Expression of opinions are communicated on social media platforms, and these records can be analyzed and applied to a variety of spatial research questions. Using geo-tagged social media posts, specifically from the microblogging application Twitter, this research investigates the following research question: Assuming that users’ perceptions of crime can be accurately extracted and analyzed from social media data, to what degree do they match with actual geo-tagged crime incidents in time and space? This research could be serviceable to officers of the law for effective administration of citywide resources with the mission to weaken crime in areas most affecting residents’ perception of their safety. The City and County of San Francisco was chosen for its availability of crime data and Twitter data. Methods applied to explore the posited research questions included preprocessing the Twitter content to remove unnecessary information such as URLs plus retweets. Sentiment analysis in RStudio separated the entire corpus of tweets into emotional categories. Finally, a series of point density and emerging hotspots maps were created to explore the relationships within the data. Comparison between the hotspots maps for the rates of crime and for the rates of tweets reveal different similarities and discrepancies based upon the time range used for crime. Based on the comparison between spatial and temporal patterns of crime and tweets showing fear, the interpretation is that the northeastern areas in San Francisco were more likely to be the site of crime on Fridays around 6 p.m., plausibly larceny. Social media like Twitter may prove as effective indicators or perhaps even predictors of crime in space and time.

This dissertation examines the current state of automated indoor mapping and modeling using point cloud data produced by close range remote sensing systems. The first part looks at reality capture techniques that convert the physical form of indoor spaces into point clouds of millions of measured points, each with an (x,y,z) coordinate value. The second part examines methods for teasing out geometries from these point clouds -- often complicated by noise and voids -- and converting them into 3D geometric models. The final part examines techniques for merging the coordinate reference systems of these indoor maps and models with those of the outdoor world, resulting in a seamless representation of space. Lessons learned in this study revealed that theories, techniques, and practices in indoor mapping remain relatively elementary compared to those for the outdoors, yet they also present significant opportunities for future research propelled by emerging developments in remote sensing and a growing demand for indoor maps.

Modern hydrologic modeling and data systems are growing in scale and resolution to answer urgent, data intensive questions related to water resources. The focus of this dissertation is how real-world entities can be represented as flexible and efficient computational elements that capture processes and ease the exchange of information. Part 1 focuses on evaluating the state-of-the-art National Water Model (NWM) and Height Above Nearest Drainage flood mapping techniques to establish a community benchmark for improvement. We propose a outline for what an intelligent emergency response system might look like with these systems in place but note challenges related to (1) data usability (2) assumptions of spatial homogeneity in explicitly heterogeneous phenomena, and (3) the application of historical conventions that are inefficient in a federated system.

Part 2 focuses on (1) improving the data format of NWM streamflow output (2) parameterizing Manning’s roughness across the national river network for improved river depth estimates, and (3) the digital representation and estimation of hydraulic relations at the national scale for increased interoperability. Part 3 utilizes these techniques to reimagine a flood forecasting system that circumvents the need to produce flood maps and instead computes feature-level flood forecasts through a relational data system that allows building-level and demographic flood assessments to be extracted from a national inventory in a matter of seconds for any area of interest.

Lastly, part 4 explores the role of land cover as the mediating layer of human, water, and energy processes. The questions explored include how land cover data is best represented at coarser scales and what are the hydrologic consequences of variable spatial allocation and static land cover data. In approaching these, we develop a new raster resampling technique, evaluate the hydrologic influence of fire over California in the last two decades, and quantify the sensitivity of hydrologic models to different land cover representations. The dissertation concludes by sharing a simulation of California land use change out to 2100 and a discussion of the potential impacts faced.