The concept of reanalysis can be tracked back to synoptic analysis in the era of manual weather forecasting, in the era of numerical weather forecasting the data assimilation methods were used to provide initial conditions for numerical weather prediction models. However, due to the rapid updating of assimilation methods and numerical weather prediction models, the application of these data in climate research is limited. Therefore, the concept of reanalysis was proposed, whose core is to use fixed assimilation methods and fixed numerical models to reanalyze historical data. It interpolates the observed data based on physical model, and can provide estimates of unobserved variables, which solves the problem of sparse and irregularly distributed observation stations. In order to improve accuracy, the major global atmospheric reanalyses assimilate as much observational data as possible, that is, all input atmospheric reanalyses. Although the numerical models and assimilation methods are fixed, the observation system has been evolving, with major changes including modern sounding observations since the 1950s and satellite observations since 1979, as well as the upgrading of ground-based and satellite observation instruments. All of these have introduced inhomogeneity into global atmospheric reanalysis products. Therefore, there has been ongoing controversy over whether global atmospheric reanalysis can be applied to climate change research, which has also given rise to the sparse input reanalysis, such as 20th century reanalysis. In order to reduce the impact of observation system evolution, this type of reanalysis only assimilated a small amount of ground observations such as atmospheric pressure, which extended reanalysis to the mid-19th century.For global atmospheric reanalysis, atmospheric sounding observations are much more important than surface observations. Early data assimilation was mainly based on the simplest optimal interpolation method, assimilating the satellite retrievals of atmospheric temperature and humidity profiles. However, the vertical resolution of satellite observations was too low, and the effects of assimilation were low or negative. This is because optimal interpolation is based on linear theory and cannot handle and analyze observations related to variables in a nonlinear manner. The variational method allows for direct assimilation of satellite radiance observations, significantly improving the value of satellite observations. But the types of observations that can be assimilated are limited to variables that can be accurately simulated. In the late 1990s, a rapid radiative transfer model was developed to accurately simulate microwave brightness temperature observations under cloudy conditions, achieving full sky assimilation of microwave radiation. But only recently attempts have been made to assimilate key but discontinuous observation parameters, and only one global reanalysis has achieved assimilation of satellite aerosol optical depth. But so far, there is no global atmospheric reanalysis that can truly assimilate precipitation observations.Global atmospheric reanalysis surface analysis products may be the most widely used data. In order to improve the accuracy and spatiotemporal resolution of land surface analysis, many global atmospheric reanalysis systems also perform offline land surface analysis. In general, these land surface reanalysis use multi-source merged precipitation data to correct the precipitation products of the atmospheric reanalysis system, because the precipitation simulated by atmospheric models still has significant errors. But ERA5 Land did not perform such correction, likely because such correction affects the real-time performance of the product, as merged precipitation products are generally delayed by more than a month.However, the existing global atmospheric reanalysis systems exhibit significant differences in the surface analysis. The European Centre for Medium Range Weather Forecasts and Japanese Meteorological Agency reanalyzed and assimilated the observed surface air temperature and humidity observations collected at weather stations over global land, with the latter further assimilating land surface wind speed observations. The reanalysis series done by the United States agencies (i. e., the National Oceanic and Atmospheric Administration (NOAA) and the National Aeronautics and Space Administration (NASA)) did not assimilate surface temperature, humidity, and wind speed observations collected by weather stations over global land. This results in lower consistency between reanalysis produced by the United States agencies and ground-based observations, while reanalysis produced by Europe and Japan agencies has higher consistency, making their applications more widespread. But until now all global atmospheric reanalysis have not included interannual variations in land cover, land use, and vegetation growth yet, which limits its accuracy in estimating surface variables. This article reviews the historical evolution of global atmospheric reanalysis done by Europe, the United States, Japan, and China, as well as the applicability of their surface analysis products in China.