5 results about "Rainfall estimation" patented technology

The invention discloses a radar quantitative rainfall estimation method based on a space-time attention model, and the method comprises the steps: carrying out the position coding, injecting position information into the embedded representation of sequence elements, explicitly representing the specific coordinates of the sequence elements in a sequence, dividing the codes into two types: fixed coding and learnable coding, and after the position coding is completed, carrying out the estimation of the radar quantitative rainfall. The method comprises the following steps of: calculating a mutual relationship between internal elements of an input sequence to realize a model architecture of information aggregation, establishing endogenous association between the elements of the sequence to realize feature interaction, synchronously calculating and integrating all position information of the sequence, and capturing space-time dependency in meteorological data through a position relationship between independent modeling time and space after calculation and integration; according to the invention, a position characterization mechanism based on three-dimensional space-time relative position coding is introduced, and space-time key features of radar echo data are effectively extracted through multiple attention modules; and designing a space-time position coding strategy capable of self-adaptive learning, and realizing joint feature representation of space-time dimensions.

The application relates to the technical field of radar data processing, and specifically discloses a quantitative rainfall estimation method based on a dual-polarization multi-radar composite technology, which comprises the following steps: obtaining standardized data by preprocessing dual-polarization radar original observation data; calculating and fusing a multi-dimensional quality index: calculating a static environment index and a dynamic physical consistency index, and fusing to obtain a comprehensive quality index; adaptively selecting an optimal rainfall estimation algorithm according to the quality of radar dual-polarization observation data to generate an initial fused rainfall field; dividing multiple quality levels, independently calculating an average field bias correction factor for each quality level, and applying the average field bias correction factor to radar pixels of the corresponding quality level to obtain a final rainfall inversion result. The application realizes source identification and filtering of non-meteorological echoes and abnormal data, avoids the propagation of false information and the amplification of system errors, and improves the reliability and robustness of radar network composition.

The invention relates to the technical field of radar rainfall estimation, and discloses a dual-polarization radar quantitative rainfall estimation method based on deep learning, and the method comprises the steps: collecting dual-polarization radar body scanning basis data, and carrying out the quality control, and obtaining a standardized polarization parameter; constructing and training a hybrid network model based on 3D-CNN + BiLSTM + Attention, and inputting a three-dimensional tensor constructed by a standardized polarization parameter into the trained hybrid network model to obtain an hourly rainfall intensity preliminary estimated value QPE of each radar grid point; positioning target radar grid points, extracting effective data pairs, and calculating a unique correction factor of each radar grid point; and calculating a single scanning height precision estimated value by combining the hourly rainfall intensity initial estimated value QPE of each radar grid point, and performing accumulation according to a preset time length to obtain an accumulated quantitative rainfall estimation product. According to the method, the dual-polarization radar parameters and the deep learning model are fused, correction is carried out in combination with the real-time rain gauge, accurate multi-period regional rainfall products are generated, and reliable support is provided for meteorological monitoring, disaster early warning and the like.

The application discloses a kind of quantitative precipitation estimation methods based on Hurdle-IMDL framework, comprising: obtaining historical precipitation measurement data and meteorological satellite observation data;Based on Hurdle model, construct biased quantitative precipitation estimation probability model, revise the biased rainfall estimation probability submodel according to the IMDL method, into experience distribution to construct experience quantitative precipitation estimation probability model and derive its negative log-likelihood function;Build AI model, use negative log-likelihood function as loss function to optimize AI model;Meteorological satellite observation data of the region to be inverted is input into the AI model trained, to obtain the estimated value of the parameter of experience quantitative precipitation estimation probability model, then estimate precipitation according to conditional expectation.The application uses Hurdle model to solve zero inflation problem, and uses IMDL learning method to deal with long tail problem, to improve the inversion accuracy of strong to extreme precipitation.

The invention belongs to the technical field of rainfall estimation, and discloses a rainfall estimation method and device based on radar data, and the method comprises the steps: obtaining minute-level accumulated rainfall information of a ground rainfall observation system, and obtaining a rainfall data source; acquiring radar product data of the weather radar to obtain a radar data source; processing the radar data source according to a preset volume scanning period and a preset format to obtain three-dimensional radar lattice point data; processing the rainfall data source based on a preset body scanning period and an effective detection distance of a weather radar to obtain rainfall label data; obtaining radar precipitation grid point field data according to the three-dimensional radar grid point data and the precipitation label data; training a deep residual network model according to the radar precipitation grid point field data to obtain a precipitation estimation model; and obtaining target three-dimensional radar grid point data and inputting the data into the rainfall estimation model to obtain a rainfall field estimation result. The rainfall can be accurately estimated based on the radar data.