Faculty Sponsor

Sara Mana, George Waddington

Status

Graduate

Publication Date

5-1-2021

Department

Geography

Description

Many volcanic fields can be found in the East African Rift (EAR), an active divergent plate boundary. Marsabit (2.32°N, 37.97°E) and Nyambeni Hills (0.42°N, 37.96°E) of Kenya are located on the eastern shoulder of the Kenyan Rift, part of the eastern branch of the EAR. Both volcanic fields formed in the late Pleistocene and Holocene, and both are host to hundreds of pyroclastic cones and maar craters. Previous research has established that trends of linear arrays and the morphologies of cones and craters can be used to establish the locations and orientations of shallow igneous intrusive systems, especially in areas where subsurface feeder dikes are not visible at the surface. The orientation of dikes are controlled by the regional tensile stress field or pre-existing lithospheric structures. Therefore, the analysis of dike orientations can supply valuable information regarding regional tectonic stresses and crustal fabric controls. Manually mapping extrusive volcanic features can be time consuming and subjective. Machine learning object detection and classification can speed up the mapping process, allowing for quicker analysis. Here we present our findings in the application of the YOLO.v.2 object detection machine learning algorithm to identify extrusive volcanic features and its analysis of volcanic fields not included in the training sites. Landsat 8 OLI imagery of Marsabit is used to create training and validation data in PASCAL VOC format. False color compositions of Marsabit and Nyambeni Hills are created from a multiplicative merge of red, NIR, and SWIR bands and the panchromatic band with a resolution of approximately 15 meters. After training YOLO V.2 with training data from Marsabit, it is used to analyze validation images of Nyambeni Hills. Preliminary results suggest that the sharp topographic relief of maars are easily identified in both volcanic fields, while cones are more likely to be identified in Marsabit. Due to eruption style and erosion, the cones of Nyambeni Hills tend to be less prominent than cones on Marsabit. However, YOLO V.2 still identified a few large cones on Nyambeni Hills.

Presentation Type

Poster

Included in

Geography Commons

COinS
 

Identifying Extrusive Volcanic Features with YOLOv2.0

Many volcanic fields can be found in the East African Rift (EAR), an active divergent plate boundary. Marsabit (2.32°N, 37.97°E) and Nyambeni Hills (0.42°N, 37.96°E) of Kenya are located on the eastern shoulder of the Kenyan Rift, part of the eastern branch of the EAR. Both volcanic fields formed in the late Pleistocene and Holocene, and both are host to hundreds of pyroclastic cones and maar craters. Previous research has established that trends of linear arrays and the morphologies of cones and craters can be used to establish the locations and orientations of shallow igneous intrusive systems, especially in areas where subsurface feeder dikes are not visible at the surface. The orientation of dikes are controlled by the regional tensile stress field or pre-existing lithospheric structures. Therefore, the analysis of dike orientations can supply valuable information regarding regional tectonic stresses and crustal fabric controls. Manually mapping extrusive volcanic features can be time consuming and subjective. Machine learning object detection and classification can speed up the mapping process, allowing for quicker analysis. Here we present our findings in the application of the YOLO.v.2 object detection machine learning algorithm to identify extrusive volcanic features and its analysis of volcanic fields not included in the training sites. Landsat 8 OLI imagery of Marsabit is used to create training and validation data in PASCAL VOC format. False color compositions of Marsabit and Nyambeni Hills are created from a multiplicative merge of red, NIR, and SWIR bands and the panchromatic band with a resolution of approximately 15 meters. After training YOLO V.2 with training data from Marsabit, it is used to analyze validation images of Nyambeni Hills. Preliminary results suggest that the sharp topographic relief of maars are easily identified in both volcanic fields, while cones are more likely to be identified in Marsabit. Due to eruption style and erosion, the cones of Nyambeni Hills tend to be less prominent than cones on Marsabit. However, YOLO V.2 still identified a few large cones on Nyambeni Hills.