Efficient Computation and Storage Techniques for Spatial data

Abstract

Considering the proliferation of geospatial and IoT solutions with a large array of sensors, there is a tremendous amount of GPS and geo-tagged IoT spatial data being generated. Analysis of spatial data is commonly used to answer queries such as places near me , or localities that have parks , closest pairs of houses or farthest pairs of shops , densely/sparsely populated localities , skyline plots for the landscape or the trends plotted as region-wise real-estate price, shopping patterns, etc. Spatial analysis, which makes use of spatial queries such as join, convex hull, etc work on a wide variety of spatial data sources. This results in a variety of spatial data formats and algorithms. The amount of spatial data being generated from the geotagged and/or IoT devices has also been projected to grow to 73 Zeta Bytes by 2025. The complexity and volume of spatial data have a huge impact on the performance of spatial queries. Shape complexity for example, large number of vertices, irregular shapes due to natural variances results in challenges in representation, storage, and computation with such data. newlineThis thesis explores the impact of shape complexity features on the accuracy of data representation and the efficiency of spatial algorithms. The thesis also aims to provide optimizations for spatial data storage and spatial data processing techniques by making use of hybrid storage systems - Hard Disk and Solid-State Drives and parallel computing models CPU and GPUs. These strategies are primarily designed to optimize the frequently used spatial operations such as spatial join, closest pair computation, and spatial index generation. Optimization strategies addressed both computational and storage hierarchy changes to achieve enhanced performance. As the first non-functional requirement towards the spatial data format conversion and impact on data accuracy, this thesis converts real-world spatial data (LULC - Land Use Land Cover) from raster to vector format and analyses the accuracy of the vectorized