application, which are used to illustrate the technical solutions of the present application, but not to limit the present application. The protection scope of the present application is not limited to this. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features, within the technical range disclosed by the present application. The modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A drought risk identification and early warning method based on multi-source satellite fused NDVI and machine learning, characterized in that, The method comprises: obtaining NDVI time series data corresponding to a research area, and extracting disaster-bearing body feature information corresponding to the research area based on the NDVI time series data, wherein the disaster-bearing body feature information comprises NDVI time series features, NDVI fluctuation features and NDVI change rate features, the NDVI time series features comprise long-term change trend features, seasonal features and residual features, the fluctuation features are used to describe short-term abnormal fluctuation of the NDVI time series data, and the NDVI change rate features are used to reflect disaster-bearing body vulnerability and disaster-bearing body resilience; obtaining water resource supply data corresponding to the research area, and extracting water resource supply features corresponding to the water resource supply data, wherein the water resource supply data comprises precipitation forecast data, or comprises precipitation forecast data and drought-resistant irrigation strategy data; predicting a disaster-bearing body NDVI prediction value of the research area at a future time based on the disaster-bearing body feature information and the water resource supply features by using a machine learning model obtained through pre-training; wherein the disaster-bearing body NDVI prediction value is used to identify drought risk of the research area in combination with the precipitation forecast data in the case that the water resource supply data only comprises the precipitation forecast data, and to dynamically adjust the drought-resistant irrigation strategy data in the case that the water resource supply data comprises the precipitation forecast data and the drought-resistant irrigation strategy data; the drought-resistant irrigation strategy data is used to describe irrigation points deployed in the research area and corresponding irrigation amount and irrigation time; the method further comprises: determining an evaluation value corresponding to the drought-resistant irrigation strategy data based on the disaster-bearing body NDVI prediction value by using an irrigation strategy evaluation function, the irrigation strategy evaluation function being a function with the objective of minimizing irrigation cost and maximizing drought risk reduction; adjusting the drought-resistant irrigation strategy based on the evaluation value; predicting a new disaster-bearing body NDVI prediction value of the research area at the future time based on the disaster-bearing body feature information, the water resource supply features and new drought-resistant irrigation strategy data by using the machine learning model, until a preset stop condition is met, to obtain optimal drought-resistant irrigation strategy data.

2. The drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning according to claim 1, characterized in that, The method comprises: performing local weighted regression processing on the NDVI time series data to extract NDVI time series features of the NDVI time series data; and determining a short-term abnormal fluctuation threshold based on the residual features, and extracting NDVI fluctuation features of the NDVI time series data according to the residual features and the short-term abnormal fluctuation threshold. and, setting a sliding window for the NDVI time series data, and determining a change rate corresponding to the sliding window based on the NDVI value corresponding to each data point within the sliding window, continuing to move the sliding window until the sliding window stops moving when it moves to the end of the NDVI time series data, and extracting the NDVI change rate feature corresponding to the NDVI time series data based on the change rates corresponding to all the sliding windows.

3. The drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning according to claim 1, characterized in that, extracting the water resource supply feature corresponding to the water resource supply data, including: extracting the cumulative precipitation feature, the precipitation anomaly feature and the precipitation fluctuation feature corresponding to the precipitation forecast data; wherein the cumulative precipitation feature is used to represent the water resource supply capacity of the study area in the short term, the precipitation anomaly feature is used to reflect the abnormal degree of precipitation in the study area, and the precipitation fluctuation feature is used to describe the uniformity of the precipitation distribution in the study area.

4. The drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning according to claim 1, characterized in that, The method further comprises: using the seasonal feature contained in the NDVI time series feature to eliminate the influence of seasonal factors on the disaster-bearing body NDVI prediction value, to obtain a new disaster-bearing body NDVI prediction value; extracting the NDVI prediction rate feature corresponding to the study area based on the new disaster-bearing body NDVI prediction value; determining the initial drought risk level corresponding to the study area according to the change trend and the positive and negative nature of the NDVI prediction rate feature; using the precipitation forecast data as an auxiliary factor to correct the initial drought risk level to obtain the target drought risk level corresponding to the study area.

5. The drought risk identification and early warning method based on multi-source satellite fusion NDVI and machine learning according to claim 1, characterized in that, The method further comprises: using the machine learning model obtained by pre-training to predict the disaster-bearing body NDVI prediction value and the target drought risk level of the study area at a future time based on the disaster-bearing body feature information and the water resource supply feature.

6. A device for identifying and warning drought risk based on multi-source satellite fused NDVI and machine learning, characterized in that, including: a disaster-bearing body feature extraction module for obtaining NDVI time series data corresponding to a study area, and extracting disaster-bearing body feature information corresponding to the study area based on the NDVI time series data, the disaster-bearing body feature information including NDVI time series features, NDVI fluctuation features and NDVI change rate features, the NDVI time series features including long-term change trend features, seasonal features and residual features, the fluctuation features being used to describe short-term abnormal fluctuations of the NDVI time series data, and the NDVI change rate features being used to reflect the vulnerability and resilience of the disaster-bearing body; a water resource supply feature extraction module for obtaining water resource supply data corresponding to the study area, and extracting water resource supply features corresponding to the water resource supply data, the water resource supply data including precipitation forecast data, or including precipitation forecast data and drought-resistant irrigation strategy data; a prediction module for using a machine learning model obtained by pre-training to predict the disaster-bearing body NDVI prediction value of the study area at a future time based on the disaster-bearing body feature information and the water resource supply feature. The hazard-bearing body NDVI prediction value is used to identify the drought risk of the research area in combination with the precipitation forecast data when the water resource supply data only includes the precipitation forecast data, and to dynamically adjust the drought-resistant irrigation strategy data when the water resource supply data includes the precipitation forecast data and the drought-resistant irrigation strategy data. The drought-resistant irrigation strategy data is used to describe the irrigation points deployed in the research area and the corresponding irrigation amount and irrigation time thereof. The irrigation strategy adjustment module is further included, which is configured to determine an evaluation value corresponding to the drought-resistant irrigation strategy data based on the hazard-bearing body NDVI prediction value through an irrigation strategy evaluation function, the irrigation strategy evaluation function being a function with the minimum irrigation cost and the maximum reduction of drought risk as the target, adjust the drought-resistant irrigation strategy based on the evaluation value, and predict a new hazard-bearing body NDVI prediction value of the research area at the future time based on the hazard-bearing body feature information, the water resource supply feature and the new drought-resistant irrigation strategy data through the machine learning model until a preset stop condition is met, so as to obtain the optimal drought-resistant irrigation strategy data.

7. An electronic device, comprising: A processor and a memory are included, the memory storing computer executable instructions capable of being executed by the processor, and the processor executes the computer executable instructions to implement the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and the computer executable instructions, when called and executed by a processor, cause the processor to implement the method of any one of claims 1 to 5.

Images