The detection of hyper-spectral atmosphere CO2 needs remote sensor to be characterized and calibrated precisely while spectral calibration is the most basic work. The high uncertainty of wavelength calibration coefficient is a big problem as to the traditional laboratory calibration methods. In order to solve this problem, the research of error correction of spectral calibration based on gas absorption principle is carried out. The method is in accordance with the using conditions of instrument and it improves the practicability of the calibration coefficient. First, theoretical spectrum and error components are simulated by using radiative transfer. Then, the experiment of atmosphere CO2 absorption spectrum measurement is performed based on the atmosphere environmental simulation calibration house. Last, spectral error is corrected and optimized with LM algorithm. The result of spectral calibration of error correction shows that the mean value of spectral error deviation decreases from 0.03 cm-1 before correction to 0.008 cm-1 after correction, and systematic and mutable errors are removed. The spectral calibration precision on the ground is improved significantly, which lays the foundation for the subsequent greenhouse gas retrieval.