Incorporation of Machine Learning in the Management of Rawal Watershed

Abstract

Watershed areas in underdeveloped countries are strategic resources for agri- culture and domestic purposes. These sheds are not adequately protected from contamination, caused by anthropogenic activities and spillways' dis- charge. The contaminated water is adversely a ecting the ecosystem. So the quality and storage parameters are of serious concern for all water resources management authorities. Conventional methods cannot cope with the root cause analysis of water reservoir contamination and discharge. For cost ef- fective and predictive water management, it is essential to analyze di erent aspects of water quality and storage with emerging modeling, mining and learning techniques. The quality indices are analyzed by the combination of supervised and un-supervised machine learning techniques. As a case-study we selected the Rawal watershed area used for irrigation and domestic pur- poses of twin city Islamabad and Rawalpindi of Pakistan. Di erent regression models based on monthly and quarterly datasets, to check the seasonal water quality trends were developed. In order to determine how much parameters satisfy the WHO quality standards, the parametric satisfactory analysis was carried out. For quality indexing, Hierarchical Clustering and Multilayer Perceptron has been found more accurate technique. Higher values of fecal coliforms were found in the months of March, June, July, and October. The predictions of hydrological parameters to manage water discharge were made for the year 2013 using regression and time series forecast models. The results show that August is a crucial month for in ow and spillways discharge; while September and October are critical for level and storage capacity. J48 tree classi cation technique has been found more accurate supervised machine learning technique for discharge management. Similarly, in order to forecast the water level, the improved SVM technique has been found to be more ac- curate. Finally, using regression and forecast models, it has been found that in year 2013, water level as well as the storage capacity will remain below the spillways gates opening threshold values.