From 1940 to 1953, red pine and weed-tree forests were converted to bamboo for the production of edible bamboo shoots and mature culms for craft use. However, bamboo shoots are now imported, and mature culms are generally no longer used for crafts. Bamboo forests have not been maintained and have expanded into semi-natural areas (satoyama) near populated zones. Thus, it is important to determine their distribution.
We mapped the bamboo distribution using seasonal ALOS/AVNIR-2 data for Nara and southern Kyoto Prefectures. To study seasonal variation in spectral reflectance, bamboo leaves were measured monthly using a spectral radiometer. Based on these data, we investigated the seasonal changes in the bamboo leaf reflectance factor corresponding to the wavelengths of the AVNIR-2 sensor, in three coefficients, and in a modified vegetation index (MVIUPD) based on the Universal Pattern Decomposition Method (UPDM). Our results showed that the MVIUPD was lowest in May due to seasonal variation in the bamboo leaf characteristics. We used spring (May) data for mapping bamboo forests using AVNIR-2 data, and winter (Jan) data for classifying deciduous vegetation. We displayed training data for bamboo forests in a scatter plot between Cv, the UPDM vegetation coefficient, and MVIUPD, resulting in a cluster with a gentle curve. We fit the relationship between Cv and MVIUPD using a natural logarithm function and labeled bamboo pixels according to this relationship. The results were verified using field survey data, and the Kappa coefficient was 0.75. Narrow or sparse bamboo forest distributions caused misidentification. The mapping results were compared with the forest stand database of Nara Prefecture and the vegetation survey dataset provided by the Ministry of the Environment of Japan. We investigated land cover in the main areas where the results differed from these datasets. About 20% of the areas were misclassified, and they included an evergreen broadleaf growing on an ancient tomb, a mixed deciduous broadleaf and red pine forest, and a red pine forest. The results show that bamboo had not been detected during the 6th vegetation survey (2001) in northern Nara Prefecture.
Thus, we conclude that AVNIR-2 data are useful for mapping bamboo forests on large spatial scales.
