5:15 PM - 7:15 PM
[ACC32-P02] Study on automatic extraction method of Aufeis using optical satellite images
Keywords:Aufeis, Remote sensing, Automatic extraction, Ulaanbaatar
1. Background and Objectives
Aufeis (icing) is a layered ice mass that forms in winter when river water, spring water, or groundwater seeping to the surface freezes. Around Ulaanbaatar, icing has been reported to occur in low-lying areas along rivers. In recent years, population concentration in this region has led to the formation of low-income residential areas in icing-prone zones, causing issues such as winter icing, spring floods, and deterioration of sanitary conditions due to wastewater mixing with icing. The distribution and scale of icing vary due to climate change and human activities. Accurate automatic extraction using remote sensing data is necessary to monitor these long-term and widespread changes. This study examines a method for the automatic extraction of icing using remote sensing data.
2. Methodology
Analysis was conducted for urban and mountainous areas of Ulaanbaatar. Sentinel-2 satellite data observed in March 2023 and the ALOS Global Digital Surface Model (AW3D30) were used for the analysis. First, a FalseColor image was created, and clustering was performed using the k-means method with the Python scikit-learn library to extract ice regions. Next, the Normalized Difference Snow Index (NDSI) was calculated using ArcGIS Pro 3.4.0, and regions with NDSI values of 0.4 or higher were extracted based on established threshold criteria. Furthermore, using AW3D30 data, a terrain index was defined, identifying valley-bottom areas where the slope was less than 5°, and the elevation difference within a 300 m radius was less than 10 m from the lowest elevation. Areas that met all these conditions were identified as icing occurrence zones.
3. Results and Discussion
In the mountainous areas of Ulaanbaatar, distinguishing between snow and icing was difficult using NDSI alone, but the use of FalseColor images allowed for the exclusion of most snow-covered areas. Additionally, applying the terrain index helped extract icing in valley bottoms.
In urban areas of Ulaanbaatar, k-means clustering failed to accurately classify ice, making FalseColor images insufficient for precise extraction. This is because urban areas contain many buildings, roads, and paved surfaces with similar reflectance characteristics to ice, leading to misclassification. However, since image data with minimal snowfall was used for the analysis, extraction using only NDSI and the terrain index produced results consistent with observed icing distributions. In the presence of snow in urban areas, NDSI struggles to differentiate between snow and ice, potentially overestimating icing. In the future, identifying factors that hinder icing extraction in urban environments and developing algorithms that account for these factors will be necessary to improve detection accuracy.
Aufeis (icing) is a layered ice mass that forms in winter when river water, spring water, or groundwater seeping to the surface freezes. Around Ulaanbaatar, icing has been reported to occur in low-lying areas along rivers. In recent years, population concentration in this region has led to the formation of low-income residential areas in icing-prone zones, causing issues such as winter icing, spring floods, and deterioration of sanitary conditions due to wastewater mixing with icing. The distribution and scale of icing vary due to climate change and human activities. Accurate automatic extraction using remote sensing data is necessary to monitor these long-term and widespread changes. This study examines a method for the automatic extraction of icing using remote sensing data.
2. Methodology
Analysis was conducted for urban and mountainous areas of Ulaanbaatar. Sentinel-2 satellite data observed in March 2023 and the ALOS Global Digital Surface Model (AW3D30) were used for the analysis. First, a FalseColor image was created, and clustering was performed using the k-means method with the Python scikit-learn library to extract ice regions. Next, the Normalized Difference Snow Index (NDSI) was calculated using ArcGIS Pro 3.4.0, and regions with NDSI values of 0.4 or higher were extracted based on established threshold criteria. Furthermore, using AW3D30 data, a terrain index was defined, identifying valley-bottom areas where the slope was less than 5°, and the elevation difference within a 300 m radius was less than 10 m from the lowest elevation. Areas that met all these conditions were identified as icing occurrence zones.
3. Results and Discussion
In the mountainous areas of Ulaanbaatar, distinguishing between snow and icing was difficult using NDSI alone, but the use of FalseColor images allowed for the exclusion of most snow-covered areas. Additionally, applying the terrain index helped extract icing in valley bottoms.
In urban areas of Ulaanbaatar, k-means clustering failed to accurately classify ice, making FalseColor images insufficient for precise extraction. This is because urban areas contain many buildings, roads, and paved surfaces with similar reflectance characteristics to ice, leading to misclassification. However, since image data with minimal snowfall was used for the analysis, extraction using only NDSI and the terrain index produced results consistent with observed icing distributions. In the presence of snow in urban areas, NDSI struggles to differentiate between snow and ice, potentially overestimating icing. In the future, identifying factors that hinder icing extraction in urban environments and developing algorithms that account for these factors will be necessary to improve detection accuracy.