The world we live in is in constant flux. Land masses are growing, shrinking and moving and modern remote sensing technology provides us the means to observe these phenomena with increased detail. A group of researchers recently published an interesting paper in the IEEE Journal of Selected Topics in Applied Earth Observation and Remote Sensing, entitled “Extraction of Glacial Lakes in Gangotri Glacier Using Object-Based Image Analysis”.
Mitkari et al., utilized eCognition’s OBIA technology to accurately identify glacial lakes from high resolution satellite data in the Gangotri glacier area of the Indian Himalayas. As global temperatures rise, glaciers are retreating, resulting in an increased number of glacial lakes which in turn give rise to outburst flood events and have an additional warming effect on the glacier. It is therefore important to regularly monitor such water bodies.
In the past, mapping efforts were based primarily on conventional field survey methods. These are “laborious, time consuming, expensive, and can be risky in high mountainous regions”. In addition, the literature survey performed by the authors revealed that traditional pixel-based remote sensing techniques, based on NDWI and band ratio approaches significantly “misclassify shadows as lakes and this further increases the task of manual removal of the shadows”.
The OBIA approach explored by Mitkari et al., integrates both high resolution (5.8m) multi-spectral (Red, Green & NIR) LISS-IV satellite data and ASTER Global DEM v2 elevation data into the glacial lake extraction workflow. eCognition allows for the combination of features from optical and elevation data into class descriptions for a more robust descriptions of the objects we see. In this case, the authors used eCognition’s tools to create a new index, the “Normalized Difference Supra Glacial Lake Index” (NDSGLI):
NDSGLI = [(ΒGreen – ΒNIR)/(ΒGreen + ΒNIR)]/Slope
The use of the NDSGLI in the eCognition Rule Set, provided the authors with the means automate the extraction of glacial lakes with an accuracy of 94.83% within the study area.
This projects demonstrates the strengths of data fusion within eCognition – the ability to easily combine features from optical and elevation data sets for improved classification accuracy.