Satellite and ground rainfall measurement value assimilation method based on neural network

Abstract

The invention provides a satellite and ground rainfall measurement value assimilation method based on a neural network method, and the method comprises the steps: 1), obtaining satellite measurement data and ground rain gauge data which are independent of each other, and carrying out the data format gridding and preprocessing operation; 2) processing and researching long-time series rainfall dataon the basis of a time level; 3) for short-time sequence rainfall research, processing class label-free data by using elevation data and rainfall data and adopting k-means + +; 4) classifying the short-time rainfall data and the long-time sequence rainfall data set, and resampling the same into a data set suitable for machine learning; and 5) adopting an improved LSTM neural network model to performi neural network training by using pure satellite data and station data as labels to obtain a training model, and then substituting pure satellite data to be tested into the training model to obtainan assimilation result. According to the invention, the system error of rainfall satellite data is optimized, and the reliability of the rainfall estimated value is further improved, so that the space-time characteristics have higher consistency.

Description

technical field [0001] The invention relates to rainfall assimilation, in particular to a satellite and ground precipitation measurement value assimilation method based on the LSTM neural network method. Background technique [0002] There are many widely used methods for measuring precipitation. One of the more intuitive ones is the surface rain gauge, which directly measures precipitation data. Its advantage is that for a specific observation point, the rainfall data measured by the rain gauge has high accuracy. But at the same time, this method also has shortcomings that cannot be ignored: because the precipitation process has great inhomogeneity both in time and in space, and the distribution density of ground rain gauges is limited, this method cannot Effectively reflect the precipitation distribution of a large area. For those remote areas, valley basins, and areas with complex terrain, it is even more impractical to directly measure with ground rain gauges. Another ...

AI Technical Summary

Problems solved by technology

But at the same time, this method also has shortcomings that cannot be ignored: because the precipitation process has great inhomogeneity both in time and in space, and the distribution density of ground rain gauges is limited, this method cannot Effectively reflect the precipitation distribution of a large area. For those remote areas, valley basins, and areas with complex terrain, it is even more impractical to measure directly with ground rain gauges

The main disadvantage of this method is that the penetration of visible light and infrared waves to the cloud layer is relatively poor, and the main information obtained during the inversion process comes from the top of the precipitation cloud layer, the error caused by this physical characteristic To a certain extent, the comparability between remote sensing information and ground observation data is reduced

Comparing several commonly used rainfall data measurement techniques at present, it can be found that the rainfall data obtained by surface rain gauges are accurate in a specific measurement space, but limited by its own distribution density, the rain gauges can The obtained rainfall data lacks continuity in space; while the method of measuring by radar and satellite has good continuity in space, but the accuracy of the measured rainfall data needs further consideration

At present, there have been many studies on monitoring rainfall data through remote sensing technology, but there are still various problems such as geographical constraints, time-consuming and labor-intensive, heavy workload, insufficient accuracy, and poor effectiveness.

Images