Performance Evaluation of MBT, OBIA and PBRF Method for High-Altitude Water Body Mapping: A Case Study of Mago River Basin of Arunachal Pradesh

• Published in: Journal of the Indian Society of Remote Sensing 2023-11, Vol.51 (11), p.2187-2200
• Main Authors: Mandal, Sameer, Bandyopadhyay, Arnab, Bhadra, Aditi
• Format: Article
• Language: English
• Subjects: Accuracy; Altitude; Earth and Environmental Science; Earth Sciences; Ground truth; High altitude; Image resolution; Mapping; Noise reduction; Performance assessment; Performance evaluation; Remote Sensing/Photogrammetry; Research Article; River basins; Rivers; Shadows; Water bodies
• ISSN: 0255-660X; 0974-3006
• DOI: 10.1007/s12524-023-01759-0

Accurate mapping of high-altitude water bodies in a hilly region is a challenging task due to the hill shadows that have a similar signature as water bodies and the presence of ample amounts of small water bodies. This study compared three techniques (MBT, PBRF, and OBIA) to map water bodies in the Mago river basin, Arunachal Pradesh, a part of the Eastern Himalaya. Spectral indices (MNDWI, NDWI, NDTI, and slope) were used as input variables. The study also assessed the performance of the 'VV' band of Sentinel-1 SAR data in MBT and PBRF (MBT v and PBRF v ) compared to methods without using the 'VV' band (MBT n and PBRF n ). An accuracy assessment was carried out using ground truth data collected through a partial field survey, high-resolution Google Earth imagery, and visual interpretation of raw data due to inaccessibility of the area. It was found that there are 224 high-altitude water bodies present in the basin with an area of 4.77 sq. km based on the ground truth datasets. All three methods were applied to map the high-altitude water bodies and an accuracy assessment was performed. The PBRF v shows the highest overall accuracy (85.3%) with a kappa of 0.83. In comparison, both the OBIA and MBT v method produces the similar result as 71.4% and 71.7% with a kappa of 0.71 and 0.70, respectively. On the other hand, the overall accuracy of the MBT n and PBRF n method was found to be 67.1% and 83.3% with kappa as 0.63 and 0.83, respectively. The total area of the water bodies using the BPRF n method (4.60 sq. km) and BBRF v (4.62 sq. km) was found more closer to ground truth data (4.77 sq. km) compared to all other methods. MBT n has the highest number of misclassified water bodies, while PBRF v showed no misclassifications. However, PBRF v missed to map a few smaller water bodies (33 in total) due to their size. The slope factor effectively reduced noise caused by hill shadows, as water bodies were mainly found on gentle to flat land. Therefore, the PBRF v approach was considered the most suitable method for accurately mapping high-altitude water bodies compared to the other technique.