REMOTE SENSING GIS BASED MACHINE LEARNING AND WEB APPROACHES FOR URBAN HYDROLOGICAL ASSESSMENT CONSIDERING LANDUSE AND CLIMATE CHANGE

Abstract

Hydrological responses are governed by climate change factors and surface dynamics. Any change in these factors will also impact the hydrological cycle of the region, also affecting humans and natural resources. The crucial factors controlling a hydrological cycle are precipitation, evapotranspiration, temperature, solar radiation, wind and humidity. These factors, ultimately, affect the streamflow and groundwater resources in a basin. On the other side, these factors also get affected by changes in spatio-temporal changes and climatological changes. A periodical assessment of these factors are important to suggest timely interventions so as to avoid any disaster possibilities. This includes faster snowmelt in Himalayan regions due to global warming, increasing temperature or variation in rainfall. So, it required to study a large region for these trends. Urban areas are rapidly expanding with landuse change. This landuse impacts the hydrological factors. This study is much relevant to urban areas which are at a risk due to climate change. However, climate change study and most hydrological factors, including landuse change, require a large amount of data analysis. In this study, most of the analysis is done on selected urban centres such as Vadodara, Chennai and Jaipur. Other analysis such as trend, landuse, urban heat and surface runoff performed on urban areas. This research investigates the impact of landuse change and climate change on the streamflow of urban catchments. Landuse change plays a vital role in the hydrological cycle. The dynamic change of surface properties impacts streamflow, evaporation, surface temperature and ultimately affecting the hydrological cycle. So, this study was necessary. An in-depth analysis of landuse changes and future landuse is performed. Future landuse is generated based on past scenarios. In the study impact of landuse change is performed till 2030, 2060 and 2090. CA-Markov and machine learning-based methods were used to generate landuse. For past landuse analysis and change detection, advance computation was used based on Supervised, Unsupervised, Random Forest, SVM and Extreme Gradient-based methods. R programming language was used to classify an image with extreme gradients, and an SVM-based approach and image classification software was used. The available satellite images from 1972 (60m resolution) to the latest 2020 (10m resolution) were used. At the beginning, trend analysis of selected cities from homogeneous monsoon regions of India was carried out considering the historical data. Future trend of Vadodara city was also carried out. Future climate data were analysed for temperature and precipitations, considering different representative concentration pathways (RCPs). Six bias-corrected and statically downscaled model were considered with RCP 4.5 and RCP 8.5. Baseline data was considered from 1990 to 2017. Future climate data for each model was divided into three periods 2016-2043 (CC1), 2044-2071 (CC2) and 2072-2099 (CC3). These scenarios were investigated on each climate model with different RCP’s. It was found that the MIROC model predicts better results for this study area. Most of the urban catchment have no gauging stations. So, this become a complex hydrological problem for validation in the absence of observed streamflow data. To resolve this issue regionalisation approach was used by considering nearby watershed as donor catchment using observed data. Parameters from the observed catchment was used to drive the flow of ungauged catchment’s simulated flow with machine learning approaches. Same approach was tested for another two catchments with observed flow data and streamflow data generated by machine learning. Higher level of accuracy was achieved for streamflow generation with machine learning. Number of statistical methods were used to test efficiency of observed and regionalised flow. By considering selected climate models and landuse a total 65 number of the simulation were performed for Vadodara city followed by Chennai and Jaipur. This includes ET, Groundwater, Water Yield, Surface runoff and discharge. Water balance is rapidly impacted by landuse. As the urban areas expands, the water balance components particularly evapotranspiration, runoff, groundwater flow etc. largely affected. This results in flow variability in urban areas reoccurring higher flows in monsoon months. Unplanned urban growth is resulting in urban flooding. To overcome this a proper management system is required by taking into account of the future streamflow. Different climate model predicted different results. But results by the MIROC model was more valuable. Most of the model estimated increasing discharge and water yield till the end of the century. Some models estimated decreasing discharge during the first Scenarios. But the second and third scenario show an increasing amount in water balance components. More extreme values were found for RCP 8.5. In this study, urban heat land analysis was also conducted for selected urban areas. The surface runoff model was developed and validated. That model also used to derive runoff for the entire country from 2012-2019. Curve number (CN) for the entire country was also estimated using developed surface runoff model. An online GIS web application was also developed to disseminate rainfall runoff layers to the public. Two GIS portals were developed. The portals used open layers, Geoserver, Apache tomcat and Javascript API. Anyone can add 32 layers of information in this portal and compare different results. The developed portal is updatable to add any new layer from the server end at any time. The user does not require any software to view these layers. This study contributes to quantifying the impact of landuse and climate change on streamflow for ungauged catchments. This study also provides an application of advance computing in hydrology, apart from development of web-portals for easy dissemination.