Daily rainfall product downscaling and fusion cooperative computing method and system based on deep learning

Abstract

The invention provides a daily rainfall product downscaling and fusion cooperative calculation method and system based on deep learning, and belongs to the technical field of data processing, and the method comprises the steps: generating a daily scale data set and a monthly scale data set based on remote sensing rainfall products from various sources, meteorological station data, and vegetation and topographic data; constructing a data space-time fusion and downscaling model, and respectively training and respectively predicting the daily scale data set and the monthly scale data set by using the data space-time fusion and downscaling model to obtain daily scale precipitation estimation and monthly scale precipitation estimation; the daily scale precipitation estimation is corrected using the monthly scale precipitation estimation on the monthly scale. According to the scheme, downscaling is carried out on the monthly scale and the daily scale respectively, vegetation information data is added into the monthly downscaling to calculate the accurate monthly precipitation, the daily downscaling is used for estimating the proportion of the daily precipitation in the whole monthly precipitation, the two are multiplied to obtain the final daily precipitation, and the accuracy of daily scale precipitation estimation is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, system and electronic device for collaborative computing of downscaling and fusion of daily precipitation products based on deep learning. Background Art

[0002] High-resolution daily precipitation data are essential for understanding regional and global precipitation patterns, simulating future climate scenarios, and assessing the ecological impacts of climate change. However, due to the strong temporal and spatial heterogeneity of daily precipitation, obtaining high-resolution daily precipitation data still faces major challenges. Traditional daily precipitation data production relies on ground monitoring data from meteorological observation stations. Although these station data can provide accurate precipitation estimates, their spatial information is limited. Specifically, meteorological observation stations are very unevenly distributed in space, and the density of stations is sparse in rural, mountainous, and forest areas, which greatly limits their reconstruction of the spatiotemporal changes in precipitation patterns. Advances in remote sensing technology provide an important source of information for obtaining high-resolution, spatially continuous daily precipitation data.

[0003] In recent years, remote sensing precipitation products based on a variety of satellite sensors have been released one after another, such as Global Satellite Mapping of Precipitation (GSMaP), Integrated Multi-satellite Reciprievals for GPM (IMERG), Climate Hazards center InfraRed Precipitation with Station data (CHIRPS) and CPC MORPHing technique (CMORPH). According to previous evaluations, GSMaP, IMERG and CMORPH have shown good daily precipitation accuracy in China. However, the resolution and accuracy of existing precipitation products are still insufficient to meet the needs of finer regional scales, especially for meteorological and hydrological studies in small areas.

[0004] Statistical downscaling methods combined with high-resolution environmental variables have been widely used in the production of daily precipitation data. Since precipitation is considered to be closely related to multiple environmental factors such as topography and vegetation, many methods have been developed to downscale monthly and annual precipitation products using high-resolution environmental data. For example, geographically weighted regression (GWR) was introduced into the downscaling of TRMM data at the monthly scale, thereby taking into account the spatial heterogeneity of the relationship between precipitation and the normalized difference vegetation index (NDVI) and digital elevation model (DEM). Studies have shown that the downscaling accuracy of support vector machine (SVM) is not only better than that of methods including random forest (RF), GWR, multivariate linear regression (MLR) and exponential regression (ER), but also has good consistency with monthly ground observation data compared to the original TRMM data; currently, these related studies mainly focus on precipitation data at the annual and monthly scales, because there is a significant correlation between precipitation and local environmental factors at these time scales. However, at finer time scales (such as daily scales), this relationship usually changes significantly or even does not exist.

[0005] To address this problem, some new methods have been introduced. A common strategy is to first accumulate precipitation estimates to a monthly or annual scale and establish their relationship with local environmental variables. After coarse-scale spatial downscaling, the daily precipitation estimates are temporally downscaled. This strategy assumes that the proportion of daily precipitation to monthly precipitation can be accurately detected at the original spatial resolution. However, this assumption is often difficult to achieve due to the limited ability of precipitation products in daily precipitation, especially extreme precipitation. Omissions or misreporting of extreme precipitation in precipitation products may lead to significant biases in daily precipitation estimates for the entire month.

[0006] Therefore, it is necessary to provide an improved technical solution to address the above-mentioned deficiencies in the prior art. Summary of the invention

[0007] The purpose of this application is to provide a method, system and electronic device for downscaling and fusion collaborative calculation of daily precipitation products based on deep learning, so as to solve or alleviate the problem of insufficient accuracy of existing daily-scale precipitation estimation, so as to produce daily precipitation products with higher resolution and higher accuracy.

[0008] In order to achieve the above objectives, this application provides the following technical solutions: In the first aspect, the present application provides a method for downscaling and fusion collaborative calculation of daily precipitation products based on deep learning, including: Generate a daily-scale dataset and a monthly-scale dataset based on remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources; wherein the monthly-scale dataset contains vegetation information, and the daily-scale dataset does not contain vegetation information; Based on deep learning, a data spatiotemporal fusion and downscaling model is constructed, and the data spatiotemporal fusion and downscaling model is trained separately using the daily scale data set and the monthly scale data set to obtain a trained model, and the trained model is used to perform predictions respectively to obtain a daily scale precipitation estimate and a monthly scale precipitation estimate; The spatial resolution of the daily-scale precipitation estimation and the monthly-scale precipitation estimation is the target resolution, and the target resolution is higher than the original spatial resolution of the remote sensing precipitation product. The data spatiotemporal fusion and downscaling model includes: a residual network ResNet module, a bidirectional long short-term memory network BiLSTM module and an attention mechanism module; the ResNet module is used to learn the spatial neighborhood features of the input data of each time step and input it to the BiLSTM module; the BiLSTM module is used to capture the temporal contextual relationship of the spatial neighborhood features, and the attention mechanism module is used to weight the output of the BiLSTM module to generate a final model output; On a monthly scale, the monthly scale precipitation estimate is used to correct the daily scale precipitation estimate to obtain high-resolution daily precipitation data.

[0009] In some possible implementations, generating a daily scale dataset and a monthly scale dataset based on the remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from the multiple sources includes: Performing spatial interpolation processing on the remote sensing precipitation products from the multiple sources to obtain spatially interpolated precipitation data; Resampling the vegetation and terrain data to obtain resampled vegetation and terrain data; The spatial resolutions of the precipitation data after spatial interpolation and the vegetation and terrain data after resampling are both the target resolution; Perform time scale accumulation processing on the interpolated precipitation data to obtain daily and monthly precipitation data; Based on the daily precipitation data and the monthly precipitation data, taking each observation station in the meteorological observation station data as the center, extracting the daily precipitation and the monthly precipitation of the surrounding N×N pixels, and extracting the vegetation information and the terrain information of the surrounding N×N pixels from the resampled vegetation and terrain data; Combining the daily-scale precipitation of N×N pixels with the terrain information to generate the daily-scale dataset; combining the monthly-scale precipitation of N×N pixels with the vegetation information and the terrain information to generate the monthly-scale dataset; Wherein, N is a positive integer greater than 1.

[0010] In some possible implementations, the terrain data is DEM data. Accordingly, taking each observation station in the meteorological observation station data as the center, the terrain information of N×N pixels around it is extracted, specifically: Taking each observation station in the meteorological observation station data as the center, the DEM data is used to extract the elevation, slope and slope direction of N×N pixels around the center point.

[0011] In some possible implementations, the vegetation information includes: Normalized Difference Vegetation Index NDVI.

[0012] In some possible implementations, the ResNet module includes a plurality of residual blocks, and the plurality of residual blocks are used to continuously compress the input data of each time step in the spatial dimension to learn its spatial neighborhood features; The BiLSTM module includes a plurality of BiLSTM layers connected in sequence, which are used to receive the spatial neighborhood features learned by the ResNet module and process them in sequence to obtain the output, hidden layer state and cell state of the corresponding layer; The attention mechanism module is used to multiply the output of the final layer of the BiLSTM module by the final hidden layer state and normalize it to obtain an attention weight, and apply the attention weight to the output of the final layer to generate the final model output.

[0013] In some possible implementations, the data spatiotemporal fusion and downscaling model further includes: a fully connected layer and a Dropout layer sequentially connected to the attention mechanism module.

[0014] In some possible implementations, after generating the daily scale dataset and the monthly scale dataset, the method further includes: dividing the daily scale dataset and the monthly scale dataset to obtain corresponding training data, verification data, and test data; Accordingly, the data spatiotemporal fusion and downscaling model is trained, including: Input the training data into the data spatiotemporal fusion and downscaling model, and repeatedly adjust the hyperparameters according to the accuracy evaluation index value of the data spatiotemporal fusion and downscaling model on the verification data until the accuracy index value reaches the preset accuracy condition, and output the trained data spatiotemporal fusion and downscaling model; The trained data spatiotemporal fusion and downscaling model is used to predict the test data, and the unbiased accuracy of the trained data spatiotemporal fusion and downscaling model is measured according to the accuracy evaluation index of the prediction result.

[0015] In some possible implementations, on a monthly scale, the daily scale precipitation estimate is corrected using the monthly scale precipitation estimate to obtain high-resolution daily precipitation data, specifically: For any specified month, normalizing the time series data consisting of the daily precipitation estimation values ​​of the specified month to obtain a normalized processing result; The normalized result is multiplied by the estimated monthly precipitation value corresponding to the specified month to obtain high-resolution daily precipitation data.

[0016] In a second aspect, this embodiment provides a daily precipitation product downscaling and fusion collaborative computing system based on deep learning, including: A data acquisition unit is configured to generate a daily scale data set and a monthly scale data set based on remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources; wherein the monthly scale data set includes vegetation information, and the daily scale data set does not include vegetation information; A model building and prediction unit is configured to process the daily scale data set and the monthly scale data set respectively using a pre-built data spatiotemporal fusion and downscaling model to obtain a daily scale precipitation estimate and a monthly scale precipitation estimate respectively; The spatial resolution of the daily-scale precipitation estimation and the monthly-scale precipitation estimation is the target resolution, and the target resolution is higher than the original spatial resolution of the remote sensing precipitation product. The data spatiotemporal fusion and downscaling model includes: a residual network ResNet module, a bidirectional long short-term memory network BiLSTM module and an attention mechanism module; the ResNet module is used to learn the spatial neighborhood features of the input data of each time step and input it into the BiLSTM module; the BiLSTM module is used to capture the temporal contextual relationship of the spatial neighborhood features, and the attention mechanism module is used to weight the output of the BiLSTM module to generate the daily scale precipitation estimate and the monthly scale precipitation estimate; The correction unit is configured to correct the daily-scale precipitation estimate on a monthly basis using the monthly-scale precipitation estimate to obtain high-resolution daily precipitation data.

[0017] In a third aspect, this embodiment provides an electronic device, comprising: a memory for storing instructions executed by one or more processors of the electronic device; and a processor, which, when the processor executes the instructions in the memory, enables the electronic device to implement the steps of the method described in any of the above embodiments.

[0018] The deep learning-based downscaling and fusion collaborative calculation method for daily precipitation products provided in the embodiment of the present application first fuses and downscales the daily and monthly precipitation, that is, directly models the daily and monthly precipitation data to obtain their respective precipitation estimates, laying the foundation for subsequent deviation correction. A data spatiotemporal fusion and downscaling model combining the residual network ResNet, the bidirectional long short-term memory network BiLSTM and the attention mechanism (ResNet-BiLSTM-Attention, referred to as the RNBLA model) is adopted. The model takes into account the relationship between precipitation and vegetation and terrain, and integrates data from a variety of remote sensing precipitation products and meteorological observation stations to improve the accuracy of precipitation estimation. Since the correlation between local environmental factors and daily precipitation is not significant on the daily scale, the daily precipitation estimate directly generated by the model is biased. However, there is a relatively stable relationship between monthly precipitation and environmental factors, so the generated monthly precipitation estimate is more accurate. Therefore, the generated monthly precipitation estimate is used to correct the deviation of daily precipitation on the monthly scale to avoid errors caused by underreporting or false reporting of extreme precipitation under traditional downscaling methods, thereby obtaining daily precipitation data products with higher resolution and higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of the process of downscaling and fusion collaborative calculation method of daily precipitation products based on deep learning.

[0020] Figure 2 A logic diagram of a deep learning-based collaborative calculation method for downscaling and fusion of daily precipitation products provided according to some embodiments of the present application.

[0021] Figure 3 Schematic diagram of the single model input area range for the data spatiotemporal fusion and downscaling model.

[0022] Figure 4 Schematic diagram of the overall structure of the data spatiotemporal fusion and downscaling model.

[0023] Figure 5 This is a schematic diagram of the specific structure of the data spatiotemporal fusion and downscaling model.

[0024] Figure 6 A schematic diagram of the structure of an electronic device provided in some embodiments. DETAILED DESCRIPTION

[0025] In order to facilitate understanding of the technical solution of the present application, the relevant terms are exemplarily explained below.

[0026] (1) Remote sensing precipitation products: Remote sensing precipitation products refer to data sets on the distribution and intensity of precipitation on the earth's surface that are collected, processed and inverted by remote sensing sensors carried on satellites. These products use satellite multi-band observations of the atmosphere and surface (such as visible light, infrared, microwaves, etc.) combined with numerical models and algorithms to generate continuous precipitation information in time and space.

[0027] As described in the background technology, at present, the remote sensing precipitation products covering the global range include: GSMaP, IMERG, CHIRPS and CMORPH, among which GSMaP is called Satellite Precipitation Mapping in Chinese, with a spatial resolution of 0.1° (about 11km) and a temporal resolution of half an hour / hour; IMERG is called Global Precipitation Measurement Multi-Satellite Integrated Inversion in Chinese, and its spatial resolution and temporal resolution are the same as GSMaP, which are 0.1° and half an hour / hour respectively; CHIRPS is called Climate Disaster Center Infrared Precipitation and Site Data Fusion Product in Chinese, with a spatial resolution of 0.05° (about 5.5km) and a temporal resolution of day / month; CMORPH is called CPC Morphological Interpolation Technology Precipitation Product in Chinese, with a spatial resolution of 8km and a temporal resolution of half an hour / hour. It can be seen that compared with traditional ground monitoring data that rely on meteorological observation stations, various remote sensing precipitation products can provide precipitation data with large coverage and spatial continuity. However, the spatial resolution of these remote sensing precipitation products is only 5.5km at most, and local-scale research usually requires a higher spatial resolution (such as a spatial resolution of 0.01° (about 1.1km) or higher). Therefore, the existing remote sensing precipitation products are difficult to meet the needs of local-scale research.

[0028] (2) Original spatial resolution and target spatial resolution: The original spatial resolution is the spatial resolution of the remote sensing precipitation product, and the target spatial resolution is the spatial resolution of the precipitation estimate obtained after downscaling. In the downscaling process, the original spatial resolution usually refers to a low spatial resolution (such as 0.1°), and the target spatial resolution is a high spatial resolution (such as 0.01°).

[0029] (3) Statistical downscaling method: The statistical downscaling method is a technology that converts low-resolution precipitation data (such as 0.1° spatial resolution of remote sensing precipitation products) into high-resolution (such as 0.01° or higher) precipitation data through mathematical and statistical models. The core idea is to use the statistical relationship between environmental variables (such as topography, vegetation index, etc.) and precipitation to spatially refine the low-resolution precipitation data, thereby generating a more refined and higher-resolution precipitation distribution.

[0030] However, the premise for the implementation of the statistical downscaling method is that there is a significant correlation between precipitation and local environmental factors, but this premise faces limitations in terms of time scale. Specifically, on the annual and monthly scales, there is a significant correlation between precipitation and local environmental factors, while on the daily scale, the correlation between precipitation and local environmental factors decreases or even does not exist.

[0031] To this end, existing technologies use data accumulation strategies to obtain high-precision daily precipitation data, that is, first accumulate precipitation estimates to monthly or annual scales, and establish their relationship with local environmental variables, and then use spatial and temporal downscaling to obtain daily precipitation data. However, when extreme precipitation events occur, such data accumulation strategies often lead to large daily precipitation estimation biases.

[0032] In view of this, the method provided in this application proposes a novel daily precipitation downscaling strategy, which adopts a more advanced deep learning module (a combination of residual network (ResNet), bidirectional long short-term memory network BiLSTM and attention mechanism (Attention) to form a deep neural network, and constitutes a data spatiotemporal fusion and downscaling model (referred to as: model). The model is first used to estimate the monthly scale and daily scale precipitation respectively, and then the generated monthly scale precipitation estimate is used to perform bias correction on the daily scale precipitation, so as to reduce the errors caused by underreporting or false reporting of extreme precipitation in remote sensing precipitation products, thereby producing daily precipitation products with higher resolution and better accuracy.

[0033] The embodiments of the present application are described below in conjunction with the accompanying drawings.

[0034] The embodiments of the present application can be applied to Figure 6 Among the electronic devices shown, the electronic device can be but is not limited to mobile terminals such as mobile phones, tablet computers, handheld computers, personal digital assistants (PDAs), smart home devices such as smart TVs and smart cameras, wearable devices such as smart bracelets, smart watches, smart glasses, or other computer devices such as desktops, laptops, notebook computers, ultra-mobile personal computers (UMPCs), netbooks, smart screens, etc.

[0035] like Figure 6 As shown, the electronic device 200 may include one or more of the following components: a processor 201, a memory 203, a communication interface 202, and a communication bus 204. The memory 203 may be connected to the processor 201 via the bus 204. The bus may transmit data between the processor 201 and the memory 203. The bus may be divided into an address bus, a data bus, a control bus, and the like.

[0036] The processor 201 may include one or more processing cores, and the processor 201 may use various interfaces and lines to connect various parts of the entire electronic device 200, and execute various functions and process data of the electronic device 200 by running or executing instructions, programs, code sets or instruction sets stored in the memory 203, and calling data stored in the memory 203. Exemplarily, the processor 201 may include an application processor (application processor, AP), a modem processor, a CPU, a graphics processor (graphics processing unit, GPU), an image signal processor (image signal processor, ISP), a controller, a video codec, a digital signal processor (digital signal processor, DSP), a field programmable gate array (field-programmable gate array, FPGA), a programmable logic array (programmable logic array, PLA) and / or a neural network processor (neural-network processing unit, NPU), etc. Among them, the CPU mainly processes the operating system, user interface and application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed; the NPU is used to implement artificial intelligence (artificial intelligence, AI) functions; and the modem is used to process wireless communications. Different processing units may be independent devices or integrated into one or more processors. For example, the multiple processing units shown above are integrated into one SoC, or the AP is a separate semiconductor chip and other processing units are integrated into one SoC, which is not limited in this application.

[0037] The memory 203 may include a random access memory (RAM), a read-only memory (ROM), or a non-transitory computer-readable storage medium. The memory 203 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 203 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, such as a daily precipitation product downscaling and fusion collaborative computing method based on deep learning, etc.; the data storage area may store data collected or generated according to the use of the electronic device 200, such as remote sensing precipitation products from multiple sources, meteorological observation station data, vegetation and terrain data input data, or generated daily scale data sets, monthly scale data sets, daily scale precipitation estimates, monthly scale precipitation estimates, etc.

[0038] In addition, those skilled in the art can understand that the structure of the electronic device 200 shown in the above drawings does not constitute a limitation on the electronic device 200, and the electronic device may include more or fewer components than shown in the drawings, or combine certain components, or arrange the components differently. For example, the electronic device 200 also includes components such as a microphone, a speaker, a radio frequency circuit, a sensor, an audio circuit, a power supply, and a Bluetooth module, which will not be described in detail here.

[0039] This embodiment provides a downscaling and fusion collaborative calculation method for daily precipitation products based on deep learning. Figure 1 As shown, the method comprises the following steps: Step S101: Generate daily-scale datasets and monthly-scale datasets based on remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources.

[0040] In this embodiment, the data set preparation requires collecting remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources.

[0041] Among them, remote sensing precipitation products include mainstream GSMaP, IMERG and CMORPH, and terrain data, vegetation data and meteorological observation station data of the study area are obtained at the same time. The above data can be downloaded from their respective official websites or other data providers. Among them, GSMaP, IMERG and CMORPH data are specifically: precipitation data with half-hour / hour time resolution from July 2000 to June 2022. The spatial resolution of GSMaP, IMERG and CMORPH refers to the previous article and will not be repeated here.

[0042] Topographic data includes: DEM data, vegetation information includes: Normalized Difference Vegetation Index NDVI.

[0043] In some embodiments, step S101 generates a daily scale dataset and a monthly scale dataset based on remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources, including: Step S111: perform spatial interpolation processing on precipitation products from multiple sources to obtain precipitation data after spatial interpolation.

[0044] Step S121 : resample the vegetation and terrain data to obtain resampled vegetation and terrain data.

[0045] Among them, the spatial resolution of precipitation data after spatial interpolation and vegetation and terrain data after resampling are all the target spatial resolution.

[0046] The purpose of the above steps S111 and S121 is to pre-process the collected data, illustratively: For remote sensing precipitation products: Use the bilinear interpolation method to interpolate the spatial resolution of all remote sensing precipitation products to the target spatial resolution, such as 0.01°.

[0047] In terms of environmental variables, NDVI and DEM were resampled to the target spatial resolution, e.g., 0.01°, and the longitude and latitude of each grid center were calculated.

[0048] Step S131: Perform time scale accumulation processing on the interpolated precipitation data to obtain daily scale precipitation data and monthly scale precipitation data.

[0049] That is to say, after spatial interpolation of the remote sensing precipitation products, on the time scale, the precipitation estimates from 12:00 noon of the previous day to 12:00 noon of today (UTC) in the remote sensing precipitation products are accumulated to obtain the accumulated daily precipitation estimates (i.e., daily precipitation data). The daily precipitation estimates are accumulated on a monthly scale to obtain monthly precipitation data. Such processing is beneficial to matching the precipitation data with the daily precipitation observations of China's meteorological observation stations.

[0050] Step S141: Based on the daily precipitation data and the monthly precipitation data, taking each observation station in the meteorological observation station data as the center, extract the daily precipitation and monthly precipitation of the surrounding N×N pixels, and extract the vegetation information and terrain information of the surrounding N×N pixels from the resampled vegetation and terrain data, where N is a positive integer greater than 1.

[0051] Here, the purpose of step S141 is to incorporate the information of the neighboring areas of the meteorological observation station into the model prediction. Specifically, taking each observation station in the meteorological observation station data as the center, the daily precipitation and monthly precipitation of the surrounding N×N pixels are extracted as the single model input area.

[0052] The size of the subgrid (i.e., input area) that meets the model input requirements is a trade-off between computational cost and accuracy. After testing and verification, 49×49 pixels can enable the model to achieve the expected accuracy while maintaining computational efficiency. Therefore, as a preferred embodiment, the value of N is 49, that is, 49×49 (0.49°×0.49°) pixels around each observation site are extracted as a single model input area.

[0053] Figure 3 A schematic diagram of the single model input area range of the data spatiotemporal fusion and downscaling model is shown, where the left figure shows the result of spatial interpolation and time scale accumulation of remote sensing precipitation products, resampling of vegetation and terrain data, and superposition with meteorological observation stations (black dots in the left figure). Taking each meteorological observation station (such as meteorological observation station i) as the center, the 49×49 pixels around it (the black grid in the left figure) are extracted to obtain the neighborhood range shown in the right figure, which is used as the single input area of ​​the model.

[0054] In some embodiments, the terrain information specifically includes: elevation, slope, and slope direction. Therefore, in step S141, taking each observation site in the meteorological observation station data as the center, the terrain information of the N×N pixels around it is extracted. Specifically, taking each observation site in the meteorological observation station data as the center, the elevation, slope, and slope direction of the N×N pixels around the center point are extracted using DEM data.

[0055] Step S151, combining the daily-scale precipitation of N×N pixels with the terrain information to generate a daily-scale data set; combining the monthly-scale precipitation of N×N pixels with the vegetation information and terrain information to generate a monthly-scale data set.

[0056] It should be noted that, since the time scale of vegetation information (such as NDVI data) is usually monthly, it can only be directly applied to the fusion and downscaling of monthly precipitation. Therefore, in step S151, the monthly scale data set contains a combination of vegetation information, monthly scale precipitation, and terrain information of N×N pixels, while the generated daily scale data set only includes a combination of daily scale precipitation and terrain information of N×N pixels.

[0057] Furthermore, after generating the daily scale dataset and the monthly scale dataset, it also includes: dividing the daily scale dataset and the monthly scale dataset to obtain corresponding training data, verification data, and test data to lay a data foundation for subsequent model training and prediction.

[0058] Step S102: construct a data spatiotemporal fusion and downscaling model based on deep learning, and use the daily scale data set and the monthly scale data set to train the data spatiotemporal fusion and downscaling model separately to obtain the trained models, and use the trained models to make predictions separately to obtain the daily scale precipitation estimate and the monthly scale precipitation estimate respectively.

[0059] Among them, the spatial resolution of daily-scale precipitation estimation and monthly-scale precipitation estimation is the target spatial resolution, and the target spatial resolution is higher than the original spatial resolution of remote sensing precipitation products.

[0060] The data spatiotemporal fusion and downscaling model includes: a residual network ResNet module, a bidirectional long short-term memory network BiLSTM module and an attention mechanism module; the ResNet module is used to learn the spatial neighborhood features of the input data at each time step and input it into the BiLSTM module (also known as the BiLSTM unit); the BiLSTM module is used to capture the temporal contextual relationship of the spatial neighborhood features, and the attention mechanism module is used to weight the output of the BiLSTM module to generate the final model output.

[0061] Specifically, the ResNet module includes multiple residual blocks, which are used to continuously compress the input data of each time step in the spatial dimension and learn its spatial neighborhood features; the BiLSTM module includes multiple BiLSTM layers connected in sequence, which are used to receive the spatial neighborhood features learned by the ResNet module and process them in sequence to obtain the output, hidden layer state and cell state of the corresponding layer; the attention mechanism module is used to multiply the output of the final layer of the BiLSTM module with the final hidden layer state and normalize it to obtain the attention weight, and apply the attention weight to the output of the final layer to generate the final model output.

[0062] Furthermore, the data spatiotemporal fusion and downscaling model also includes: a fully connected layer and a Dropout layer sequentially connected to the attention mechanism module.

[0063] In this embodiment, in terms of model design, in order to capture the spatial and temporal correlation between meteorological observation stations and satellite estimates, the RNBLA model uses the ResNet module to learn spatial neighborhood features and the BiLSTM module to capture temporal contextual relationships. In addition, in order to selectively emphasize the most relevant and informative parts of the input sequence to achieve better prediction results, an attention mechanism module is introduced after the BiLSTM module to achieve weighting.

[0064] Figure 4 Figure 1 is a schematic diagram of the overall structure of the data spatiotemporal fusion and downscaling model. Figure 4 As shown in the figure, first, for the target time step T, data for a total of n time steps before and after it are collected. At each time step from T1 to Tn, the remote sensing precipitation products and environmental variables (topography and vegetation) in the model input area are spliced ​​into a three-dimensional array. Secondly, these three-dimensional arrays are respectively subjected to one convolution and maximum pooling, and then continuously compressed in the spatial dimension through five residual blocks, while being enhanced at the feature level. Next, the obtained n feature vectors form a time series and are sequentially input into a BiLSTM module containing three BiLSTM layers to generate three outputs: BiLSTM output, hidden layer state, and cell state. Then, the attention weight (Attention weight) is obtained by multiplying the BiLSTM output with the final hidden layer state and normalizing it, and it is applied to the BiLSTM output to generate the final output (that is, the Attention weight is normalized and weighted summed to obtain the Attention output). Finally, several fully connected layers are used to capture the relationship and pattern between the outputs of the attention module and remap them to the required output length. In other words, in order to prevent the model from overfitting, fully connected layers and Dropout layers are used after the attention mechanism module.

[0065] Figure 5 This is a schematic diagram of the specific structure of the data spatiotemporal fusion and downscaling model. Figure 5 , the model combines convolutional neural network (CNN), residual block, bidirectional LSTM (BiLSTM), attention mechanism (Attention) and fully connected layer (Linear), and is a hybrid model. The complete process executed at each time step Ti from T1 to Tn and the related hyperparameters are as follows: The input part of the model has four different data sources, namely remote sensing precipitation products and environmental variables (NDVI, DEM), and location information. The data dimension of each is 49×49, but with different numbers of channels. The shape of the remote sensing precipitation product is (49,49,3), NDVI is a single channel, DEM has three channels, and the location information is two channels; then the Concatenate operation is used to splice these data in the channel dimension to obtain the spliced ​​tensor; the next is the convolution layer Convolution2D, which uses a 5×5 convolution kernel and outputs 16 channels, followed by ReLU activation and MaxPooling2D (2×2); then it is input to multiple residual blocks for processing, and the number of output channels of multiple residual blocks is 32, 64, 128, 256, and 512 respectively. The structure of each residual block includes two 3×3 convolution layers, and the output channel of each convolution layer is h, corresponding to different numbers of channels, each convolutional layer is followed by BatchNorm and ReLU, and then the input and output are added (i.e. residual connection); multiple residual blocks are followed by Flatten layers, and the feature map output by the last residual block is flattened through the Flatten operation, and the flattened feature map of each time step is obtained. The feature time series composed of the feature maps of all time steps enters the BiLSTM layer in turn. There are three BiLSTM layers, which have 64, 256, and 512 hidden units respectively; followed by the Attention layer, which performs an attention mechanism on the output of the BiLSTM to refine the weighting of the BiLSTM output, specifically by multiplying the BiLSTM output with the final hidden layer state and normalizing it to obtain the attention weight; then multiple fully connected layers and Dropout layers are connected, and the fully connected layer includes 256, 128, 64, 32, and 1 neurons. There is ReLU activation between the fully connected layer and the Dropout layer, and the Dropout layer uses a ratio of 0.1; finally, a value is output, using the ReLU activation function, which is used for regression tasks, i.e., predicting precipitation.

[0066] After the RNBLA model is constructed, step S102 includes: Step S112: input the training data into the data spatiotemporal fusion and downscaling model, and repeatedly adjust the hyperparameters according to the accuracy evaluation index value of the data spatiotemporal fusion and downscaling model on the verification data until the accuracy index value reaches the preset accuracy condition, and output the trained data spatiotemporal fusion and downscaling model.

[0067] Step S122: Use the trained data spatiotemporal fusion and downscaling model to predict the test data, and measure the unbiased accuracy of the trained data spatiotemporal fusion and downscaling model according to the accuracy evaluation index of the prediction result.

[0068] Based on the above description, we know that the time scale of NDVI data is monthly, and it can only be directly applied to the fusion and downscaling of monthly precipitation. Therefore, the RNBLA model is used to fuse and downscale the daily and monthly precipitation data, respectively. When processing the daily dataset, NDVI data is not input, and the daily and monthly precipitation estimates with the target resolution (such as 0.01°) are obtained.

[0069] Step S103: On a monthly scale, the daily scale precipitation estimate is corrected using the monthly scale precipitation estimate to obtain high-resolution daily precipitation data.

[0070] In some embodiments, step S103, on a monthly scale, using the monthly scale precipitation estimate to correct the daily scale precipitation estimate to obtain high-resolution daily precipitation data, specifically includes the following steps: Step S113: for any specified month, normalize the time series data consisting of the daily precipitation estimation value of the specified month to obtain a normalized processing result.

[0071] Step S123: multiply the normalized processing result by the estimated monthly precipitation value corresponding to the specified month to obtain high-resolution daily precipitation data.

[0072] Specifically, in step S103, the daily precipitation estimate is bias-corrected on a monthly scale, and the correction process includes: normalizing the daily precipitation estimate sequence of a given month and multiplying it by the monthly precipitation estimate of that month.

[0073] It should be noted that traditional normalization processing is usually used to eliminate dimension or data differences, while the normalization processing in this application is to convert the daily precipitation estimate into the proportion of daily precipitation to the total monthly precipitation, which is used to reflect the relative contribution of daily precipitation in the entire specified month. The normalization result is then multiplied by the monthly precipitation estimate corresponding to the specified month. The high-precision monthly precipitation estimate is used as a benchmark to correct the daily precipitation. The stability of the monthly precipitation is utilized, and the temporal dynamic characteristics of the daily precipitation are combined to avoid the problem of precipitation estimation deviation caused by extreme precipitation errors, thereby improving the consistency of the monthly and daily precipitation.

[0074] Through the above steps, the RNBLA model successfully integrated three mainstream remote sensing precipitation products (GSMaP, IMERG and CMORPH) and station observation data to generate high-resolution (spatial resolution of 0.01°) or higher resolution daily precipitation distribution data in the study area (such as China). Compared with the original spatial resolution of remote sensing precipitation products (such as 0.1° for GSMaP, 0.1° for IMERG and 8 km for CMORPH), the daily precipitation estimates calculated by the RNBLA model have obtained richer spatial precipitation details than the original products, which is more suitable for regional detailed research.

[0075] The unbiased accuracy of the trained data spatiotemporal fusion and downscaling model is measured according to the accuracy evaluation index of the prediction results. This example uses CC, MAE, RMSE, Bias, POD, FAR, and CSI as accuracy evaluation indicators. Through verification and comparison with ground observation sites, compared with GSMaP, IMERG and CMORPH products, the CC of daily precipitation data obtained by the method provided in this embodiment is improved by 9.21%, MAE is reduced by 16.02%, RMSE is reduced by 18.14%, and the absolute value of Bias is reduced by 69.57%. The detailed results are shown in Table 1. Table 1 is as follows: Table 1 Overall accuracy comparison between the RNBLA model and different precipitation products

[0076] The results show that the method proposed in this embodiment successfully improves the estimation accuracy of daily precipitation data and enhances the simulation accuracy of daily precipitation fields. In addition, the comprehensive comparison of the three indicators of POD, FAR and CSI shows that this method can improve the forecast accuracy of precipitation data. The experiment also plotted the spatial distribution map of each evaluation index. The analysis shows that the precipitation estimation of the RNBLA model generally obtains lower errors in space. The MAE at the station location is reduced by 16%~32% compared with the original precipitation product, and the RMSE is reduced by 18%~35%. The Bias of the RNBLA model is closer to 0 in 78.33% of the stations, which alleviates the overestimation or underestimation bias of remote sensing precipitation products to a certain extent.

[0077] In order to further understand the above process, Figure 2 The overall logic of this embodiment is reorganized. Figure 2 The following is a logic diagram of a method for downscaling and fusion collaborative calculation of daily precipitation products based on deep learning according to some embodiments of the present application. Figure 2 As shown, the method provided in this embodiment can be performed according to the following steps: (a) Dataset preparation, including remote sensing precipitation products, DEM, NDVI, location and meteorological observation station data. Multi-source satellite precipitation data from different sources provide precipitation information of different precisions; spatial coordinate features are provided through location information (latitude and longitude), vegetation information is provided through NDVI, and digital elevation model DEM provides terrain features such as elevation, slope, and slope aspect. In short, by integrating multi-source remote sensing data with ground observation data and integrating spatiotemporal heterogeneity characteristics, the model's modeling ability for complex environments (such as terrain and vegetation) is improved.

[0078] (b) Data preprocessing, including interpolation of remote sensing precipitation products, accumulation of precipitation products from half-hour / hour time scale to daily and monthly scale, resampling of NDVI and DEM data, and extraction of subgrids that meet the model input requirements. Through data preprocessing, we can fill in the data gaps in space or time, unify the spatial and temporal resolutions of different data sources, and enhance the model's ability to capture spatial context information through local subgrid extraction.

[0079] (c) Model training and hyperparameter adjustment, including repeatedly adjusting hyperparameters according to the performance of the model on the validation data until the accuracy is optimal and a trained model is obtained. The RNBLA model adopts a hybrid architecture of ResNet + BiLSTM + Attention, combining the local feature extraction capability of CNN with the temporal modeling capability of BiLSTM to adapt to the spatiotemporal dynamics of precipitation data and enhance the sensitivity of the model to key features through the attention mechanism.

[0080] (d) Precipitation assessment and prediction: The proposed RNBLA model is used to fuse and downscale remote sensing precipitation products, and the unbiased accuracy of the model is measured using evaluation indicators on test data.

[0081] (e) Daily precipitation correction, which uses the monthly precipitation estimate to correct the daily precipitation estimate on a monthly scale.

[0082] In summary, although a large number of studies have confirmed that there is a significant correlation between precipitation and vegetation, and the use of known vegetation information for precipitation downscaling can improve accuracy. However, the inventor believes that the correlation is only significant on the annual and monthly scales, and is greatly weakened or even disappears on the daily scale. At present, the vegetation NDVI index is only reliable on the monthly scale. These two reasons make it difficult for traditional daily precipitation methods to directly use vegetation information. The method provided in this embodiment is to downscale at both the monthly and daily scales, wherein vegetation information data is added to the monthly scale to calculate the accurate monthly precipitation, while the daily scale downscaling does not consider vegetation information, and is used to estimate the proportion of daily precipitation to the total monthly precipitation. Finally, the monthly scale precipitation is multiplied by the daily precipitation ratio to obtain the final daily precipitation. The advantages of this downscaling framework are: 1) It cleverly uses vegetation information at the monthly scale and integrates it into the final daily precipitation estimation during the correction stage; 2) The original remote sensing precipitation product is obviously deficient in extreme precipitation, and often has large errors or even omissions. This error will completely change the proportion of daily precipitation in the total monthly precipitation. In this embodiment, a model estimation of daily precipitation is first performed separately to improve the accuracy of the proportion of daily precipitation to the total monthly precipitation; 3) Monthly scale data is relatively stable and less affected by accidental factors, while daily precipitation data is greatly affected by random factors and local effects. By correcting the data at the monthly scale, these random fluctuations can be smoothed and the random errors of the data can be reduced. Therefore, the daily precipitation downscaling results calculated by this framework are better than the direct daily precipitation results, and also better than the results of temporal downscaling using the original values ​​of precipitation products after monthly precipitation.

[0083] Based on the same inventive concept, this embodiment provides a daily precipitation product downscaling and fusion collaborative computing system based on deep learning, which includes: A data acquisition unit configured to generate a daily scale dataset and a monthly scale dataset based on precipitation products from multiple sources, meteorological observation station data, vegetation and terrain data; A model building and prediction unit is configured to process the daily scale data set and the monthly scale data set respectively using a pre-built data spatiotemporal fusion and downscaling model, and obtain a daily scale precipitation estimate and a monthly scale precipitation estimate respectively; Among them, the spatial resolution of daily-scale precipitation estimation and monthly-scale precipitation estimation is the target spatial resolution, and the target spatial resolution is higher than the original spatial resolution of precipitation products; The data spatiotemporal fusion and downscaling model includes: a residual network ResNet module, a bidirectional long short-term memory network BiLSTM module, and an attention mechanism module; the ResNet module is used to learn the spatial neighborhood features of the input data at each time step and input it into the BiLSTM module; the BiLSTM module is used to capture the temporal contextual relationship of the spatial neighborhood features, and the attention mechanism module is used to weight the output of the BiLSTM module to generate daily-scale precipitation estimates and monthly-scale precipitation estimates; The correction unit is configured to correct the daily-scale precipitation estimate on a monthly basis using the monthly-scale precipitation estimate to obtain high-resolution daily precipitation data.

[0084] The daily precipitation product downscaling and fusion collaborative computing system based on deep learning provided in this embodiment can implement the process and steps of the daily precipitation product downscaling and fusion collaborative computing method based on deep learning provided in any of the above embodiments, and achieve the same technical effects, which will not be repeated here one by one.

[0085] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for downscaling and fusion collaborative calculation of daily precipitation products based on deep learning, characterized in that: include: Generate a daily-scale dataset and a monthly-scale dataset based on remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources; wherein the monthly-scale dataset contains vegetation information, and the daily-scale dataset does not contain vegetation information; Based on deep learning, a data spatiotemporal fusion and downscaling model is constructed, and the data spatiotemporal fusion and downscaling model is trained separately using the daily scale data set and the monthly scale data set to obtain a trained model, and the trained model is used to perform predictions respectively to obtain a daily scale precipitation estimate and a monthly scale precipitation estimate; The spatial resolution of the daily-scale precipitation estimation and the monthly-scale precipitation estimation is the target resolution, and the target resolution is higher than the original spatial resolution of the remote sensing precipitation product; The data spatiotemporal fusion and downscaling model includes: a residual network ResNet module, a bidirectional long short-term memory network BiLSTM module and an attention mechanism module; the ResNet module is used to learn the spatial neighborhood features of the input data of each time step and input it to the BiLSTM module; the BiLSTM module is used to capture the temporal contextual relationship of the spatial neighborhood features, and the attention mechanism module is used to weight the output of the BiLSTM module to generate a final model output; On a monthly scale, the daily scale precipitation estimate is corrected using the monthly scale precipitation estimate to obtain high-resolution daily precipitation data.

2. The method according to claim 1, characterized in that Based on the remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from various sources, daily and monthly scale datasets are generated, including: Performing spatial interpolation processing on the remote sensing precipitation products from the multiple sources to obtain spatially interpolated precipitation data; Resampling the vegetation and terrain data to obtain resampled vegetation and terrain data; The spatial resolutions of the precipitation data after spatial interpolation and the vegetation and terrain data after resampling are both the target resolution; The interpolated precipitation data are processed by time scale accumulation to obtain daily and monthly precipitation data; Based on the daily precipitation data and the monthly precipitation data, taking each observation station in the meteorological observation station data as the center, extracting the daily precipitation and the monthly precipitation of the surrounding N×N pixels, and extracting the vegetation information and the terrain information of the surrounding N×N pixels from the resampled vegetation and terrain data; Combining the daily-scale precipitation of N×N pixels with the terrain information to generate the daily-scale dataset; combining the monthly-scale precipitation of N×N pixels with the vegetation information and the terrain information to generate the monthly-scale dataset; Wherein, N is a positive integer greater than 1.

3. The method according to claim 2, characterized in that The terrain data is DEM data. Accordingly, taking each observation station in the meteorological observation station data as the center, the terrain information of the surrounding N×N pixels is extracted, specifically: Taking each observation station in the meteorological observation station data as the center, the DEM data is used to extract the elevation, slope and slope direction of N×N pixels around the center point.

4. The method according to claim 2, characterized in that: The vegetation information includes: Normalized Difference Vegetation Index NDVI.

5. The method according to claim 1, characterized in that The ResNet module includes a plurality of residual blocks, and the plurality of residual blocks are used to continuously compress the input data of each time step in the spatial dimension and learn its spatial neighborhood features; The BiLSTM module includes a plurality of BiLSTM layers connected in sequence, which are used to receive the spatial neighborhood features learned by the ResNet module and process them in sequence to obtain the output, hidden layer state and cell state of the corresponding layer; The attention mechanism module is used to multiply the output of the final layer of the BiLSTM module by the final hidden layer state and normalize it to obtain an attention weight, and apply the attention weight to the output of the final layer to generate the final model output.

6. The method according to claim 5, characterized in that The data spatiotemporal fusion and downscaling model also includes: a fully connected layer and a Dropout layer connected to the attention mechanism module in sequence.

7. The method according to claim 1, characterized in that After generating the daily and monthly scale datasets, it also includes: Dividing the daily scale data set and the monthly scale data set to obtain corresponding training data, verification data, and test data; Accordingly, the data spatiotemporal fusion and downscaling model is trained, including: Input the training data into the data spatiotemporal fusion and downscaling model, and repeatedly adjust the hyperparameters according to the accuracy evaluation index value of the data spatiotemporal fusion and downscaling model on the verification data until the accuracy index value reaches the preset accuracy condition, and output the trained data spatiotemporal fusion and downscaling model; The trained data spatiotemporal fusion and downscaling model is used to predict the test data, and the unbiased accuracy of the trained data spatiotemporal fusion and downscaling model is measured according to the accuracy evaluation index of the prediction result.

8. The method according to claim 1, characterized in that On a monthly scale, the daily precipitation estimate is corrected using the monthly precipitation estimate to obtain high-resolution daily precipitation data, specifically: For any specified month, normalizing the time series data consisting of the daily precipitation estimation values ​​of the specified month to obtain a normalized processing result; The normalized result is multiplied by the estimated monthly precipitation value corresponding to the specified month to obtain high-resolution daily precipitation data.

9. A daily precipitation product downscaling and fusion collaborative computing system based on deep learning, characterized in that: include: A data acquisition unit is configured to generate a daily scale data set and a monthly scale data set based on remote sensing precipitation products, meteorological observation station data, vegetation and terrain data from multiple sources; wherein the monthly scale data set contains vegetation information, and the daily scale data set does not contain vegetation information; A model building and prediction unit is configured to process the daily scale data set and the monthly scale data set respectively using a pre-built data spatiotemporal fusion and downscaling model to obtain a daily scale precipitation estimate and a monthly scale precipitation estimate respectively; The spatial resolution of the daily-scale precipitation estimation and the monthly-scale precipitation estimation is the target resolution, and the target resolution is higher than the original spatial resolution of the remote sensing precipitation product. The data spatiotemporal fusion and downscaling model includes: a residual network ResNet module, a bidirectional long short-term memory network BiLSTM module and an attention mechanism module; the ResNet module is used to learn the spatial neighborhood features of the input data of each time step and input it into the BiLSTM module; the BiLSTM module is used to capture the temporal contextual relationship of the spatial neighborhood features, and the attention mechanism module is used to weight the output of the BiLSTM module to generate the daily scale precipitation estimate and the monthly scale precipitation estimate; The correction unit is configured to correct the daily-scale precipitation estimate on a monthly basis using the monthly-scale precipitation estimate to obtain high-resolution daily precipitation data.

10. An electronic device, characterized in that: include: a memory for storing instructions executed by one or more processors of the electronic device; The processor, when executing the instructions in the memory, can enable the electronic device to implement the steps of any one of the methods according to claims 1 to 8.

Images