Research on Extreme Learning Algorithm and Its Application to Atmospheric Nonlinear Systems

Abstract

Abstract. Extreme learning machine (ELM) algorithm, which becomes more and more popular in the area of artificial intelligence for the past few years, is faster than the traditional machine learning algorithms, especially than the single hidden layer feed-forward neural networks (SLFNs). However, ELM is merely commonly used in the field of computer science or other hot areas. This paper investigates the ability of ELM to emulate the atmospheric nonlinear systems. The performance of ELM on emulating the nonlinear chaotic system -Lorenz63 is analyzed. The results show that ELM can accurately and quickly simulate the Lorenz63 forecast field at different forecast length, and thus providing a new idea for solving kinds of atmospheric nonlinear equations.

Numerical experimental research on assimilation of cloud-free radiances from fengyun-3D microwave sounding observations over land surface

“…Firstly, the key scientific problems such as real-time pixel by pixel dynamic retrieval of land surface emissivity [5], and the new cloud detection scheme of dioxygen channels based on artificial intelligence algorithms are solved. The Gaussian process regression GPR method is used to find some intrinsic brightness temperature relationship function between the dioxygen channel at 50-60GHz and 118.75GHz, and a new cloud detection scheme is developed [6], which realizes the cloud detection of FY-3D temperature and humidity assimilated under clear sky conditions. The three-dimensional variational data assimilation system for FY-3D land surface-sensitive microwave vertical sounding data is established, and the direct assimilation of MWTS-2 and MWHS-2 radiance data is realized.…”