Modeling and Analysis of Groundwater Level in Major Cities of Pakistan using Satellite Imagery and Machine Learning

Abstract

Nature has gifted mankind with finite resources and groundwater is one of these. However, due to population explosion, urbanization and uncontrolled exploitation of groundwater reserves, groundwater resources of this country are at risk. Therefore,

this research focuses on mapping of groundwater levels of the major cities of Pak- istan i.e. Islamabad, Lahore and Karachi using Remote sensing techniques. In this

research, Landsat 8 Operational Land Imager (OLI) top-of-atmosphere (TOA) Collec- tion 2 imagery was used as an input to calculate 3 indices i.e. Near Infrared Reflectance

Vegetation (NIRV), Normalized Difference Water Index (NDWI) and Normalized Dif- ference Moisture Index (NDMI). NIRV has been proven as an indicator of presence of

groundwater through research and a map of groundwater levels for 3 major cities was generated on the basis of these indices. These revealed alarming results as groundwater depletion was highlighted in once preserved areas of Pakistan. Moreover, groundwater levels were found to be reaching alarming levels in the residential zones especially in

Zone I of Islamabad, Old City Lahore and South Karachi. A correlation was found be- tween water index (NDWI) and moisture index (NDMI) with NIRV and it was found to

be 0.88 for Islamabad for both NDMI and NDWI. For Lahore, a correlation of 0.78 and 0.67 was found between NIRV & NDWI and NIRV & NDMI respectively. For Karachi, moisture content was found not to be a significant feature associated with NIRV and groundwater with a correlation value of -0.08. Therefore, the understanding of other climate variables for this region can further improve the performance of our proposed pipeline. The study highlighted the alarming results of groundwater depletion where the depletion has reached 34%, 40% and 27% in the residential zones of the cities of Islamabad, Lahore and Karachi respectively. The Random Forest Classifier modeled

v

as a part of this research, generates results with high accuracy as it correctly classifies areas into groundwater level classes i.e. low water table, moderate water table and high water table zones. Our model achieved an accuracy of 68.7%, 71.5% and 76.8% for the cities of Islamabad, Lahore and Karachi respectively. The only outliers that were detected are due to the fact that the values are near the class boundaries and constant transition between low and moderate groundwater table. Inclusion of other factors like rainfall level, temperature and lithology characteristics can further enhance the performance of this machine-learning model.