5:15 PM - 7:15 PM
[HGM04-P06] Estimating deformation rates of deep-seated landslides using NDVI ratios derived from Sentinel-2 imagery: a case study of Hokkaido larch forest
Keywords:Deep-seated landslide, NDVI ratio, Sentinel-2, Japanese larch, UAV, LiDAR
Deep-seated landslides commonly occur in areas with weak geological formations. These landslides typically deform slowly, enabling forest growth on their surfaces, but can accelerate rapidly due to triggers, posing risks to surrounding areas. Since many exhibit continuous deformation before acceleration, early detection through accessible methods could aid in disaster prevention. This study explores whether landslide deformation rates can be estimated using NDVI derived from open-source satellite imagery.
Trees on active landslides tend to exhibit increased tilting over time, potentially reflecting the deformation history of the land mass. More tilted trees may show lower NDVI values in their crowns. If these values correlate with Sentinel-2 satellite imagery, satellite data could be used to estimate landslide deformation history. When combined with detailed topographic records from airborne LiDAR survey, this data could enable calculation of landslide deformation rates over time. While previous research has shown that active landslides generally display lower NDVI values compared to surrounding areas, the relationships between NDVI, deformation history, and rates remain understudied.
This study focuses on the Otarimappu landslide in Biratori, Hokkaido, a weathered rock landslide with a slip surface 8 meters below ground level. The 3-hectare site consists of 7 blocks with varying deformation rates. Airborne LiDAR surveys were conducted by manned aircraft in 2010 and 2014 (Hokkaido Development Bureau), followed by UAV surveys from 2021. The average annual deformation rate of each block from 2010 to 2024 ranges from 0.14 to 0.60 m/year. The area was planted with Japanese larch in 1956, which now forms the upper canopy layer. Within the Otarimappu landslide, many larches show trunk tilting and fallen trees with no regular pattern, while those on the adjacent inactive landslide grow straight at approximately 5-meter intervals. The study analyzed 77 identifiable larches within the Otarimappu landslide from June 2024 UAV imagery and 20 randomly selected larches from the inactive landslide. June was selected for imagery acquisition to capture the period when new leaves are fully developed. Trunk tilt measurements were taken using ground-based laser scanning from August to September 2024. NDVI values for these tree crowns were calculated using UAV imagery from June of each year from 2021 to 2024. NDVI ratios were obtained for individual trees and block averages relative to the mean NDVI of larches in the inactive landslide. Similarly, NDVI ratios were calculated from Sentinel-2 satellite imagery (10m resolution) from June of each year from 2021 onwards.
Results showed that blocks with greater movement since 2010 had more tilted larches, suggesting tree tilt as an indicator of long-term landslide deformation. A negative correlation was found between tree tilt and NDVI ratios calculated from June 2024 UAV imagery. NDVI ratios from individual larches and corresponding Sentinel-2 imagery from June 2021 to 2024 showed strong correlation (R² = 0.91), demonstrating the potential for using satellite imagery to estimate landslide deformation history. For each block, the moving distance D (m) from 2010 to 2024 could be expressed using the June 2024 satellite-derived NDVI ratio, s, as D = -46.9s + 53.2 (R² = 0.81), and the average annual rate V (m/year), as V = -3.35s + 3.80. These relationships indicate that NDVI from Sentinel-2 imagery can be used to estimate deep-seated landslide deformation rates in Japanese larch plantations.
While promising, further research is needed to strengthen the observed relationships, particularly regarding local tree variations within blocks and their association with fine-scale geomorphic features and seasonal NDVI changes. Future studies will focus on expanding the dataset across multiple sites to better understand the relationships between satellite-derived NDVI, landslide deformation history, and rates.
Trees on active landslides tend to exhibit increased tilting over time, potentially reflecting the deformation history of the land mass. More tilted trees may show lower NDVI values in their crowns. If these values correlate with Sentinel-2 satellite imagery, satellite data could be used to estimate landslide deformation history. When combined with detailed topographic records from airborne LiDAR survey, this data could enable calculation of landslide deformation rates over time. While previous research has shown that active landslides generally display lower NDVI values compared to surrounding areas, the relationships between NDVI, deformation history, and rates remain understudied.
This study focuses on the Otarimappu landslide in Biratori, Hokkaido, a weathered rock landslide with a slip surface 8 meters below ground level. The 3-hectare site consists of 7 blocks with varying deformation rates. Airborne LiDAR surveys were conducted by manned aircraft in 2010 and 2014 (Hokkaido Development Bureau), followed by UAV surveys from 2021. The average annual deformation rate of each block from 2010 to 2024 ranges from 0.14 to 0.60 m/year. The area was planted with Japanese larch in 1956, which now forms the upper canopy layer. Within the Otarimappu landslide, many larches show trunk tilting and fallen trees with no regular pattern, while those on the adjacent inactive landslide grow straight at approximately 5-meter intervals. The study analyzed 77 identifiable larches within the Otarimappu landslide from June 2024 UAV imagery and 20 randomly selected larches from the inactive landslide. June was selected for imagery acquisition to capture the period when new leaves are fully developed. Trunk tilt measurements were taken using ground-based laser scanning from August to September 2024. NDVI values for these tree crowns were calculated using UAV imagery from June of each year from 2021 to 2024. NDVI ratios were obtained for individual trees and block averages relative to the mean NDVI of larches in the inactive landslide. Similarly, NDVI ratios were calculated from Sentinel-2 satellite imagery (10m resolution) from June of each year from 2021 onwards.
Results showed that blocks with greater movement since 2010 had more tilted larches, suggesting tree tilt as an indicator of long-term landslide deformation. A negative correlation was found between tree tilt and NDVI ratios calculated from June 2024 UAV imagery. NDVI ratios from individual larches and corresponding Sentinel-2 imagery from June 2021 to 2024 showed strong correlation (R² = 0.91), demonstrating the potential for using satellite imagery to estimate landslide deformation history. For each block, the moving distance D (m) from 2010 to 2024 could be expressed using the June 2024 satellite-derived NDVI ratio, s, as D = -46.9s + 53.2 (R² = 0.81), and the average annual rate V (m/year), as V = -3.35s + 3.80. These relationships indicate that NDVI from Sentinel-2 imagery can be used to estimate deep-seated landslide deformation rates in Japanese larch plantations.
While promising, further research is needed to strengthen the observed relationships, particularly regarding local tree variations within blocks and their association with fine-scale geomorphic features and seasonal NDVI changes. Future studies will focus on expanding the dataset across multiple sites to better understand the relationships between satellite-derived NDVI, landslide deformation history, and rates.