Malaria occurrence risk prediction method and device

Abstract

The invention discloses a malaria occurrence risk prediction method and device, and relates to the technical field of disease risk prediction, and the method comprises the steps: determining a plurality of ground investigation points in a to-be-predicted region, the ground investigation points being used for representing whether a malaria case has occurred; according to the plurality of ground survey points, extracting a time sequence of multi-source environmental factors for driving malaria occurrence from the multi-source earth observation data; analyzing the time sequence of the multi-source environmental factors to obtain an optimal time lag period of the multi-source environmental factors; and constructing a training data set according to the optimal time delay period of the multi-source environmental factors, and inputting the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model. According to the method, the time sequence of the multi-source environmental factors is extracted, and model training is carried out by using the optimal time lag stage, so that the malaria occurrence risk prediction model can fully consider the lag effect and comprehensively capture the complex relation between the environmental factors and malaria, thereby realizing high-precision prediction of the malaria risk.

Description

Technical Field

[0001] The present application relates to the technical field of disease risk prediction, and in particular to a method and device for predicting the risk of malaria occurrence. Background Art

[0002] Malaria is an infectious disease spread by mosquitoes. Its prevention and control work faces many challenges, especially in some areas where medical resources are relatively scarce. Although global malaria prevention and control has made some progress in recent years, it is difficult to prevent and control due to its complex transmission mechanism and the interweaving of multiple factors.

[0003] With the continuous development of remote sensing technology, researchers have widely applied remote sensing technology to investigate malaria mosquitoes and disease ecology. Remote sensing technology collects data through satellite sensors, covering the spatial variation of meteorological and vegetation variables, which is crucial for the study of malaria transmission. Meteorological factors such as temperature, humidity and rainfall, as well as the growth of vegetation, directly affect the survival and reproduction of malaria mosquitoes. Remote sensing technology can cover a large geographical area in a short period of time and provide high-resolution data, which is of great significance for the monitoring and prediction of malaria epidemics.

[0004] At present, among the prediction methods of malaria transmission, statistical models, mathematical models and machine learning each have their own advantages and disadvantages. Statistical models such as generalized linear models and ARIMA can analyze and predict the trend of malaria incidence based on historical data. Mathematical models such as the Ross-Macdonald model and the SIR model can reveal the transmission mechanism and influencing factors of malaria in more depth. Machine learning methods, such as support vector machines and random forests, can handle more complex data relationships and improve the accuracy and stability of predictions. However, the existing malaria transmission test methods are still unable to achieve high-precision prediction of malaria risk. Summary of the invention

[0005] In view of this, the main purpose of the present application is to provide a method and device for predicting the risk of malaria, with the aim of achieving high-precision prediction of the risk of malaria.

[0006] The first aspect of the present application provides a method for predicting the risk of malaria occurrence, the method comprising:

[0007] Determine multiple ground survey points in the area to be predicted, and the ground survey points are used to characterize whether malaria cases have occurred;

[0008] Extract the time series of multi-source environmental factors driving malaria from multi-source earth observation data based on multiple ground survey points;

[0009] Analyze the time series of multi-source environmental factors to obtain the optimal time lag of multi-source environmental factors;

[0010] A training data set is constructed according to the optimal time lag of multi-source environmental factors, and the training data set is input into a supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0011] In some implementations of the first aspect of the present application, the ground survey points include: points where malaria cases occurred and points where malaria cases did not occur. After determining the plurality of ground survey points, the method further includes:

[0012] Dividing the area to be predicted into multiple grids of preset spatial resolution;

[0013] The malaria-free points whose distances from the malaria-free points to the malaria-free points are less than the preset spatial resolution are screened out, so that the malaria-free points and the malaria-free points belong to different grids.

[0014] In some implementations of the first aspect of the present application, the method further includes:

[0015] A plurality of points where malaria cases have not occurred are randomly generated outside a preset value of the distance from the malaria case occurrence point, wherein the preset value is greater than a preset spatial resolution.

[0016] In some implementations of the first aspect of the present application, a time series of multi-source environmental factors driving the occurrence of malaria is extracted from multi-source earth observation data according to multiple ground survey points, including:

[0017] According to the latitude and longitude coordinates of multiple ground survey points, environmental variables are extracted from multi-source earth observation data for preprocessing to obtain preprocessed environmental variables;

[0018] The time series of multi-source environmental factors driving malaria occurrence are extracted from the pre-processed environmental variables.

[0019] In some implementations of the first aspect of the present application, the environmental variables include: remote sensing images, and the preprocessing includes:

[0020] The remote sensing images are declouded and the missing pixels after declouding are filled by using multi-temporal image synthesis technology and / or inverse distance weighted spatial interpolation technology.

[0021] In some implementations of the first aspect of the present application, the preprocessing includes:

[0022] Taking the temporal frequency and spatial resolution of the ground survey points as the reference standard, the environmental variables are resampled to make them have the same temporal frequency and spatial resolution as the ground survey points.

[0023] In some implementations of the first aspect of the present application, the multi-source environmental factors include dynamic environmental factors and static environmental factors, wherein the time window length of the time series of the dynamic environmental factors is shorter than that of the static environmental factors.

[0024] In some implementations of the first aspect of the present application, analyzing the time series of multi-source environmental factors includes:

[0025] Spearman correlation analysis was performed on the time series of multi-source environmental factors.

[0026] In some implementations of the first aspect of the present application, the dynamic environmental factors include: precipitation, normalized difference vegetation index, enhanced vegetation index and surface temperature, and the static environmental factors include: land use / cover, elevation DEM and travel time.

[0027] The second aspect of the present application provides a device for predicting the risk of malaria occurrence, the device comprising:

[0028] A survey point determination module is used to determine multiple ground survey points in the area to be predicted, and the ground survey points are used to characterize whether malaria cases have occurred;

[0029] The time series extraction module is used to extract the time series of multi-source environmental factors driving malaria from multi-source earth observation data based on multiple ground survey points;

[0030] The time series analysis module is used to analyze the time series of multi-source environmental factors to obtain the optimal time lag of multi-source environmental factors;

[0031] The model training module is used to construct a training data set according to the optimal time lag of multi-source environmental factors, and input the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0032] The technical solution provided by this application has the following beneficial effects:

[0033] In the technical solution proposed in this application, a plurality of ground survey points in the area to be predicted are first determined to mark whether there have been cases of malaria. Then, according to the locations of these ground survey points, the time series of various environmental factors closely related to the occurrence of malaria are extracted from multi-source earth observation data. These environmental factors are key factors in the spread of malaria, and their time series can reveal the seasonal fluctuations and long-term trends of environmental factors. Afterwards, in view of the possible time lag effect between environmental factors and the occurrence of malaria, that is, the changes in certain environmental factors may only show their impact on the occurrence of malaria after a period of time, this application further analyzes the time series of multi-source environmental factors to determine the optimal time lag between each environmental factor and the occurrence of malaria, thereby more accurately grasping the time association between environmental factors and malaria. Finally, a training data set is constructed based on the determined optimal time lag, and it is input into a supervised learning model for training. The model will learn the complex relationship between environmental factors and the occurrence of malaria, and then form a malaria occurrence risk prediction model. The trained malaria occurrence risk prediction model can use the new prediction data set to accurately predict the risk of malaria in the predicted area. In general, this application extracts time series of multi-source environmental factors and uses the optimal time lag period for model training, so that the prediction model of malaria risk can fully consider the lag effect and comprehensively capture the complex relationship between environmental factors and malaria, thereby achieving high-precision prediction of malaria risk. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a process for predicting the risk of malaria occurrence provided in an embodiment of the present application;

[0035] Figure 2 A schematic diagram of a flow chart of another method for predicting the risk of malaria occurrence provided in an embodiment of the present application;

[0036] Figure 3 A schematic diagram of a flow chart of another method for predicting the risk of malaria occurrence provided in an embodiment of the present application;

[0037] Figure 4 A schematic diagram of a flow chart of another method for predicting the risk of malaria occurrence provided in an embodiment of the present application;

[0038] Figure 5 A schematic diagram of a flow chart of another method for predicting the risk of malaria occurrence provided in an embodiment of the present application;

[0039] Figure 6 A schematic diagram of the overall technical route provided for the embodiments of the present application;

[0040] Figure 7 A schematic diagram of a region to be predicted provided in an embodiment of the present application;

[0041] Figure 8 A schematic diagram of the results of the correlation analysis of the time-lagged variables of the dynamic environmental factors that affect the occurrence of malaria provided in the embodiments of the present application;

[0042] Fig. 9 A schematic diagram of the probability of malaria occurrence risk in a to-be-predicted area and ground survey results provided in an embodiment of the present application;

[0043] Fig.10 A schematic diagram of another malaria occurrence risk probability in a to-be-predicted area and ground survey results provided in an embodiment of the present application;

[0044] Fig.11 A schematic diagram of the structure of a malaria risk prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] Glossary:

[0047] Optimal lag response time: Optimal lag response time is used in time series analysis to determine how the change of a variable at a certain point in the past has the best impact on the current or future target variable. It is used to find the lag time interval that can make the model work best when dealing with lag effects.

[0048] The embodiments of the present application provide a method and device for predicting the risk of malaria occurrence.

[0049] See also Figure 1 As shown, the embodiment of the present application provides a method for predicting the risk of malaria occurrence, comprising the following steps:

[0050] S101: Determine a plurality of ground survey points in the area to be predicted, where the ground survey points are used to characterize whether malaria cases have occurred.

[0051] In an embodiment of the present application, a plurality of ground survey points are determined in the area to be predicted, each ground survey point represents a geographical location of a known state, and is used to identify whether a malaria case has occurred at the location. In addition, the ground survey point is used as a label for subsequent supervised learning model training, where a malaria case has occurred is a positive sample for model training, and a malaria case that has not occurred is a negative sample.

[0052] S102: Extract the time series of multi-source environmental factors driving the occurrence of malaria from multi-source earth observation data based on multiple ground survey points.

[0053] In an embodiment of the present application, multi-source environmental factors refer to a variety of environmental factors that drive the occurrence of malaria, and the multi-source environmental factors are derived from multi-source earth observation data. Multi-source earth observation data refers to various types of environmental data related to the earth's surface and atmosphere collected by remote sensing technology, which may include remote sensing data, meteorological data, geographic data and other data types. The time series of multi-source environmental factors refers to a data set of multiple environmental factors arranged in chronological order within a period of time, and each environmental factor has a corresponding value at each time point; it should be noted that each environmental factor may have a different time update frequency.

[0054] By extracting the time series of environmental factors, the relationship between multi-source environmental factors and the occurrence of malaria can be established. The time series of multi-source environmental factors can reflect the dynamic changes of environmental characteristics before the spread of malaria, thereby providing a data basis for determining the time lag period and training the model.

[0055] S103: Analyze the time series of multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors.

[0056] In the embodiments of the present application, the effect of the change of environmental factors on the occurrence of malaria is not immediately apparent, but there is a certain lag effect. For example, an increase in precipitation may cause the spread of malaria due to the peak of mosquito breeding in a few weeks. Therefore, the present application calculates the correlation between the time series of multi-source environmental factors and the incidence of malaria by statistical methods, and finds the most significant optimal time lag period related to malaria. Correspondingly, each environmental factor in the multi-source environmental factors will correspond to an optimal time lag period.

[0057] The optimal time lag refers to the time period when the environmental factors have the most significant impact on the occurrence of malaria. The optimal time lag for each environmental factor may be different. For example, precipitation may take 4 weeks to affect the spread of malaria, while temperature changes may show an impact on the occurrence of malaria after 6 weeks.

[0058] It should be noted that by determining the optimal time lag, the temporal effect of environmental factors on malaria transmission can be accurately captured, making the subsequently constructed training data set more in line with the laws of malaria transmission and improving the accuracy of the prediction model.

[0059] S104: constructing a training data set according to the optimal time lag of the multi-source environmental factors, and inputting the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0060] In an embodiment of the present application, a training data set refers to a data set used to train a supervised learning model. Specifically, the present application extracts data of a corresponding time period from multi-source earth observation data according to the optimal time lag of multi-source environmental factors, and constructs a training data set based on the occurrence of malaria cases at ground survey points.

[0061] After the training data set is constructed, it will be input into the supervised learning model for training. The supervised learning model is a machine learning method that can learn the relationship between features and labels by inputting labeled training data. Supervised learning models can include: support vector machine SVM, random forest RF, neural network NN, etc.

[0062] In some implementations of the embodiments of the present application, support vector machines are used. This supervised learning model is widely used in classification and regression analysis. For linear inseparable problems, the principle is to introduce the inner product operation of the kernel function in the high-dimensional space, nonlinearly map the input vector to a high-dimensional feature vector space, and construct the optimal classification hyperplane in the feature space. Based on this, the support vector machine model is used as a malaria prediction model, and malaria training, verification and prediction are realized through data-driven.

[0063] The supervised learning model after training is a malaria occurrence risk prediction model, which can receive a new prediction data set and output a prediction structure of malaria occurrence risk. It can be understood that the structure of the prediction data set is similar to that of the training data set, and the only difference lies in the selection of the time range and whether the ground survey points contain known labels.

[0064] exist Figure 1In the process shown, multiple ground survey points in the area to be predicted are first determined to mark whether there have been cases of malaria. Then, based on the locations of these ground survey points, time series of various environmental factors closely related to the occurrence of malaria are extracted from multi-source earth observation data. These environmental factors are key factors in the spread of malaria, and their time series can reveal the seasonal fluctuations and long-term trends of environmental factors. Afterwards, in view of the possible time lag effect between environmental factors and the occurrence of malaria, that is, the changes in certain environmental factors may not show their impact on the occurrence of malaria until a period of time has passed, the present application further analyzes the time series of multi-source environmental factors to determine the optimal time lag between each environmental factor and the occurrence of malaria. This allows a more accurate grasp of the temporal association between environmental factors and malaria. Finally, a training data set is constructed based on the determined optimal time lag, and is input into a supervised learning model for training. The model will learn the complex relationship between environmental factors and the occurrence of malaria, and then form a malaria occurrence risk prediction model. The trained malaria occurrence risk prediction model can use a new prediction data set to accurately predict the malaria risk in the predicted area. In general, this application extracts time series of multi-source environmental factors and uses the optimal time lag period for model training, so that the prediction model of malaria risk can fully consider the lag effect and comprehensively capture the complex relationship between environmental factors and malaria, thereby achieving high-precision prediction of malaria risk.

[0065] See also Figure 2 As shown, the embodiment of the present application further provides a method for predicting the risk of malaria occurrence, which may specifically include the following steps:

[0066] S201: Determine a plurality of ground survey points in the area to be predicted, the ground survey points being used to characterize whether malaria cases have occurred, and the ground survey points include: malaria case occurrence points and malaria case non-occurrence points.

[0067] S202: Divide the area to be predicted into a plurality of grids with a preset spatial resolution.

[0068] In an embodiment of the present application, the area to be detected is rasterized according to a preset spatial resolution. Rasterization is the process of dividing the area to be detected into regular grids. Each grid is called a grid. Each grid represents a specific geographical area and is used to form a corresponding relationship with multi-source earth observation data, malaria occurrence data and other data in the area. The preset spatial resolution is the size of the grid, for example, it can be 5km×5km, 2km×2km, etc. The smaller the resolution, the higher the accuracy. Preferably, the preset spatial resolution is determined based on the maximum flight radius of the malaria mosquito.

[0069] By rasterizing the area to be detected, the entire area to be detected can be divided into grids of uniform size. Each grid can be used as a basic unit of analysis, which is convenient for combining environmental factor data with ground survey points. At the same time, rasterization can refine the data granularity, enabling the model to capture the local characteristics of malaria occurrence and output the probability of malaria risk occurrence in each grid, which is the model prediction result.

[0070] S203: Screening out the malaria-free points whose distances from the malaria-free points to the malaria-free points are less than a preset spatial resolution, so that the malaria-free points and the malaria-free points belong to different grids.

[0071] In the embodiment of the present application, the ground survey points are divided into malaria case occurrence points and malaria case non-occurrence points. In order to prevent the malaria case occurrence points and the malaria case non-occurrence points from being too close to each other and falling into the same grid, thereby causing data deviation and model misjudgment, the malaria case non-occurrence points that are too close are screened out. Among them, the preset spatial resolution is used as the judgment standard for the distance being too close. If the distance between the malaria case non-occurrence point and the malaria case occurrence point is less than the preset spatial resolution, they will be considered to be within the same grid range, so the corresponding malaria case non-occurrence point needs to be deleted, so as to achieve the screening of the malaria case non-occurrence point in the grid to which the malaria occurrence point belongs.

[0072] S204: Randomly generate a plurality of points where malaria cases have not occurred outside a preset value of the distance from the malaria case occurrence point, wherein the preset value is greater than a preset spatial resolution.

[0073] In the embodiment of the present application, there may be an uneven population distribution in the area to be detected, which will lead to an uneven distribution of samples and thus affect the global prediction ability of the supervised learning model for the area to be detected. Therefore, it is necessary to balance the sample distribution by increasing the number of points where malaria cases have not occurred. Specifically, a plurality of points where malaria cases have not occurred are randomly generated outside a preset value of the distance from the malaria case occurrence point, and the preset value is greater than the preset spatial resolution, ensuring that the newly generated points where malaria cases have not occurred do not fall into the grid to which the malaria case occurrence point belongs.

[0074] S205: Extracting the time series of multi-source environmental factors driving the occurrence of malaria from multi-source earth observation data based on multiple ground survey points.

[0075] S206: Analyze the time series of multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors.

[0076] S207: Constructing a training data set according to the optimal time lag of the multi-source environmental factors, and inputting the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0077] exist Figure 2 In the process shown, the validity of data distribution is ensured by rasterizing the area to be predicted and screening out points where malaria cases have not occurred, thereby improving the distinguishing ability of the malaria occurrence risk prediction model; in addition, by randomly adding points where malaria cases have not occurred outside a preset distance to balance the sample distribution, the global prediction ability of the malaria occurrence risk prediction model is improved. Overall, the embodiment of the present application improves the prediction accuracy of the malaria occurrence risk prediction model by rasterizing the area to be predicted, screening out points where malaria cases have not occurred within a specified distance, and adding new points.

[0078] See also Figure 3 As shown, the embodiment of the present application further provides a method for predicting the risk of malaria occurrence, which may specifically include the following steps:

[0079] S301: Determine a plurality of ground survey points in the area to be predicted, where the ground survey points are used to characterize whether malaria cases have occurred.

[0080] S302: Environmental variables are extracted from multi-source earth observation data according to the latitude and longitude coordinates of multiple ground survey points, remote sensing images in the environmental variables are declouded, and multi-temporal image synthesis technology and / or inverse distance weighted spatial interpolation technology are used to fill in the missing pixels after declouding to obtain the preprocessed environmental variables.

[0081] In an embodiment of the present application, environmental variables corresponding to the latitude and longitude coordinates of ground survey points are extracted from multi-source earth observation data, part of the remote sensing image in the environmental variables is subjected to cloud removal, and multi-temporal image synthesis technology and / or inverse distance weighted spatial interpolation technology are used to fill in the missing pixels that still exist after the cloud removal processing, ultimately generating high-quality preprocessed environmental variables.

[0082] The purpose of performing cloud removal in the above process is that some areas, especially the areas to be detected in tropical regions, are frequently covered by clouds, and the clouds will block the satellite's observation of the surface, resulting in some areas in the remote sensing image being unable to be observed. Therefore, the present application eliminates the interference of clouds on remote sensing images through cloud removal and restores the real surface information to a greater extent. Preferably, the embodiment of the present application is based on the Google Earth Engine (GEE) cloud platform to perform pixel-level cloud removal on remote sensing images.

[0083] After declouding part of the remote sensing image, there may still be missing pixels that are not completely filled in the image, which is not conducive to the integrity of the predicted structure, so further completion is needed.

[0084] Specifically, the embodiments of the present application may use multi-temporal image synthesis technology and / or inverse distance weighted spatial interpolation technology. Multi-temporal image synthesis technology uses observation data of the same area at different times to synthesize a complete image. Specifically, by selecting multi-temporal remote sensing data of the area to be predicted, such as observation images of the 16 days before and after or the same period last year, the mean of the time series image can be calculated in the pixel area covered by clouds to generate the completed pixel value. Its advantage is that it can use historical data to accurately restore the long-term change trend. The inverse distance weighted spatial interpolation technology uses the weighted pixel values ​​of the position near a missing pixel on a single-phase image to calculate the average value as the missing pixel value. The weight assigned to the neighboring pixels is proportional to the distance of the missing pixel position, and different power exponents can be specified. The larger the power exponent, the greater the weight, which is more suitable for local completion.

[0085] In the embodiment of the present application, when the multi-temporal image synthesis technology and the inverse distance weighted spatial interpolation technology are used, the preferred order is first the multi-temporal image synthesis technology and then the inverse distance weighted spatial interpolation technology. That is, when the multi-temporal image synthesis cannot fill all the gaps, the inverse distance weighted spatial interpolation technology is used to continue to fill them.

[0086] S303: Extracting the time series of multi-source environmental factors driving the occurrence of malaria from the preprocessed environmental variables.

[0087] S304: Analyze the time series of multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors.

[0088] S305: Constructing a training data set according to the optimal time lag of the multi-source environmental factors, and inputting the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0089] exist Figure 3 In the process shown, the data quality of remote sensing images is improved through cloud removal and missing pixel completion. High-quality pre-processed environmental variables can provide accurate input for supervised learning models and improve the accuracy of malaria risk prediction. At the same time, it can better meet the special needs of areas with high incidence of malaria.

[0090] See also Figure 4 As shown, the embodiment of the present application further provides a method for predicting the risk of malaria occurrence, which may specifically include the following steps:

[0091] S401: Determine a plurality of ground survey points in the area to be predicted, where the ground survey points are used to characterize whether malaria cases have occurred.

[0092] S402: Extract environmental variables from multi-source earth observation data based on the latitude and longitude coordinates of multiple ground survey points, and resample the environmental variables using the time frequency and spatial resolution of the ground survey points as reference standards so that the environmental variables have the same time frequency and spatial resolution as the ground survey points, thereby obtaining preprocessed environmental variables.

[0093] In an embodiment of the present application, environmental variables related to ground survey points are extracted from multi-source earth observation data. These environmental variables are resampled in time frequency and spatial resolution to make them consistent with the ground survey points, and finally preprocessed environmental variables that meet the analysis requirements are generated.

[0094] Specifically, based on the longitude and latitude coordinates of the ground survey point, the environmental variable values ​​of the corresponding location are obtained from the multi-source earth observation data. For example: for the longitude and latitude coordinates (X, Y) of a ground survey point, the temperature, precipitation, humidity and other data of the location within a specific time range are extracted. Then, since the temporal frequency and spatial resolution of different data sources may be inconsistent, this difference will cause the data to be unable to be directly matched in time and space, thus affecting subsequent analysis. Therefore, it is necessary to resample the extracted environmental variables and adjust the data time interval of the environmental variables to be consistent with the ground survey point. For example, daily frequency data is aggregated into weekly frequency data, and the spatial resolution of the environmental variables is adjusted to match the ground survey point. For example, the resolution of 2000 meters is increased to 1000 meters. The preprocessed environmental variables thus obtained have the same temporal frequency and spatial resolution as the temporal frequency and spatial resolution of the ground survey point.

[0095] S403: Extracting the time series of multi-source environmental factors driving the occurrence of malaria from the preprocessed environmental variables.

[0096] S404: Analyze the time series of multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors.

[0097] S405: Constructing a training data set according to the optimal time lag of the multi-source environmental factors, and inputting the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0098] exist Figure 4 In the process shown, the differences in temporal and spatial resolution between earth observation data and ground survey point data are eliminated through resampling, so that the two can be effectively matched, thereby ensuring the accuracy of data analysis.

[0099] See also Figure 5 As shown, the embodiment of the present application further provides a method for predicting the risk of malaria occurrence, which may specifically include the following steps:

[0100] S501: Determine a plurality of ground survey points in the area to be predicted, where the ground survey points are used to characterize whether malaria cases have occurred.

[0101] S502: Extract the time series of multi-source environmental factors that drive the occurrence of malaria from multi-source earth observation data based on multiple ground survey points. The multi-source environmental factors include dynamic environmental factors and static environmental factors. The time window length of the time series of dynamic environmental factors is shorter than that of static environmental factors. The dynamic environmental factors include precipitation, normalized difference vegetation index, enhanced vegetation index and surface temperature. The static environmental factors include land use / cover, elevation and travel time.

[0102] In the embodiments of the present application, dynamic environmental factors refer to environmental factors that change rapidly in spatial distribution or value over time, including precipitation environmental factors Rain that affect the formation of mosquito breeding grounds, normalized vegetation index NDVI that reflects the influence of vegetation coverage on the habitat of mosquitoes, enhanced vegetation index EVI that is more suitable for densely vegetated areas, and land surface temperature LST that affects the growth and development of mosquitoes. It is understandable that dynamic environmental factors change more frequently, so the time window of their time series is shorter, such as weeks or months.

[0103] Static environmental factors are those that remain relatively stable over a long period of time, including: land use / cover LULC that affects human activities and mosquito habitat distribution, elevation DEM that indirectly affects temperature and precipitation patterns, and traffic time that affects population mobility and thus disease transmission. It is understandable that static environmental factors have a longer change cycle and a relatively larger time window, such as years.

[0104] S503: Perform Spearman correlation analysis on the time series of multi-source environmental factors to obtain the optimal time lag of multi-source environmental factors.

[0105] In the embodiments of the present application, Spearman Correlation Analysis is used to find the correlation between the time lag of different multi-source environmental factors and the occurrence of malaria cases, thereby determining the optimal time lag of each factor, that is, the time interval that best reflects the malaria transmission relationship.

[0106] Spearman correlation is a nonparametric correlation analysis method based on data sorting, which is suitable for describing nonlinear relationships. Since there is a complex nonlinear relationship between environmental factors and whether malaria occurs, the Spearman correlation coefficient is used to measure how the time-lagged changes in environmental factors affect whether malaria occurs. Among them, the Spearman correlation coefficient ranges from -1 to 1, where -1 indicates a complete negative correlation, +1 indicates a complete positive correlation, and 0 indicates no correlation at all. Assuming that whether malaria occurs is a binary variable Y, taking two values ​​of 0 and 1, and the four dynamic environmental factors are independent variables Xi, i∈{1,2,3,4}, then the Spearman correlation coefficient between each environmental factor Xi and Y is:

[0107]

[0108] Among them, d k =R(X k )-R(Y k ) represents the rank difference between the environmental factor and the binary variable in the kth sample, and n is the total number of samples.

[0109] Assume that the incidence of malaria in a certain area is related to precipitation and surface temperature. Spearman correlation analysis is performed on the precipitation time series [Rain_lag1, Rain_lag2, ..., Rain_lag12] and malaria case data. The results show that the maximum value is reached at Rain_lag4, indicating that malaria transmission is most related to precipitation 4 weeks ago. Analysis of the surface temperature time series [LST_lag1, LST_lag2, ..., LST_lag12] shows that the maximum value is reached at LST_lag9, indicating that malaria transmission is most related to temperature 9 weeks ago.

[0110] S504: constructing a training data set according to the optimal time lag of the multi-source environmental factors, and inputting the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0111] exist Figure 5 In the process shown, the time series of dynamic environmental factors captures the lag effect of malaria transmission, such as high precipitation that may lead to mosquito breeding within a few weeks, and static environmental factors provide background environmental information of the study area, such as elevation data revealing the long-term distribution characteristics of regional temperature and precipitation, thus laying the foundation for model training. The combination of dynamic and static factors enables the malaria transmission model to capture short-term dynamic characteristics and combine long-term stable environmental characteristics, thereby improving prediction accuracy.

[0112] Below, the technical solution provided in the embodiment of the present application is further explained in combination with specific application scenarios.

[0113] The overall technical route of this application can be as follows Figure 6 As shown. Based on the driving mechanism of external factors affecting the occurrence of malaria, a multi-source environmental indicator system of meteorological, remote sensing, geographical, and transportation conditions is constructed, and a response mechanism model of the indicator system and malaria events is further established, so as to achieve 1-2 weeks in advance prediction of whether malaria will occur in an area with a resolution of 1km. The entire prediction process can be roughly divided into:

[0114] (1) Extract the multi-source environmental factors that drive malaria and their time series;

[0115] (2) To study the lag between environmental variables and malaria incidence, to analyze the optimal time lag of multi-source environmental factors in the time series of malaria occurrence through statistical methods, and to establish a malaria prediction indicator system;

[0116] (3) Create a data set based on the indicator system and input it into the constructed malaria prediction model for training and verification;

[0117] (4) Create a prediction data set and input it into the trained malaria prediction model to complete the prediction.

[0118] In the present application, the example uses Sao Tome and Principe, where the risk of malaria is relatively high, as the study area, i.e., the area to be predicted. Figure 7 As shown. Sao Tome and Principe is located in the Gulf of Guinea, with a total area of ​​1,001 square kilometers, about 400 kilometers from the west coast of Africa. Principe Island is 173 kilometers away from Sao Tome Island, and the population is mainly concentrated on the plains on the east coast of Sao Tome Island. Sao Tome and Principe has a tropical rainforest climate. Because it is close to the equator, the annual average temperature is relatively high. The seasons are distinct, and the year is divided into two main seasons: the dry season and the rainy season. The dry season usually lasts from June to August, and the rainy season lasts from September to May of the following year. During the rainy season, short but intense rainfalls often occur on the islands. Due to its geographical location and climate, the islands of Sao Tome and Principe have rich tropical rainforest vegetation and a diverse ecological environment. For the above study area, the following schemes can be adopted in practical applications:

[0119] The first step is to determine the spatial resolution.

[0120] According to the survey, adult mosquitoes may show a high diffusion rate between villages, with the longest flight distance reaching 5 kilometers, but half of the flights are within a 1-kilometer radius. Therefore, the spatial resolution of multi-source data and ground survey points was finally determined to be 1km.

[0121] The second step is to screen ground survey points for malaria cases.

[0122] There are two types of ground survey points, namely, malaria case occurrence points and malaria case non-occurrence points. According to the above-mentioned setting of a spatial resolution of 1km, the area to be predicted is divided into a 1km×1km grid. If the malaria case occurrence point and the malaria case non-occurrence point are geographically close, they may fall into the same 1km×1km grid at the same time. The embodiment of the present application mainly focuses on the malaria case occurrence point. Therefore, the distance from the malaria case non-occurrence point to the malaria case occurrence point can be calculated based on the latitude and longitude coordinates of the ground survey point, and the non-occurrence points that are less than the set threshold are removed. The set threshold is set to 1, so different types of ground survey points can be divided into different grids after screening.

[0123] The third step is to generate pseudo-random points.

[0124] Since people are mainly concentrated in the northeast of Sao Tome Island, some people also live in Principe Island in its southeastern coast and northwest, resulting in an uneven population distribution. The embodiment of the present application focuses on the risk of malaria occurrence on the entire island, so ground survey points are also needed in sparsely populated areas, where there are basically no malaria cases, so pseudo-random points are generated according to certain rules based on the malaria transmission habits, that is, points where malaria cases do not occur. According to the above-mentioned adult mosquito flight maximum distance of up to 5 kilometers, malaria cases are randomly generated 5 kilometers away from the ground survey point.

[0125] The fourth step is to extract the environmental variable values ​​of the corresponding points according to the longitude and latitude of the ground survey points.

[0126] According to Table 1, we can get the land surface temperature LST, normalized difference vegetation index NDVI, enhanced vegetation index EVI, precipitation, elevation data DEM, land use / cover LULC and travel time, etc. Here, travel time specifically refers to the time to travel to cities or the time to walk to medical facilities (Walking Only Travel Time To Healthcare), etc.

[0127] Among them, the land surface temperature LST, the normalized difference vegetation index NDVI, and the enhanced vegetation index EVI are calculated according to the following formula to obtain the required indicators or normal value ranges.

[0128] LST=(LST_Day_1km+LST_Night_1km) / 2*0.02-273.15

[0129] Among them, LST represents the land surface temperature, LST_Day_1km represents the daytime land surface temperature, and LST_Night_1km represents the nighttime land surface temperature.

[0130] NDVI=(sur_refl_b02-sur_refl_b01) / (sur_refl_b02+sur_refl_b01),-1≦NDVI≦1

[0131] Among them, NDVI represents the normalized difference vegetation index, sur_refl_b01 represents the red band, and sur_refl_b02 represents the near infrared band.

[0132] EVI=EVI*0.0001,-1≦EVI≦1

[0133] Among them, EVI stands for Enhanced Vegetation Index.

[0134] Table 1 Summary of data used in the examples of this application

[0135]

[0136] The fifth step is to preprocess the environment variable values, including de-clouding, gap filling, and resampling.

[0137] Based on the GEE cloud platform, pixel-level cloud removal is performed on remote sensing images. There will be a large number of image gaps in the images after cloud removal, which is not conducive to the integrity of the prediction results. Therefore, two methods, multi-temporal image synthesis and inverse distance weighted spatial interpolation, are used to fill the gaps. Multi-temporal image synthesis uses images within 16 days before and after and images of the same period last year to calculate the average of the images that need to be filled, and uses composite images to fill the gaps after cloud removal. When multi-temporal image synthesis is unable to fill all the gaps, inverse distance weighted spatial interpolation technology is used to continue filling. Taking the temporal frequency and spatial resolution of ground survey data as reference standards, the environmental variables are resampled to a temporal resolution of 7 days and a spatial distribution rate of 1km.

[0138] The sixth step is to extract the multi-source environmental factors and their time series that drive the occurrence of malaria based on ground survey points.

[0139] According to the time update frequency of relevant factors, they can be divided into two categories: one is dynamic environmental factors, which change rapidly in spatial distribution and value over time, such as precipitation, temperature, vegetation information, etc.; the other is static environmental factors, whose spatiotemporal distribution is relatively stable, and changes can usually be observed in years, including elevation DEM, land use / cover LULC and traffic data. According to the ground survey points processed above, the dynamic environmental factor data of 12 weeks before the sampling date are selected, recorded as [lag12, lag11, lag10, lag09, ..., lag02, lag01], while the selection of static environmental factors only needs to select the data of the corresponding year according to the sampling date.

[0140] The seventh step is to use statistical methods to analyze the optimal lag period of multi-source environmental factors.

[0141] Since there is a complex nonlinear relationship between environmental factors and whether malaria occurs, the present embodiment uses the Spearman correlation coefficient to measure how the time-lagged changes of environmental factors affect whether malaria occurs. By performing Spearman correlation analysis on the four dynamic environmental factors and whether malaria events occur, the hysteresis response characteristics of the dynamic environmental factors can be determined, such as Figure 8 As shown. Among them, precipitation Rain showed a high significant positive correlation with whether a malaria event occurred during the lag08-lag06 period, indicating that if rainfall occurred 42 to 56 days before the occurrence of malaria, the possibility of malaria would increase; NDVI showed a positive correlation with whether a malaria event occurred during the lag05 period, and the possibility of affecting the occurrence of malaria during this period was relatively high; the enhanced vegetation index EVI showed a positive correlation with whether a malaria event occurred during the lag12-lag08 period, indicating that if the EVI value gradually increased from 56 to 84 days before the occurrence of malaria, the risk of malaria would increase; LST showed a high significant positive correlation with whether a malaria event occurred during the lag09 period, indicating that the possibility of malaria would be affected in the 9th week before the occurrence of malaria. This is mainly because temperature can affect the growth and development of malarial parasites in malaria mosquitoes. In the suitable temperature range of 16℃-30.2℃, the increase in temperature will accelerate the development of malarial parasites, thereby spreading through the bites of malarial mosquitoes and causing human illness, which is in line with the law of malaria transmission. In the conclusions drawn through correlation analysis, the lag period of precipitation and ground temperature LST is consistent with the lag period range obtained by Maquins et al. in their analysis of malaria in western Kenya. The lag period of the normalized difference vegetation index NDVI is consistent with the lag period range obtained by AbiodunMorakinyo Adeola et al. and PeterHaddawy et al. in their malaria analysis. The EVI lag period is also within a reasonable range. This further confirms the authenticity and reliability of data processing and provides support for model training.

[0142] The eighth step is to extract the precipitation in the lag08-lag06 period, the normalized vegetation index in the lag05 period, the enhanced vegetation index in the lag12-lag08 period, and the surface temperature in the lag09 period in combination with the optimal time lag period. And together with static environmental factors including elevation data, land use / cover LULC, time to the city, time to walk to medical facilities, ground survey points in the first 4 weeks, and ground survey points in the same period last year, an index system is established, and input samples are made and passed into the constructed model. Here, the embodiment of the present application uses four indicators of prediction accuracy (Accuracy), recall rate (Recall), F1 score (F1-score), and area under the curve (Area Under the Curve, AUC) to evaluate the risk prediction results of the support vector machine model for malaria case sampling points from the 31st week in 2022 to the 30th week in 2023, as shown in Table 2. The prediction accuracy can measure the overall correctness of the model, the recall rate focuses on the recognition ability of the model's positive samples, the F1 score is used to strike a balance between precision and recall, and the area under the curve is used to evaluate the model's overall ability to distinguish between positive and negative samples. Combined with relevant research, it was found that the overall prediction accuracy of the model was 0.72, the recall rate of positive samples was 0.65, the F1 score was 0.67, and the area under the curve was 0.74. On a weekly scale, this prediction level can provide a reference for weekly predictions of malaria. See Table 3 here. Table 3 is a further statistical analysis of Table 2. According to the perpetual calendar, the weekly results are counted into months. It can be seen that only the prediction accuracy of March 2023 is low, and the results of other months are good, especially the prediction accuracy of August, September, and October has reached more than 80%. From the perspective of AUC indicator analysis, the AUC values ​​of most months are above 0.7, indicating that the model has a good ability to distinguish between positive and negative samples; similarly, from the perspective of F1 score, except for the low F1 score in March, the F1 scores of other months are passing or even excellent; from the perspective of recall rate, except for the low recall rate in September, the recall rate of other months exceeds 50%, and more than half of the recall rates exceed 70%.

[0143] Table 2 Weekly prediction results based on SVM model

[0144]

[0145] Table 2 (Continued) Weekly prediction results based on SVM model

[0146]

[0147] Note: 202231 represents the 31st week of 2022.

[0148] Table 3 Statistics of monthly prediction results based on SVM model

[0149]

[0150] Table 3 (continued) Monthly forecast results based on SVM model

[0151]

[0152] Next, the risk prediction probability results and ground observation results for the entire study area of ​​São Tomé and Príncipe Island are presented at weekly and monthly scales, respectively. Fig. 9 and Fig.10 In the weekly scale results display, due to the large number of Fig. 9 Only the results from weeks 20 to 27, which have more survey points and newer dates, are shown. The results show that the risk of malaria is higher in the northeastern and coastal areas of São Tomé Island, and there are also small high-risk areas in the southeastern and northwest coasts; compared with São Tomé Island, the probability of malaria cases in Principe Island is lower. Except for the forecast results of weeks 26 and 27 in 2023, the recall rate of positive samples is low, and the recall rate of positive samples in other weeks is above 0.70, which can provide information support for the prevention and control of malaria cases in the next 1-2 weeks. Fig.10 The results of the risk probability at the monthly scale are presented, ranging from August 2022 to July 2023. The results show that malaria occurrence is mainly concentrated in the northeast of São Tomé Island. From September to December, the risk of malaria increases month by month, and the risk probability of malaria occurrence reaches a relatively serious level in December; from January to May 2023, the range of malaria occurrence risk in the study area is also gradually expanding. The reason is that the islands of São Tomé and Principe are in the rainy season during this period, and the climate is humid, which provides favorable conditions for the breeding of malaria mosquitoes. Once infected malaria mosquitoes spread between human hosts through bites, they will trigger the occurrence of malaria. In June and July 2023, the range of malaria occurrence risk began to narrow, but the risk probability in the northeast of São Tomé Island is still increasing, which may be related to the spread of the disease.

[0153] In summary, based on the entire time series of malaria cases, the Spearman correlation coefficient was used to analyze the time lag response of dynamic environmental factors, and the optimal lag period for precipitation was 6-8 weeks, the optimal lag period for NDVI was the 5th week, the optimal lag period for EVI was the 8th-12th week, and the optimal lag period for LST was the 9th week. The optimal lag characteristics of the corresponding period of dynamic environmental factors were extracted and combined with static environmental factors as the input features of the support vector machine model to predict the probability of malaria risk. The risk prediction and forecast of occurrence can be achieved 1-2 weeks in advance, providing strong support for public health departments in formulating prevention and control measures. The northeastern part of Sao Tome Island is a perennial place for malaria cases. It is densely populated and has convenient transportation. The northwest coastal and southeast coastal areas are accompanied by local malaria risks. The risk of malaria in Principe Island is seasonal and generally occurs in the rainy season. In addition, according to the topography of Sao Tome and Principe Island, malaria often occurs in areas with low altitudes, wetlands, shrubs, broad-leaved forests and other land cover types.

[0154] See also Fig.11 As shown, the embodiment of the present application provides a device for predicting the risk of malaria occurrence, the device comprising:

[0155] A survey point determination module 1101 is used to determine a plurality of ground survey points in the area to be predicted, where the ground survey points are used to characterize whether malaria cases have occurred;

[0156] A time series extraction module 1102 is used to extract the time series of multi-source environmental factors driving the occurrence of malaria from multi-source earth observation data based on multiple ground survey points;

[0157] The time series analysis module 1103 is used to analyze the time series of multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors;

[0158] The model training module 1104 is used to construct a training data set according to the optimal time lag of the multi-source environmental factors, and input the training data set into the supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

[0159] In some implementations of the embodiments of the present application, the ground survey points include: points where malaria cases occur and points where malaria cases do not occur. After determining the plurality of ground survey points, the device further includes:

[0160] A rasterization module, used to divide the area to be predicted into multiple grids of preset spatial resolution;

[0161] The screening module is used to screen out the malaria-free points whose distance from the malaria-free points to the malaria-free points is less than a preset spatial resolution, so that the malaria-free points and the malaria-free points belong to different grids.

[0162] In some implementations of the embodiments of the present application, the device further includes:

[0163] The pseudo sample generation module is used to randomly generate a plurality of malaria case non-occurrence points outside a preset value of the distance from the malaria case occurrence point, wherein the preset value is greater than a preset spatial resolution.

[0164] In some implementations of the embodiments of the present application, a time series of multi-source environmental factors driving the occurrence of malaria is extracted from multi-source earth observation data according to multiple ground survey points, including:

[0165] According to the latitude and longitude coordinates of multiple ground survey points, environmental variables are extracted from multi-source earth observation data for preprocessing to obtain preprocessed environmental variables;

[0166] The time series of multi-source environmental factors driving malaria occurrence are extracted from the pre-processed environmental variables.

[0167] In some implementations of the embodiments of the present application, the environmental variables include: remote sensing images, and the preprocessing includes:

[0168] The remote sensing images are declouded and the missing pixels after declouding are filled by using multi-temporal image synthesis technology and / or inverse distance weighted spatial interpolation technology.

[0169] In some implementations of the embodiments of the present application, preprocessing includes:

[0170] Taking the temporal frequency and spatial resolution of the ground survey points as the reference standard, the environmental variables are resampled to make them have the same temporal frequency and spatial resolution as the ground survey points.

[0171] In some implementations of the embodiments of the present application, the multi-source environmental factors include dynamic environmental factors and static environmental factors, wherein the time window length of the time series of the dynamic environmental factors is shorter than that of the static environmental factors.

[0172] In some implementations of the embodiments of the present application, analyzing the time series of multi-source environmental factors includes:

[0173] Spearman correlation analysis was performed on the time series of multi-source environmental factors.

[0174] In some implementations of the embodiments of the present application, the dynamic environmental factors include: precipitation, normalized difference vegetation index, enhanced vegetation index and surface temperature, and the static environmental factors include: land use / cover, elevation and travel time.

[0175] Finally, it should be noted that, in the embodiments of the present application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements that are not clearly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements limited by the statement "comprise one..." do not exclude the presence of other identical elements in the process, method, article or equipment including the elements.

[0176] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the risk of malaria occurrence, characterized in that: The method comprises: Determine a plurality of ground survey points in the area to be predicted, wherein the ground survey points are used to characterize whether malaria cases have occurred; extracting the time series of multi-source environmental factors driving the occurrence of malaria from multi-source earth observation data based on the plurality of ground survey points; Analyzing the time series of the multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors; A training data set is constructed according to the optimal time lag of the multi-source environmental factors, and the training data set is input into a supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

2. The method according to claim 1, characterized in that The ground survey points include: points where malaria cases occurred and points where malaria cases did not occur. After determining a plurality of the ground survey points, the method further includes: Dividing the area to be predicted into a plurality of grids of a preset spatial resolution; The malaria-free points whose distances from the malaria-free points to the malaria-free points are less than the preset spatial resolution are screened out, so that the malaria-free points and the malaria-free points belong to different grids.

3. The method according to claim 2, characterized in that The method further comprises: A plurality of points where malaria cases have not occurred are randomly generated outside a preset value of the distance from the malaria case occurrence point, wherein the preset value is greater than the preset spatial resolution.

4. The method according to claim 1, characterized in that: The step of extracting the time series of multi-source environmental factors driving the occurrence of malaria from multi-source earth observation data based on the plurality of ground survey points includes: Extracting environmental variables from multi-source earth observation data according to the latitude and longitude coordinates of the plurality of ground survey points for preprocessing to obtain preprocessed environmental variables; The time series of the multi-source environmental factors driving the occurrence of malaria are extracted from the pre-processed environmental variables.

5. The method according to claim 1, characterized in that The environmental variables include: remote sensing images, and the preprocessing includes: The remote sensing image is subjected to cloud removal processing, and the missing pixels after the cloud removal processing are filled by using multi-temporal image synthesis technology and / or inverse distance weighted spatial interpolation technology.

6. The method according to claim 1, characterized in that The pre-processing comprises: The environmental variables are resampled with the temporal frequency and the spatial resolution of the ground survey points as reference standards, so that the environmental variables are the same as the temporal frequency and the spatial resolution of the ground survey points.

7. The method according to claim 1, characterized in that The multi-source environmental factors include dynamic environmental factors and static environmental factors, wherein the time window length of the time series of the dynamic environmental factors is shorter than that of the static environmental factors.

8. The method according to claim 1, characterized in that The analyzing the time series of the multi-source environmental factors comprises: The time series of the multi-source environmental factors were subjected to Spearman correlation analysis.

9. The method according to claim 7, characterized in that: The dynamic environmental factors include: precipitation, normalized difference vegetation index, enhanced vegetation index and surface temperature, and the static environmental factors include: land use / cover, elevation and travel time.

10. A device for predicting the risk of malaria occurrence, characterized in that: The device comprises: A survey point determination module is used to determine a plurality of ground survey points in the area to be predicted, wherein the ground survey points are used to characterize whether malaria cases have occurred; A time series extraction module, for extracting the time series of multi-source environmental factors driving the occurrence of malaria from multi-source earth observation data according to the plurality of ground survey points; A time series analysis module is used to analyze the time series of the multi-source environmental factors to obtain the optimal time lag of the multi-source environmental factors; A model training module is used to construct a training data set according to the optimal time lag period of the multi-source environmental factors, and input the training data set into a supervised learning model for training to obtain a malaria occurrence risk prediction model, wherein the malaria occurrence risk prediction model is used to predict the malaria occurrence risk in the area to be detected based on the prediction data set.

Images