A qualitative accuracy assessment for land cover classification was performed to evaluate (i) the performance when applied to topographically and non-topographically corrected Landsat ETM+ images, and (ii) the relative applicability of multi-temporal Landsat ETM+ images in the mountainous area of northern Japan. Five Landsat ETM+ images from a single year were used to characterize six categories : water, conifer forest, deciduous forest, agriculture, paddy and urban. The removal of topographic effects from Landsat ETM+ image before the classification resulted in only slightly more accurate mapping. Topographic correction was not essential for land cover classification using Landsat ETM+ image, because habitat formation of several species already had strong relations with geographic factors, such as slope, aspect and elevation, in heavy snowfall area of Japan. On the other hand, the use of multi-temporal Landsat ETM+ images significantly increased the classification accuracy. Overall accuracy and kappa coefficient rose from 66.6±2.8% and 0.519±0.044 with a single image up to 73.7% and 0.628 with five images, respectively. The best performance was attained when combining all five images. Multi-temporal analyses enhanced the ability to discriminate categories that are inseparable in a single date image. Moreover, as clouds were rarely at the same place, an overlay procedure that used multi-temporal images was useful for creating a cloud-free composite image and thus monitoring land cover of large areas. However, the classification accuracy did not proportionally increase with increasing the number of images. Classification accuracy demonstrated a tendency to saturate with three images. At least two or three images were acquired to get more accurate land cover map from the perspective of cost optimization. These findings underline how multi-temporal analyses of Landsat ETM+ images can be used as a tool for rapid operation for land cover classification without the need to employ complex and time-consuming topographic correction techniques. This type of research and resultant information would be critical for utilization of remotely sensed data to the fullest extent.
In this report, we acquired hyper spectrum data, LiDAR data and digital photography data by airborne survey to investigate the distribution/the expansion situation of bamboo stands in the Satoyama area of Kakuma campus of Kanazawa University in Kanazawa, Ishikawa Pref. Precision of classification became higher by using the “standardized” spectrum of hyper spectral data. In addition, the precision was improved more by adding the information of height of vegetation by the LiDAR data. We also found that other trees surviving in the bamboo expansion area were taller than the bamboo stands. In conclusion, by the high-resolution investigation by airborne remote sensing, we could investigate distribution of bamboo stand and the expansion situation well.