Machine Learning Based Prediction Model for Suspended Sediment Concentration (SSC) Under the Influence of Anthropogenic Factors in India

Abstract:

SSC plays a vital role in the river system as it affects the riverbank stabilization, ecology of the river; also important for planning, design, and management of the hydraulic structure. Globally, it is significant for maintaining coastal estuaries, coral reefs, geochemical cycle and coastal lines. Several studies predicted SSC based on discharge using Sediment Rating Curve (SRC). Recent decades witnessed the exponential growth of population and economic activities in India leading to significant changes in Land Use Land Cover (LULC). To ensure food security, forest land is converted to cropland/ grassland. The construction of water retaining structures is increased due to additional water demands and urban areas are augmented with the proliferation of population. The anthropogenic factors must be considered in addition to discharge for a more accurate prediction of SSC in India. Considering these factors, a machine learning based regression model viz. eXtreme Gradient Boosting (XGBoost) is formulated to predict SSC using discharge, rainfall, LULC classes and water retaining structures as independent variables. The model is trained and tested on 75% and 25% of data respectively at each gauge station. Daily discharge and SSC of 140 gauge stations mainly in the Peninsular India and data for the dams, barrages, etc. are collected from the Central Water Commission (CWC), Government of India (GoI). Gridded rainfall data is obtained from the Indian Meteorological Department (IMD), GoI. We use the dataset of LULC at a yearly scale from the peer reviewed journal and National Remote Sensing Centre (NRSC), GoI. The performance of the model is assessed at each gauge stations, using R-squared value. The model is predicting SSC satisfactorily at most of the gauge stations with R-squared value more than 0.6 during testing. The model can be used for the future prediction of SSC under different hydro-climatic and anthropogenic scenarios. Policymakers can use this model to analyze the impact of anthropogenic factors on SSC and take preventive measures.

Keywords: Suspended Sediment Concentration (SSC), machine learning, XGBoost, anthropogenic action.