Studying erosion, sedimentation and flooding pattern of Brahmaputra River using Google Earth Engine

The Brahmaputra River is amongst the world’s largest sand-braided rivers which has high erosion accretion due to annual flooding. This research aims to conduct morphological analysis in one of the river reaches of Brahmaputra to identify the areas prone to erosion-accretion through spatial-temporal analysis to be used in decision making. Cloud based Google Earth Engine (GEE) is an open access platform which gives us the opportunity to use existing algorithms to conduct morphological analyses in rivers such as the Brahmaputra. Use of supervised classification in identifying water pixels in a satellite image is called surface water extraction. Surface water extraction is very important in understanding the ecological and hydrological process. The study has been conducted in the river reach of about 80 km, situated between Tezpur and Guwahati. This reach faces heavy floods almost every year and is used for irrigation, navigation by the people living in the floodplains. The methodology consists of two parts, a morphological study of the river reach and the application of a supervised algorithm to identify water and non-water pixels in satellite images. In the first part we choose the best water index and then perform a bank migration and erosion-accretion analysis for two different years. The second part consist of choosing the most suitable classifier in supervised classification for surface water extraction and then using the chosen classifier for water index classification. For morphological analysis the best water index for classification was chosen from Normalized Difference Water Index (NDWI), Normalized Difference Moisture Index (NDMI), Modified Normalized Difference Water Index (MNDWI), Water Ratio Index (WRI), Automated Water Extraction Index (AWEIsh), Automated Water Extraction Index (AWEInsh), Tasseled Cap Wetness (TCW), Water index-2015 (WI2015), Simple Water Index (SWI) and Multi-Band Water Index (MBWI). Among them, NDWI was chosen as the best index through it’s high score in overall accuracy (OA) 97%, Kappa-coefficient (k) 0.91 and Overall success index (OSI) 0.91, producer’s accuracy(PA) 98% for non-water, 92% for water, user’s accuracy (UA) 97% for non-water, 93% for water. The morphological analysis from 1990 to 2020 show erosion accretion are related directly with the occurrence of flood. The 1998-1999 have highest annual accretion 147.7km2 and 2004-2005 have highest annual erosion 143.3km2. Average bank migration has been used to identify volatile areas along the river reach and mark them to be used while decision making for river training. The narrower sections show higher bank migration. The places near Roumari, Laharighat and Baghpori Chapori, Chereng Chapori and Bhuragaon have been classified as most fluctuating. For surface water extraction supervised classification algorithm has been built using (blue, green, NIR and SWIR1) bands. Three different classifiers: classification and regression tree (CART), random Forest (RF) and support vector machine (SVM) have been used to classify Landsat images from 2010 to 2018. These algorithms can also extrapolate classification outside the study area. The CART and SVM does not perform as well as RF in classifying all the images from 2010 to 2018. The RF classifier performs the best in overall validation accuracy, kappa co-efficient, producer’s accuracy and user’s accuracy for water and non-water classification. Then RF classifier and NDWI was used to build training data to be used for supervised classification for each year. The classification results using water index derived best results with all the accuracy results ranging from 1 to 0.96. Water area was calculated from 2010 to 2018 using the pixels classified as water. The area of water is highest 233.26km2 in 2018 and lowest 151.22km2 in 2010. Accuracy assessment of the NDWI classified images was done using Joint Research Centre (JRC) monthly water history maps and they show high resemblance between extracted surface water. It can be concluded that supervised classification is a very simple, effective and versatile technique for surface water extraction using GEE