Meteorological parameter fine scale conversion method based on deep bimodal

Abstract

The invention discloses a meteorological parameter fine scale conversion method based on deep bimodal. The method comprises the following steps: S1, determining a research target variable; s2, related covariables are selected according to three aspects of spatial variation, driving force and influence factors; s3, collecting coarse resolution assimilation data, fine-scale grid and/or survey or measurement data and covariable data; s4, establishing a deep network structure according to the coarse resolution and the fine resolution; s5, determining a loss function and a restrictive condition, and sorting and pairing the data; s6, training a scale conversion model; s7, performing restrictive optimization on the coarse scale background and/or domain knowledge; s8, storing the trained model and parameters and test precision thereof; and S9, carrying out scale conversion application on the model. According to the method, the background with the coarse resolution ratio is combined with the grids with the fine scale or the measurement data, and the scale refining effect of the assimilation data of the meteorological grids is improved through bimodal deep learning modeling.

Description

technical field [0001] The invention relates to a refinement method, in particular to a method for fine-scale conversion of meteorological parameters based on depth dual modes. Background technique [0002] The existing meteorological parameters mainly come from ground observation data or meteorological assimilation data. The former comes from measured data on the ground, but the station data are extremely limited. For example, there are only about 693 monitoring stations for meteorological wind speed observation stations in mainland China; while the latter integrates meteorological observation stations, model predictions and multi-source remote sensing data. Data aggregation has high reliability, but the resolution is relatively coarse, and it is difficult to directly apply it to fine-scale estimation, monitoring and forecasting. [0003] When different raster data have different spatial resolution data, it is necessary to convert the data of different resolutions into a u...

AI Technical Summary

Problems solved by technology

However, the main defect of the assimilated data is that the ground resolution is too coarse to reflect the distribution of meteorological parameters at a fine scale

Although we have some monitoring points of surface meteorological parameters, due to the limited sample points and the weak generalization function of traditional modeling techniques, the real distribution of the surface cannot be well derived

[0008] The existing coarse-resolution raster data fine-scale conversion methods mainly include the use of regression and interpolation methods: 1) The former directly extracts sample points from fine-scale covariates to model regression, less consideration is given to the distribution of measured sample points, Or the restrictive conditions of coarse-scale products or other domain knowledge may lead to deviations in estimation results. Although some methods also consider the background distribution of coarse-scale data, the generalization ability of the modeling method is limited; 2) The interpolation method is According to the monitoring sample points, the method of spatial regression such as kriging is used for interpolation. Generally, the second-order stationary condition required by the spatial variation extracted by interpolation is difficult to meet in actual conditions. The interpolation space is too smooth to reflect the actual situation. The background data Distribution constraints are also lacking in consideration

[0009] In summary, the main disadvantages of existing technologies include the inapplicability of super-resolution techniques, the lack of fine-scale measured samples leads to the inapplicability of supervised convolutional network methods, and the lack of background parameters or limitations of domain knowledge in the results lead to poor scaling results. Deviation etc.

Images