To estimate the amount of carbon stored in forests is a huge task to make the global carbon cycle clear. To understanding the carbon cycle in forests is a challenge for helping to solve lots of environmental issues such as the global warming. Lidar (light detection and ranging) remote sensing can be a technology for estimation of biomass. Thus the instrument started to be used for it in forests lately. In previous studies, the reliability of Lidar remote sensing had been discussed a lot and proved the performance today. However, there are some parts which need to be fixed to improve the accuracy remaining; it still has lots of errors under the conditions like complex geographical features, irregular stand structure, and mixed species composition. In this study, we retest the performance comparing the relationship between observed canopy structures (data collected by hand) and predicted above ground biomass (data collected by Lidar) in natural mixed forests of boreal conifers and broad-leaved tree species in Hokkaido, Japan. Then we use different spatial resolutions following topography (ridge, slope, stream) and vegetation type (conifer, broadleaved) to derive a suitable resolution for each factor. We propose improvements of the technique to increase the accuracy of the estimation.