Drought cause index screening method

Abstract

The invention discloses a drought cause index screening method, and relates to the technical field of drought cause index screening, and the method comprises the steps: obtaining multi-scale drought data in time and space, and extracting the drought features of the multi-scale drought data; constructing a comprehensive drought index based on the drought characteristics; obtaining a data set based on the comprehensive drought index and the multi-scale drought data; analyzing the data set based on a drought evaluation model to obtain a loss value, inputting the loss value into an SHAP model to obtain an SHAP value, and screening drought cause indexes based on the SHAP value to obtain a screening result; the problems that in the process of obtaining the causal relationship between observation data and drought, an existing method neglects the influence of underlying surface factors on drought events, the relationship between agricultural and hydrological drought conditions and observation data cannot be mined, spatial features are neglected, and the spatial evolution features of the drought process are difficult to reflect can be solved.

Description

Technical Field

[0001] The invention relates to the field of screening drought cause indicators, and in particular to a method for screening drought cause indicators. Background Art

[0002] Drought is a natural disaster that has a long duration, severe losses and a wide range of impacts. It seriously affects agricultural production and human livelihoods. In recent years, against the backdrop of global climate change, the development of local droughts is accelerating. For example, the intensification of sudden droughts will seriously hinder the development of human society and threaten food security.

[0003] Using drought indicators to monitor and predict drought events is an effective way to prevent drought disasters. In the field of drought monitoring, the causal relationship between various observation data and drought is usually analyzed to select the most appropriate features for subsequent drought prediction, understand the principles of the physical process of comprehensive drought events, and improve the accuracy of drought prediction.

[0004] The causal relationship between observation data and drought is currently mostly accomplished by using easily accessible meteorological drought monitoring indicators such as SPI, SPEI and MCI, using correlation coefficients, model simulation and other methods. However, this method has the following disadvantages:

[0005] 1) The impact of underlying surface factors on drought events is usually ignored, and the relationship between agricultural and hydrological drought conditions and observational data cannot be explored;

[0006] 2) The temporal characteristics of the drought process are taken into account to a certain extent, but the spatial characteristics are ignored, making it difficult to reflect the spatial evolution characteristics of the drought process. Summary of the invention

[0007] In order to solve the problem that in obtaining the causal relationship between observation data and drought, the existing methods ignore the influence of underlying surface factors on drought events, fail to explore the relationship between agricultural and hydrological drought conditions and observation data, ignore spatial characteristics, and are difficult to reflect the spatial evolution characteristics of the drought process, the present invention provides a method for screening drought cause indicators, the method comprising:

[0008] Acquire multi-scale drought data in time and space, extract drought characteristics of the multi-scale drought data; construct a comprehensive drought index based on the drought characteristics; obtain a data set based on the comprehensive drought index and the multi-scale drought data; analyze the data set based on a drought assessment model to obtain a loss value, input the loss value into a SHAP model to obtain a SHAP value, and screen drought cause indicators based on the SHAP value to obtain a screening result.

[0009] This method adds the drought characteristics of the underlying surface factors and considers more comprehensive factors to improve the accuracy of drought prediction; multiple indicators are used to construct a comprehensive drought index, which can reflect the comprehensive drought index of various types of agriculture, meteorology and hydrology. The index can integrate the characteristic information of different drought types, more effectively characterize the comprehensive situation of drought, and provide a more comprehensive and accurate basis for drought monitoring; the drought assessment model is used to extract features, and the spatiotemporal features are extracted through the ConvLSTM module. Then the VGG module and the attention mechanism module are added, which can adaptively allocate weights to the extracted spatiotemporal features, enhance the ability to extract feature information, improve the model's attention to the channels and intrinsic characteristics of spatiotemporal features, and enhance the model's prediction accuracy for drought events; the SHAP model is constructed by loss value, and drought characteristics with higher contribution are screened in combination with drought levels, which can obtain more effective drought influencing factors, and the SHAP model can calculate marginal contributions to realize quantitative causal analysis of the influencing factors of drought events, and screen out an accurate, efficient drought index with spatiotemporal characteristics, which can provide effective help for the monitoring and prediction of subsequent drought events.

[0010] The underlying surface factor refers to the surface at the bottom of the atmosphere that can interact with the atmosphere during the exchange of heat and moisture. The underlying surface factors mainly include soil, grassland and water bodies, which play an important role in climate formation.

[0011] Furthermore, the specific steps of constructing a comprehensive drought index include: constructing a standardized anomaly index based on the drought characteristics, dimensionlessly processing the standardized anomaly index to obtain first data, and obtaining a correlation coefficient matrix based on the first data; obtaining principal components based on the correlation coefficient matrix and the standardized anomaly index; obtaining a contribution rate of the principal component based on the correlation coefficient matrix, and obtaining a weight coefficient of the principal component based on the contribution rate; and obtaining the comprehensive drought index based on the principal component and the weight coefficient.

[0012] Constructing a comprehensive drought index that can reflect various types of agricultural, meteorological, and hydrological droughts can explore the relationship between agricultural and hydrological drought conditions and observational data.

[0013] Furthermore, according to the calling order, the drought assessment model includes: a ConvLSTM module and a VGG module, and the VGG module includes a CBAM module.

[0014] Based on the ConvLSTM model for extracting spatiotemporal features, the VGG model and attention mechanism module are added to improve the model's attention to the channels and intrinsic features of spatiotemporal features.

[0015] Furthermore, the specific steps of obtaining the loss value include: extracting the spatiotemporal features in the data set based on the ConvLSTM module, the VGG module performing a convolution operation on the spatiotemporal features to obtain a first feature; the CBAM module obtains an attention map based on the spatiotemporal features and the first feature, and obtains the loss value based on the attention map.

[0016] Furthermore, the specific steps of obtaining the loss value based on the attention map include: obtaining a regression loss based on the attention map and a first loss function, obtaining a classification loss based on the attention map and a second loss function, and obtaining the loss value based on the regression loss and the classification loss.

[0017] An improved Smooth L1Loss-Cross Entropy Loss (cross entropy loss function) hybrid loss function is constructed to comprehensively consider the regression and drought classification characteristics of the model, and integrate spatiotemporal features in the calculation to better measure the performance of the model.

[0018] Furthermore, the specific steps for obtaining the SHAP value include: presetting the drought level, weighting the loss value based on the drought level to obtain a weighted loss value; obtaining an output result based on the weighted loss value, obtaining input conditions based on the drought characteristics, inputting the output result and the input conditions into the SHAP model to obtain the SHAP value of the drought characteristics.

[0019] The GradientExplainer interpreter is constructed based on the SHAP theory. For the baseline model parameters of the interpreter, the loss value of the fused multi-scale drought characteristics is used as the explanation target to screen drought characteristics with higher contribution, so as to solve the problem that the convolution model for multi-source spatiotemporal data is difficult to calculate the marginal contribution.

[0020] Furthermore, the specific steps for obtaining the screening results include: based on the drought level, dividing the SHAP value to obtain the level SHAP value; based on the time dimension and the space dimension, obtaining the average value of the level SHAP value to obtain the mean SHAP value; based on the SHAP value and the mean SHAP value, screening the drought cause indicators to obtain the screening results.

[0021] For each drought level, the corresponding SHAP values ​​are collected and averaged in the time and space dimensions to obtain the comprehensive average contribution of each feature and the average contribution to a specific drought level. By visualizing the spatial distribution of the cumulative values ​​of the features before the drought occurs, the spatial distribution of the drought events is verified, and the significance of different features for the occurrence of drought events is analyzed.

[0022] Furthermore, the method also includes: obtaining the correlation between the comprehensive drought index and the meteorological drought index and the agricultural drought index, and obtaining a verification result based on the correlation.

[0023] The correlation between the screened comprehensive drought index and historical drought events was verified from the perspective of temporal and spatial characteristics, and the analysis showed that the screened drought index can better monitor and predict drought events.

[0024] Furthermore, the standardized anomaly index includes a standardized precipitation anomaly index, a standardized soil moisture anomaly index and a standardized surface runoff anomaly index.

[0025] Furthermore, the first loss function is a SmoothL1Loss loss function, and the second loss function is a cross entropy loss function.

[0026] The regression and classification characteristics of the model are comprehensively considered, and spatiotemporal features are integrated in the calculation to better measure the performance of the model. The hybrid loss function can improve the robustness to outliers while maintaining the accuracy of the model.

[0027] One or more technical solutions provided by the present invention have at least the following technical effects or advantages:

[0028] 1. The drought characteristics of the underlying surface factor are added to consider more comprehensive factors to improve the accuracy of drought prediction; multiple indicators are used to construct a comprehensive drought index, which can reflect the comprehensive drought index of various types of agriculture, meteorology and hydrology. This index can integrate the characteristic information of different drought types, more effectively characterize the comprehensive situation of drought, and provide a more comprehensive and accurate basis for drought monitoring.

[0029] 2. The drought assessment model is used to extract features. The spatiotemporal features are extracted through the ConvLSTM module. Then the VGG module and the attention mechanism module are added. This can adaptively assign weights to the extracted spatiotemporal features, enhance the ability to extract feature information, improve the model's attention to the channels and intrinsic characteristics of spatiotemporal features, and enhance the model's prediction accuracy for drought events.

[0030] 3. The SHAP model is constructed based on the loss value, and drought characteristics with higher contributions are screened in combination with the drought level, so that more effective drought influencing factors can be obtained. The SHAP model can calculate the marginal contribution and realize the quantitative causal analysis of the influencing factors of drought events. An accurate, efficient drought index with spatiotemporal characteristics is screened, which can provide effective assistance for the monitoring and prediction of subsequent drought events.

[0031] 4. Construct an improved Smooth L1Loss-Cross Entropy Loss hybrid loss function, comprehensively consider the regression and drought classification characteristics of the model, and integrate spatiotemporal features in the calculation to better measure the performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the present invention, and do not constitute a limitation on the embodiments of the present invention;

[0033] Figure 1 It is a schematic flow chart of a method for screening drought cause indicators in the present invention;

[0034] Figure 2 is a schematic diagram of the structure of the drought assessment model in the present invention; wherein X1, X2...X t Represents the input data of the current time series, C0, C1…C t Represents the unit state at the previous moment, H0, H1…H t represents the hidden layer state at the previous moment, X1 inputs the corresponding C0 and H0 at the moment and then uses the zero vector for initialization, ConvLSTM represents the ConvLSTM module, conv+relu represents the convolution layer combined with the ReLu activation function, max pooling represents the maximum pooling, CBAM represents the attention mechanism, Sigmoid represents the Sigmoid activation function, Fully connectes+relu represents the fully connected layer combined with the ReLu activation function, and SDCI_pre represents the comprehensive drought index;

[0035] Figure 3 It is a schematic diagram of using the SHAP model to calculate the characteristic importance of the drought disaster assessment model in a typical drought event in 2010 in the present invention. DETAILED DESCRIPTION

[0036] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0037] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those within the scope of this description. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0038] Embodiment 1

[0039] refer to Figure 1-Figure 3This embodiment provides a method for screening drought cause indicators, the method comprising:

[0040] Acquire multi-scale drought data in time and space, and extract drought characteristics of the multi-scale drought data; for example, use China's global land surface reanalysis data (CRA), gross domestic product data (GDP), meteorological station data, land use data, disaster data and elevation data. CRA data, land use data and elevation data are all raster data, and deep learning can be used to perform image convolution operations to improve the prediction accuracy of drought events.

[0041] In this embodiment, the multi-scale drought data can also be preprocessed, and drought characteristics can be extracted based on the preprocessed multi-scale drought data. The preprocessing can include format conversion, normalization, and average value filling, etc. For example, the collected NC format data is converted into tif format data, the spatial resolution of the data is unified to 34km, and the data is normalized and averaged.

[0042] In this embodiment, drought characteristics may include soil heat flux, latent heat flux, instantaneous value of surface temperature, vegetation transpiration, total canopy water storage, instantaneous value of soil temperature, 2-meter specific humidity, 10-meter wind speed and bare soil evaporation rate, etc.

[0043] The standardized deviation index is a method used to compare the deviations between different data sets. It obtains a standardized deviation value by dividing the distance of each data point from the mean by the standard deviation. This value indicates the degree of deviation of the data point from the mean in the entire data set.

[0044] A PCA (principal component analysis) method is used to construct a comprehensive drought index based on the drought characteristics. The specific steps include: constructing a standardized anomaly index based on the drought characteristics, wherein the standardized anomaly index includes a standardized precipitation anomaly index, a standardized soil moisture anomaly index, and a standardized surface runoff anomaly index. The calculation method can be:

[0045]

[0046] Among them, StX m,n represents the standardized anomaly index, m and n represent the nth day in the mth year, and X m,n It represents the target data required on the nth day of the mth year. represents the average value of the nth day in the set year, σ n It represents the standard deviation of the nth day in the set years in the inter-annual time series.

[0047] Nondimensionalization is an important processing idea in scientific research. It removes part or all of the units of an equation involving physical quantities by replacing a suitable variable in order to simplify the experiment or calculation. Nondimensionalization is an important step in the comprehensive evaluation process, which is used to eliminate the influence of dimensions and enable indicators of different magnitudes to be compared and integrated.

[0048] The standardized deviation index is dimensionless processed to ensure that each variable has the same scale, and the first data is obtained; the calculation method thereof can be:

[0049]

[0050] Among them, x′ ij represents the jth data value in the i-th element after standardization, x ij represents the jth data value in the i-th element, σ i represents the standard deviation of the ith element, It represents the average of the i-th element, and i and j both represent serial numbers.

[0051] Based on the first data, a correlation coefficient matrix is ​​obtained; the correlation coefficient matrix between variables is calculated, and the standardized multidimensional data is formed into a two-dimensional matrix. Subsequently, the covariance matrix, i.e., the correlation coefficient matrix, is calculated based on the matrix. The calculation method can be:

[0052]

[0053] Among them, S ij represents the correlation coefficient matrix, represents the series mean of the i-th element after standardization, represents the series mean of the hth element after standardization, x′ ij represents the jth data value in the i-th element after standardization, x′ hj It represents the jth data value in the hth element after standardization, m represents the number of samples, and i and j both represent serial numbers.

[0054] The principal component is obtained based on the correlation coefficient matrix and the standardized anomaly index; the calculation method thereof can be:

[0055] D i =v1P r +v2P f +v3S m ; (4)

[0056] Among them, D i represents the principal component, P r represents the standardized precipitation anomaly index, P frepresents the standardized surface runoff anomaly index, S m represents the standardized soil moisture anomaly index, v1, v2 and v3 represent the eigenvectors of the correlation coefficient matrix of precipitation index, surface runoff index and soil moisture index respectively, and i represents the serial number.

[0057] The contribution rate of the principal component is obtained based on the correlation coefficient matrix; the calculation method thereof can be:

[0058] |S ij -λ i I|=0; (5)

[0059]

[0060] Among them, S ij represents the correlation coefficient matrix, I represents the identity matrix, λ i represents the eigenvalue, g i represents the contribution rate, p represents the number of eigenvalues, k represents the number of principal components, and i represents the sequence number.

[0061] The weight coefficient of the principal component is obtained based on the contribution rate; the calculation method thereof may be:

[0062]

[0063] Among them, w i represents the weight coefficient, gi represents the contribution rate, p represents the number of eigenvalues, and i represents the sequence number.

[0064] Based on the principal component and the weight coefficient, the comprehensive drought index is obtained. The calculation method thereof can be:

[0065]

[0066] Among them, Index represents the comprehensive drought index, w i Denotes the weight coefficient, D i represents the principal component, p represents the number of eigenvalues, and i represents the ordinal number. The lower the Index value, the more severe the drought, and the higher the value, the less severe the drought.

[0067] Obtaining a data set based on the comprehensive drought index and the multi-scale drought data;

[0068] The data set is analyzed based on a drought assessment model to obtain a loss value, wherein, in accordance with the calling order, the drought assessment model includes: a ConvLSTM module and a VGG module, and the VGG module includes a CBAM module.

[0069] ConVLSTM is an architecture that combines convolutional neural networks (CNN) and long short-term memory networks (LSTM) to process time series data. Unlike traditional LSTM, ConvLSTM applies convolution operations at each time step, which helps capture spatial information in time series data.

[0070] VGG is a deep convolutional neural network (CNN) proposed by Karen Simonyan and Andrew Zisserman of the Visual Geomety Group (VGG) at the University of Oxford in 2014. VGG is famous for its simplicity and consistency of structure, which is mainly composed of stacked small 3x3 convolution kernels and 2x2 maximum pooling layers.

[0071] Convolutional Attention Module (CBAM) is a feed-forward convolutional neural network attention module. Given an intermediate feature map, CBMA sequentially infers attention maps along two separate dimensions (channel and spatial), and then multiplies the attention map by the input feature map for adaptive feature refinement. Moreover, CBAM is a lightweight universal module that can be seamlessly integrated into any CNN architecture.

[0072] ConvLSTM uses convolution operations to extract spatial pattern information from images, extracts the spatiotemporal features of input data, and converts them into tensors. At the same time, the convolution gate analyzes the temporal information in a similar way to the traditional LSTM. ConvLSTM can achieve this goal with fewer parameters and higher computational efficiency. VGG is used to replace the fully connected layer in the ConvLSM model, and the CBAM module is introduced in the VGG model. After adding the convolution layer in VGG, the channel attention features are first extracted, and then the spatial attention features are extracted. Finally, the data with the extracted features is put back into the convolution layer of VGG for calculation, and finally VGG outputs the prediction results. Using the improved SmoothL1Loss-CrossEntropyLoss as the loss function when training the drought assessment model can solve the problem that the convolution model for multi-source spatiotemporal data is difficult to calculate the marginal contribution.

[0073] The specific steps of obtaining the loss value include: extracting the spatiotemporal features in the data set based on the ConvLSTM module; and extracting the temporal features and spatial features in the data set using ConvLSTM. The calculation method can be:

[0074]

[0075] Among them, f t Represents the output of the forget gate, x t Indicates the input at the current moment, h t-1represents the hidden layer state at the previous moment, C t-1 represents the unit state at the previous moment, W xf , W hf and W cf Both represent the weight of the forget gate, b f represents the bias term of the forget gate, σ represents the activation function, i t and Respectively represent the two output results of the input gate, W xi , W hi and W ci Both represent the weight of the input gate, b i represents the bias term of the input gate, tanh represents the activation function, and W xc and W hc Both represent the weight of the unit state, b c The bias term representing the cell state, C t represents the current cell state, ot represents the output result of the output gate, W xo , W ho and W co Both represent the weight of the output gate, b o represents the bias term of the output gate, h t represents the hidden layer state at the current moment, represents element-wise multiplication, also known as the Hadamard product, and * represents convolution.

[0076] The VGG module performs a convolution operation on the spatiotemporal features to obtain a first feature; VGG is used to perform a convolution operation again on the spatiotemporal feature tensors extracted by ConvLSTM to extract their spatial features, thereby enhancing the model's ability to extract spatial features.

[0077] The CBAM module obtains an attention map based on the spatiotemporal feature and the first feature, and obtains the loss value based on the attention map. The CBAM convolutional attention module is introduced into the VGG module, and the attention map is inferred from two different dimensions, channel and space, and the obtained attention map is multiplied with the input feature map to perform adaptive feature refinement. This step can improve the attention of the entire model to different channels and enhance the prediction accuracy of the model for drought events.

[0078] Since the prediction results are based on the regression value of the comprehensive drought index and the drought level can be classified according to the range of the regression value, the loss function of the drought assessment model is improved.

[0079] The specific steps of obtaining the loss value based on the attention map include:

[0080] The regression loss is obtained based on the attention map and the first loss function; for the regression loss, the SmoothL1Loss loss function with a penalty coefficient is used for calculation. For the prediction results with a greater distance between the regression value range and the label, a larger penalty coefficient is used to reflect the loss impact of the drought level classification based on the regression value range on the prediction results. The calculation method can be:

[0081] l(x,y)=L={l1,…,l N} T ; (10)

[0082]

[0083] Among them, l() represents the SmoothL1Loss loss function, L represents the loss vector, l N Represents the loss value of a single sample, T represents transposition, x represents the predicted value of the model, y represents the true label value, and SmoothL1Loss() represents the SmoothL1Loss loss function.

[0084] Based on the attention map and the second loss function, the classification loss is obtained; for the classification loss, the Cross Entropy Loss loss function with a penalty coefficient is used for calculation. The regression value divided by drought level is converted into a one-hot value as a score (Logits), and the score is weighted by the penalty coefficient based on the distance between the regression value and the label, and the weighted result is calculated using cross entropy. The calculation method can be:

[0085]

[0086] Among them, H() represents the cross entropy loss function, y1 represents the one-hot encoding vector of the true label, x1 represents the probability distribution vector predicted by the model, C represents the number of categories, and y i represents the one-hot encoding vector of the i-th label, x i Represents the probability of the i-th category predicted by the model, where i represents the sequence number.

[0087] The loss value is obtained based on the regression loss and the classification loss.

[0088] Among them, the first loss function is the SmoothL1Loss loss function, and the second loss function is the cross entropy loss function.

[0089] Smooth L1 Loss is a quadratic function when |xy|<1, which makes it more sensitive to small errors and can better fit the data; and it is a linear function when |xy|≥1, which makes it relatively less penalized for large errors (outliers) and shows better robustness in the presence of outliers. When dealing with regression tasks, Smooth L1 Loss treats all errors in a more balanced way and does not overly focus on a certain part of the data. This is especially beneficial for small class samples in unbalanced data sets, which can prevent the model from overfitting to the majority class samples.

[0090] The logarithmic function in the cross entropy loss function has a larger penalty for small probability values ​​and a smaller penalty for large probability values. This means that when the model predicts a very low probability for a sample (i.e., a prediction error), the loss will increase significantly, but the impact of outliers will not be overly amplified. The cross entropy loss function can also handle unbalanced data by weighting the losses of different categories. For example, for minority class samples, higher weights can be given so that the model pays more attention to the prediction of minority class samples. Therefore, the combination of these two loss functions has unique advantages in handling outliers and unbalanced data, which is very important for processing complex drought data.

[0091] Smooth L1 Loss is a commonly used loss function, mainly used for regression tasks. It is a combination of L1 Loss and L2 Loss, aiming to balance the advantages of both while overcoming their shortcomings.

[0092] By combining the losses of regression and classification tasks, the loss function reflects both the performance of the model in drought prediction and the accuracy of the model in classifying different drought levels, thus providing a more comprehensive drought disaster assessment.

[0093] The loss value is input into the SHAP model to obtain the SHAP value; since the drought characteristics in different regions may be different, by combining the loss function of multiple features for weighted processing, the interpreter can better adapt to the drought characteristics of different regions and improve the generalization ability of the model. At the same time, since the output of SHAP must be a scalar when using certain gradient calculation methods in the calculation process, using the loss value as the output of the explanation model can solve the problem that the convolution model of multi-source spatiotemporal data is difficult to calculate the marginal contribution.

[0094] The specific steps of obtaining the SHAP value include: presetting a drought level, weighting the loss value based on the drought level, and obtaining a weighted loss value;

[0095] The prediction results in different types of loss functions are weighted according to the drought level. The regression loss function uses a dynamic weighting method to dynamically adjust the weight according to the error between the predicted value and the true value. The weighting rule can be:

[0096] R_Loss=(Loss1*0.1+1)·w; (13)

[0097] Among them, R_Loss represents, Loss1 represents the loss value output by the Smooth L1 Loss function, and w represents the fixed weight corresponding to the error distance between the predicted value and the true value after the drought level division.

[0098] The classification loss function uses an exponential function to adjust the weights so that the weights of categories with larger errors grow faster. The weighting rule can be:

[0099] C Loss =Loss2·(1+e β·(d-1) ), β>0; (14)

[0100] Among them, C_Loss represents, Loss2 represents the loss value output by the cross entropy loss function, β represents the parameter that controls the weight growth rate, and d represents the drought level distance between the predicted value and the true value.

[0101] The weighting mechanism enhances the sensitivity to different drought levels, ensuring that the model pays more attention to drought levels with larger prediction errors, making the model more flexible and effective in processing drought data, which helps to improve the accuracy and practicality of the model in practical applications; dynamically adjust the weights according to the performance of the model to further optimize the model performance and ensure that the model can be continuously improved during the training process; the drought characteristics of different regions may be different. By combining multiple features and weighting them, the mixed loss function can better adapt to the drought characteristics of different regions and improve the generalization ability of the model.

[0102] An output result is obtained based on the weighted loss value, an input condition is obtained based on the drought characteristic, and the output result and the input condition are input into the SHAP model to obtain a SHAP value of the drought characteristic. The calculation method can be:

[0103]

[0104] Among them, φj() represents the SHAP value of feature j for input x2, f represents the prediction function of the model, x2 represents the input sample, that is, the input feature vector of the model, S represents the feature subset that does not contain the feature, M represents the total number of features, N represents the set of all features, |S| represents the number of elements in the set S, and f x (S) represents the predicted value of the model when only the features in set S are used, fx (S∪{j}) represents the predicted value of the model when only the features in set S and feature j are used, ! represents the factorial, and j represents the feature.

[0105] SHAP (SHapleyAdditive exPlanations) is a game theory method for explaining the output of any machine learning model. It uses the classic game theory Shapley value and its related extensions to combine optimal credit assignment with local explanation. The closer the SHAP value is to zero, the smaller the contribution of the feature to the prediction; and the farther the SHAP value is from zero, the greater the contribution of the feature to the prediction. To interpret the model in SHAP, you need to create an explainer first. SHAP supports many types of explainers, such as deep, gradient, kernel, tree, and sampling.

[0106] The SHAP model provides an intuitive way to understand how features affect model predictions, providing interpretability for the drought factor screening process in this method.

[0107] In the SHAP value visualization process, in order to further analyze the importance of features under different drought levels, the prediction results of the model are divided according to the defined drought intervals. For each drought level, the corresponding SHAP values ​​are collected and averaged in the time and space dimensions to obtain the comprehensive average contribution of each feature and the average contribution to a specific drought level.

[0108] The drought cause index is screened based on the SHAP value to obtain a screening result. The specific steps include: based on the drought level, the SHAP value is divided to obtain a grade SHAP value; for example, the prediction result of the model is divided into no drought, light drought, moderate drought, severe drought and extreme drought, thereby dividing the SHAP value; based on the time dimension and the space dimension, the average value of the grade SHAP value is obtained to obtain a mean SHAP value; based on the SHAP value and the mean SHAP value, the drought cause index is screened to obtain the screening result.

[0109] Embodiment 2

[0110] On the basis of the first embodiment, in this embodiment, the method further includes:

[0111] The correlation between the comprehensive drought index and the meteorological drought index and the agricultural drought index is obtained, and a verification result is obtained based on the correlation.

[0112] For example, the SPI index representing meteorological drought is selected as the verification target, the relationship between SDCI (comprehensive drought index) and SPI is analyzed, and the correlation coefficient between SDCI and SPI within a certain period of time is calculated based on Pearson correlation analysis; the soil relative humidity index representing agricultural drought is selected to calculate the Pearson correlation coefficient with SDCI; the trend of changes in the total amount of surface water resources is selected for analysis, and the trend of changes in the SDCI drought area within a certain period of time is statistically analyzed.

[0113] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0114] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for screening drought cause indicators, characterized in that: The method comprises: Acquire multi-scale drought data in time and space, and extract drought characteristics of the multi-scale drought data; Based on the drought characteristics, construct a comprehensive drought index; based on the comprehensive drought index and the multi-scale drought data, obtain a data set; The data set is analyzed based on a drought assessment model to obtain a loss value, the loss value is input into a SHAP model to obtain a SHAP value, and drought cause indicators are screened based on the SHAP value to obtain a screening result.

2. A method for screening drought cause indicators according to claim 1, characterized in that: The specific steps to construct a comprehensive drought index include: constructing a standardized anomaly index based on the drought characteristics, performing dimensionless processing on the standardized anomaly index to obtain first data, and obtaining a correlation coefficient matrix based on the first data; Obtaining principal components based on the correlation coefficient matrix and the standardized anomaly index; Obtaining a contribution rate of the principal component based on the correlation coefficient matrix, and obtaining a weight coefficient of the principal component based on the contribution rate; Based on the principal components and the weight coefficients, the comprehensive drought index is obtained.

3. A method for screening drought cause indicators according to claim 1, characterized in that: According to the calling order, the drought assessment model includes: a ConvLSTM module and a VGG module, and the VGG module includes a CBAM module.

4. A method for screening drought cause indicators according to claim 3, characterized in that: The specific steps to obtain the loss value include: Based on the ConvLSTM module, the spatiotemporal features in the data set are extracted, and the VGG module performs a convolution operation on the spatiotemporal features to obtain a first feature; the CBAM module obtains an attention map based on the spatiotemporal features and the first feature, and obtains the loss value based on the attention map.

5. A method for screening drought cause indicators according to claim 4, characterized in that: The specific steps of obtaining the loss value based on the attention map include: A regression loss is obtained based on the attention map and the first loss function, a classification loss is obtained based on the attention map and the second loss function, and the loss value is obtained based on the regression loss and the classification loss.

6. A method for screening drought cause indicators according to claim 5, characterized in that: The specific steps to obtain the SHAP value include: Preset a drought level, and weight the loss value based on the drought level to obtain a weighted loss value; An output result is obtained based on the weighted loss value, an input condition is obtained based on the drought characteristic, and the output result and the input condition are input into the SHAP model to obtain a SHAP value of the drought characteristic.

7. A method for screening drought cause indicators according to claim 6, characterized in that: The specific steps to obtain screening results include: Based on the drought level, dividing the SHAP value to obtain a level SHAP value; Based on the time dimension and the space dimension, obtaining the average of the level SHAP values ​​to obtain the mean SHAP value; Based on the SHAP value and the mean SHAP value, drought cause indicators are screened to obtain the screening result.

8. A method for screening drought cause indicators according to claim 1, characterized in that: The method further comprises: The correlation between the comprehensive drought index and the meteorological drought index and the agricultural drought index is obtained, and a verification result is obtained based on the correlation.

9. A method for screening drought cause indicators according to claim 2, characterized in that: The standardized anomaly index includes the standardized precipitation anomaly index, the standardized soil moisture anomaly index and the standardized surface runoff anomaly index.

10. A method for screening drought cause indicators according to claim 5, characterized in that: The first loss function is a SmoothL1Loss loss function, and the second loss function is a cross entropy loss function.

Images