Abstract: This study used adaptive algebraic reconstruction technique to reconstruct 2-D temperature and H₂O concentration in the multi-peak distribution measurement with different grid divisions based on laser absorption spectroscopy technology. The research found that the reconstruction error at different grid divisions is not related to θ (the ratio of the number of rays to the number of grids), but mainly to the projection angle. When the number of rays is constant at a single Angle, the more projection angles there are, the lower the reconstruction error will be. By adding random errors, it was found that the reconstruction error increases with the increase of random errors. Meanwhile, the reconstruction errors for 10×10 grid were significantly higher than that of the other two groups of grid division situations with the same random errors. When the random error reaches 1.5%, the reconstruction errors of the temperature and H₂O concentration with 10×10 grid can no longer converge. On this basis, by using the neighborhood mean smoothing method to optimize the reconstruction process, the anti-noise performance of the AART during the reconstruction process can be significantly improved. When the random error is 1.5%, the reconstruction errors of the two-dimensional temperature and H₂O concentration obtained by the 10×10 grids are 7.51% and 5.17% respectively. Compared with before the neighborhood mean smoothing method, the reconstruction errors were significantly reduced.