[H-TT15] Environmental Remote Sensing

[HTT15-P02] Land-use/land-cover change (LULCC) in the Peruvian Andes-Amazon since 2000 based on spatial-temporal fused composites of multiple satellite data

*Ziping Yang¹ (1.Graduate School of Science and Engineering, Chiba University)

キーワード：Land cover, Spatial-temporal fusion, Data reconstruction, Change detection, Forest change

Ecosystems in the Peruvian Andes-Amazon are among the most vulnerable to the effects of climate change and human activities such as agriculture, deforestation and fires, although they are preserved by large differences in elevation and a variety of environmental conditions. Identification of land-use/land-cover change (LULCC) by remote sensing can monitor ecosystem services and endemic biodiversity, which is essential. However, the high precipitation in the Amazon rainforest area and the low temporal resolution of the satellite will affect the availability of remote sensing imagery. Therefore, mapping the distribution of land-cover types in the region is full of challenges. Here, we performed a high spatial resolution land-cover change assessment for Peruvian Andes-Amazon with spatial-temporal fused composites of multiple satellite data. Our goals were 1) to examine the effect of spatial-temporal fused composites on land-cover classification for mountainous and rainforest regions, and 2) to assess land-cover changes since 2000 across the Peruvian Andes-Amazon based on Landsat data. First, the method of spatial-temporal fusion was performed to reconstruct high temporal resolution, high availability composite using MODIS, Sentinel and Landsat data. We compared supervised classification using a single type of imagery with supervised classification using composite of different imagery, and assessed the accuracy of both. Second, we believed that different features should be used for different land-cover types in supervised classification. Therefore, we carried out vegetation classification before supervised classification. For the vegetation type, NDVI time series is used for further classification; and for the non-vegetation, spectral data combined with nighttime light, altitude and other data are used. We also compared the accuracy of land-cover types with and without vegetation classification. Third, we derived land-cover classification every year from 2000 to 2020, based on class supervised classification and SAM. In regard to our first goal, spatial-temporal fused composites improved the overall accuracy of the classification by up to 10%. In regard to our second goal, we found that forest loss was the most prevalent change in Peruvian Andes-Amazon. Forest loss has a lot to do with illegal logging. Under the combined influence of the Peruvian cold current and the Andes Mountains, there is almost no precipitation in the western part of Peru, which makes both the population and the arable land continue to move eastward. Natural forestry resources, hidden mineral resources, and fertile land in the Amazon rainforest attract people to deforestation. Our results underscore the importance and feasibility of spatial-temporal fusion and pre-classification for accurate land-cover classification in alpine and rainforest regions.