New Thinking in GIScience

This essay discusses theoretical perspectives in GIScience in representing and computing geographic information. Grounding the discussion is the need for new ways of thinking about new facts. Information and geospatial technologies continue acquiring new facts of various kinds. New ways of thinking about these new facts are essential to theoretical advances. Geographic representation encodes new facts to evoke new ways of thinking about them. Geocomputation carries out analytical and modeling procedures to realize these new ways of thinking. Discussions follow the proposed object-field continuum and event-process continuum to capture the essence of geographic representation and computational thinking. While much progress has been made, theories in GIScience research mostly apply existing ones from other disciplines or surround conceptual, logical, or ontological arguments. The lack of a well-defined theory for geographic information presents an excellent research opportunity. Theories for statistics and machine learning are exemplars.

Geographic entities exist in a spatio-temporal continuum, yet traditional geographic information system represents this dynamic geographic world using a map model which is often static in nature. This discrepancy between the dynamitic nature of the real world and the static representation scheme calls for a new system which can capture and represent the dynamic nature of the spatio-temporal continuum. This chapter presents such a system, referred to as Holo-Spatial Information System (HSIS). HSIS consists of two major components: an object-oriented representation scheme, which captures and represents the spatial entities in the dynamic world as multi-granular spatio-temporal objects (MGSTO), and the information management framework, which comprehensively manages the multi-dimensional information in adaptive transformation under different objectifications and granularity abstractions. The development of HSIS will drive innovations in theoretical, methodological, technological and system framework perspectives, which should lead to a drastic change in the landscape of geographic information science.

With past more than 20 years of development, virtual geographic environments (VGE) had gradually matured and formed its own supporting theories and remarkable characteristics. During this period, the remarkable steps forwards of VGE were often inseparable from the promotion of new technologies. Recently, the term of digital twins has emerged and attracted researchers from the community of geographic information sciences to discuss what the digital twins of the physical geographic environments should be alike. This chapter focuses on discussing the conceptual connotations and typical characteristics of both virtual geographic environments and digital twins, analyzes the basic requirements for building digital twins of physical geographic environments, and summarizes whether VGE can match the framework of digital twins of physical geographic environments. The final conclusions of this chapter declare that: The concepts and framework of VGE are essentially consistent with those of digital twins; The characteristics of VGE can absolutely meet the basic requirements of digital twins of physical geographic environments; What’s more, VGE has been more than a digital twin of the physical geographic environments, for instance, it can extensively fit well with the conceptual framework of metaverse of geographic environments which have eight characteristics including identity, friends, immersive, low friction, variety, anywhere, economy, civility.

The latest improvement of sensor resolutions has led to the emergence of various big remote sensing data, which provides great potentials for extracting valuable geospatial information. However, traditional perceptions of these data could not fully capitalize their benefits. In pursuing better solutions to the challenges posed by the volume and variety of new big remote sensing data, we processed the data from different perspectives by transforming the data into curves in frequency domain. The derived curves enabled us to develop a new thinking about big remote sensing data, based on which a theoretical framework was established with new directions on forming commensurate variable types, compatible processing units, as well as matching strategies and algorithms to process and fuse big remote sensing data. The new thinking bears significant theoretical implication and will have widespread adoption with its easiness to be applied to different features of the emerging data.

The term “Geographical Information Science” (GIScience) was formally introduced in 1992, after 30-year development of Geographical Information Systems (GIS). The authors believe that it is the appropriate time to reexamine what GIScience should actually be, as it has reached an age of 30 years. In this article, it is noted that GIScience at its current content is focused on the “G” aspect and deals with the theoretical aspect of spatial data handling. However, it is argued that GIScience should also be a type of specialized information science (or a branch of information science) as geographical information is a special type of information. Then, it is pointed out that the foundation of developing GIScience as a branch of information science has been laid down already and it is time to develop theories behind such a science. This article provides an insight into the future development of GIScience.

The space-place dichotomy has long been discussed in human geography, digital humanity, and more recently in cartography and geographic information science. Place-based GIS are not yet well developed, although there is an increasing interest in semantic and ontological approaches. In this chapter, I present the technological building blocks towards the implementation of an operational place-based GIS that requires the input of platial data from crowdsourced data streams, the understanding of place characteristics and associated human activities and cognition, the support of representation and computational models of place, and the development of platial analysis and visualization. Based on the literature review, I found that the platial analysis functionalities with regard to their spatial counterparts were not sufficiently implemented yet. Therefore, more researches are needed into the development of platial operators for place-based GIS.

We see that GIS is under a major expansion of incorporating more bottom-up methods. The bottom-up approach does not seek to build general/global and therefore likely complicated and delicate models or problem solvers. Instead, it employs local and simple operations, and resorts to intensive computation to achieve the global solution. The burgeoning and adoption of the bottom-up approach are motivated by the contemporary application problems dealt with by GIS, featuring complex systems and high uncertainty, and facilitated by the explosive advancement of modern computing capacity. We use problems of classification, assessment, estimation, and prediction to illustrate the distinction between the top-down and bottom-up approaches. We also point out that an outcome of this new expansion of GIS is that mapping is receding from its center-stage position in GIS.

There are many definitions for geography, most contain the word space or place. In order to foresee the future of geography, let us first examine the presence of the discipline, in particular, its variation in space. This chapter illustrates the distribution of global leading higher education institutions and compare that with the distribution of those leading the study of geography. Are they mostly overlapping? Or in some countries, do they deviate from each other? Among the leading institutions for the study of geography, are they focusing on physical geography, human geography, geographic information science, or all sub-disciplines? Among the leading institutions that are not strong in the study of geography, what are the related disciplines they choose to focus on? Is there a geographic variation in the composition of geographic education? If yes, how to describe it, and how to explain it? Do these patterns reveal any insight to the future of the discipline?

Geography is a comprehensive discipline that studies spatial–temporal patterns, evolution processes, and interaction mechanisms of geographic objects and phenomena. With the evolution of the world from a binary space to a ternary space, it is urgent to deepen and expand the understanding, expression, and mining of geographic information. Most current GIS models use the geometry + combination to express geographic information. Geographic processes, including interplay among features, cannot be directly modeled under the above notion. Geography analyzes spatial and temporal structure of macroscopic patterns as a whole and studies evolutionary processes from the perspective of comprehensive integration, and reveals system structures from the perspective of the integrated role of multiple elements. Based on the concept of ternary space, we identify seven dimensions of geographic information elements, which include semantics, spatial location, geometric structure, attribute, interrelationship, evolution process, and interplay mechanism. We also discussed how such representation framework can be employed under a geometric algebra approach to represent geographic scenes and to achieve a unified representation of the seven dimensions.

Laws are statements of relation of phenomena that currently hold under given conditions and powerful ways for people to communicate and even advance human understanding about the world around us. Currently, three general principles in geography have been named as Law of Geography. The first is the spatial autocorrelation principle, which describes the relation among the attribute values of a given geographic variable over distance. The second is the spatial heterogeneity principle, describing the uncontrolled variance of geographic variables. The third is the geographic similarity principle which describes the resemblance of geographic phenomena under similar geographic configurations (geographic contexts). The Third Law of Geography is different from the first two in that it emphasizes the individual representation of single samples using the similarity in geographic configuration. This focus on individual representation offers a completely new perspective on geographic analyses and knowledge discovery. There are three key issues to be addressed for the Third Law of Geography to fully manifest itself in this capacity: the characterization of geographic configuration; the integration of the individual representation-based techniques with the average model-based techniques; and application of the Third Law in the broader range of geographic subfields and related disciplines.

The neighborhood effect averaging problem (NEAP) was discovered in 2018. It arises when human mobility is ignored when assessing individual exposures to environmental factors (e.g., noise and air pollution). Neighborhood effect averaging occurs because most people move around in their daily life, and as a result, their mobility-based exposures would tend toward the average of the population or participants of the study area. Assessments of individual exposures or their health impacts based only on residential neighborhoods do no capture people’s exposures in non-residential neighborhoods and thus may lead to erroneous findings (because people’s daily mobility may amplify or attenuate the exposures they experienced in their residential neighborhoods). To date, there has been limited research on the NEAP and its effects on research findings. This chapter provides a succinct overview of the NEAP and relevant recent studies on the problem. It also highlights the need to mitigate the NEAP in research and its policy implications, especially concerning the situations of socially disadvantaged groups.

So What is about justifying contribution to knowledge. It is often presented as the relevance, significance, and broader value of the research. Building upon some successful formats from education, medicine, and geography, we present a style to form and answer the So What question in GIScience. The format is Where? Why? How? and SO What? (WWHO). It is also referred to as the “Gazing on the Peak” format, inspired by the famous poem by Du Fu.

Although causal reasoning is a tradition in geographic inquiry, adaptations of statistical and computational causal inference frameworks developed in the past decades have been limited in GIS. To facilitate spatial causal analysis, GIS should develop new data models and software tools for the discovery of causal structures as well as the identification and estimation of causal effects. Event-based and scenario-based spatio-temporal models are promising concepts. Spatially explicit causal models can be developed by integrating spatial statistical models with existing computational and statistical models for causal analysis. There are limits to quantitative approaches to causal inferences; a comprehensive causal analysis should include qualitative analysis.

This chapter provides an applied introduction to model two types of point-based geospatial data using Bayesian methods. Unlike frequentist inference, Bayesian inference describes unknown statistical parameters with a prior distribution. With this foundation, Bayesian approach provides a valuable alternative to analyze geospatial data. We begin the chapter by introducing the basic concepts and benefits of Bayesian inference and survey four selected Bayesian models and methods, including Bayesian spatial interpolation, spatial epidemiology/disease mapping, Bayesian hierarchical models, and Bayesian spatial autoregressive models, for their applications in geospatial data analysis. Then we discuss some popular software packages to perform Bayesian analysis. We conclude the chapter by encouraging geospatial researchers and practitioners to add Bayesian methods in their toolboxes.

Over the past four decades, GIS software products have evolved from workstations to desktops to client–server systems, culminating in today's SaaS-based Cloud computing platforms (Platform as a Service—PaaS). Previous generations of GIS products were mostly self-contained with very limited interactions outside the system. The Geospatial database built on top of the traditional relational database (RDBMS) was usually part of the self-contained system as well. In the first decade of the twenty-first century, Web applications were often lightweight with limited GIS functionality. However, in this new era of cloud computing with SaaS, the whole paradigm has shifted to having full-blown GIS capabilities running in a browser or smart device. In the future, cloud computing will replace much of the desktop functionality we see today. This leads to a whole new way in which a GIS software product should be developed. The traditional waterfall approach to developing a product—from defining specifications, to prototyping, to reviewing with stakeholders—will not survive this rapidly changing era of SaaS. In this chapter, we will discuss the contemporary challenges of cloud computing, particularly with SaaS. We will focus on the various aspects of the development process including design, development, testing, and monitoring of a product system. We will also briefly discuss the critical importance of a team in the success of product development.

The transformation of spatial data into knowledge and understanding through spatial analysis has become an important and ubiquitous element of research and education in numerous fields, especially with support provided by geographic information science and systems (GIS). However, as the complexity and size of spatial data and sophistication of associated analysis approaches have significantly increased, spatial analysis has become increasingly computationally intensive. The focus of this chapter is to address the fundamental challenge of representing and evaluating computational requirements for optimal use of cyberGIS to enable computationally intensive spatial analysis. The chapter describes a computational intensity map (CIM) approach to representing computational requirements of spatial analysis and guiding cyberGIS-enabled spatial analysis. Computational intensity maps (CIMs) are conceptualized to apply the analytical capabilities of cartographic maps and critical spatial thinking to the representation of computational requirements. This map-based formalization allows for the exploitation of critical spatial thinking to evaluate computational requirements for cyberGIS-enabled spatial analysis.

This chapter discusses the challenges of traditional spatial analytical methods in their limited capacity to handle big and messy data, as well as mining unknown or latent patterns. It then introduces a new form of spatial analytics—geospatial artificial intelligence (GeoAI)—and describes the advantages of this new strategy in big data analytics and data-driven discovery. Finally, a convergent spatial analytical framework is suggested as a potential future pathway for spatial analysis.

With rapid advances of geospatial data acquisition technologies, spatiotemporal data have become increasingly available. As the geography and spatial science community is shifting rapidly to embrace the data-rich era, the long-standing challenges facing the spatiotemporal analysis remain not only unsolved but of increasing prominence in producing geographic knowledge out of the rich data. This chapter reviews these challenges posed by the big spatiotemporal data and discusses the recent progresses in addressing them with a particular focus on the promises of deep learning and GeoAI methods. The chapter is then concluded with a discussion on possible future directions.

Geographical modeling has been recognized as a powerful way to solve complex geographic problems. However, its wide applicability is increasingly hindered by its complexity in domain knowledge required and the procedures involved. In this chapter, we argue that domain knowledge plays a key role in making geographical modeling intelligent. Domain-knowledge-based intelligent geographical modeling would not only solve wide geographical problems in an easy-to-use manner on the premise of the effectiveness of the built model specific to the application context, but also contribute to research in artificial intelligence.

VGI (volunteered geographic information) observations are often spatially biased, which degrades the quality of inferences drawn from field sample sets consisting of VGI observations. This chapter presents a novel representativeness-directed approach to mitigating spatial bias in VGI for spatial modeling and prediction. The approach, based on the Third Law of Geography (the similarity principle), defines the representativeness of a field sample set as the degree to which the field sample locations capture the spatial variability of environmental covariates in the study area. Sample representativeness is then quantified as the overlap between the probability distribution of covariate values over sample locations and the distribution over the whole study area. Adjusting the weights for individual sample locations towards increasing the overlap thus mitigates spatial bias in the sample locations and improves sample representativeness. Applications of the approach to species habitat suitability mapping and digital soil mapping demonstrate its effectiveness in mitigating spatial bias to improve the accuracy of spatial modeling and prediction.

Unstructured geospatial data become more and more important in the big data era. Compared with classic spatial data and other big data, the special characteristics of the unstructured geospatial data are investigated and summarized. The key technologies and challenges in data storage, management, analysis, mining, and high-performance computing are evaluated. Finally, future GIS are characterized as smart GIS with real-time sensing, ubiquitous interconnection, deep integration, and intelligent services integrated.

Cartography’s roles in environmentally sustainable development are twofold: first, expressive maps can communicate and narrative environmentally sustainable development; second, digital maps themselves, as device-dependent digital tools, can be more energy efficient in minimizing our impact on the environment. In this chapter, we discuss the concept of green cartography that encourages aligned map design and use for energy awareness. First, we investigate how map design and use impact energy consumption. Then, we discuss the possible ways in which digital maps can be energy-aware, including how to make and how to use energy-aware maps, outlining a series of possible research opportunities.

Shortwave remote sensing signals acquired from vegetation contain information not only for vegetation structure, such as leaf area index and clumping index, but also for leaf biochemical parameters, such as pigments, nitrogen content, water content, dry matter, etc. However, the retrievals of these two types of parameters are generally carried out separately without considering the influence of one type of parameters on the spectral signals used to retrieve the other type of parameters. Since green leaves would be very different from brown leaves in performing photosynthesis and transpiration, we suggest that a next step in vegetation remote sensing be directed towards synergetic retrievals of these two types of parameters for the purpose of improving regional and global carbon and water cycle estimation.

The three-dimensional (3D) structure of forests has long been recognized to have profound effects on forest ecosystems. However, the use of spectral and radar remotely sensed data for forest structure quantification is insensitive to changes in forest vertical structure. LiDAR has emerged as a robust means to measure forest structures. Numerous studies have been devoted to accurately quantifying forest structures from LiDAR data at various scales (from tree branches level to global level) and revolutionized the way we consider forest structure in ecosystem studies. In this chapter, we outline how LiDAR sheds light on forest ecosystem studies and discuss current challenges and perspectives of LiDAR applications.

Earth observation satellites provide important data for monitoring land surface dynamics. In recent years, with the development of new satellite constellations, supercomputing, artificial intelligence, and cloud computing, remote sensing studies of land surface changes have been gradually shifted from sparse time series analysis to dense time series anslysis. Dense satellite image time series dramatically improve our capability for capturing frequent changes in the land surface. It has changed the research questions, data processing techniques, and applications compared with the traditional sparse time series analysis. This chapter discussed the opportunities, challenges, and future directions of dense satellite time series data analysis. It can help researchers from the remote sensing community or other disciplines apply dense satellite time series analysis to solve real-world problems.

Remote sensing collects the primary data for Earth observations. Social sensing especially citizen science offers crowdsourced volunteered geographic information (VGI) as patchworks of geospatial data infrastructure. Digital Earth integrates remote sensing and social sensing by employing Big Earth Data approaches. Via integration, geospatial information can be improved in four domains: spatial (coverage vs. details), temporal (timeliness), social (contextual), and data (credibility). While facing significant challenges in harnessing the soaring amount of spatial and social data, Digital Earth holds great opportunities for geospatial analytics to assist sustainable decision making.

The application of GIS technologies has extended from natural sciences to social sciences. The emerging spatial–temporal big data supported by GIS has broad applications in social governance. Through a unified time–space reference, multi-source big data from different departments can be linked and organized, forming a block data. A cloud platform based on the block data is developed for data processing, data fusion, data analysis, and data mining. This cloud platform can support the management of specific public affairs, such as natural resources management, urban and rural planning, and urban construction. In the future, we need to further explore and use spatial–temporal big data to constantly improve our spatial governance capabilities.

Humanities and social sciences (HSS) are undergoing the transformation of quantification and spatialization. Geo-computation provides effective computational methods and tools for processing geographic information. Geo-computation for humanities and social sciences (GHSS) is a field coupling geo-computation with humanities and social sciences. This chapter introduces the concept of GHSS, and introduces the origin and development, the related theories and methods, and some applications of GHSS. At the end of the chapter, the future development directions of GHSS are discussed.

This chapter outlines four methodological themes in spatial analytics with broad applications in social sciences and public policy, all grouped under the umbrella of “Computational Spatial Social Science”. Spatial accessibility measures the relative ease by which the locations of activities or services can be reached, and serves as a major matric for location advantages. Regionalization constructs regions by merging small areas that are similar in attributes or are tightly connected. The former forms homogenous regions and the latter defines functional regions. Both can be scale flexible and thus produce a series of area units to support analysis, management, and planning. Spatial simulation imitates real-world social, economic, and human environments, behaviors and interactions in a lab setting, and empowers social scientists for discovery and cost-effective policy experiments. Finally, the maximal accessibility equality problem (MAEP) is proposed as a new location-allocation paradigm in spatial optimization to plan public resources and services.

The adoption of spatially integrated approaches has become an increasing trend in social sciences. Concurring with the spatial turn in social sciences, there has been a social turn in geography. Inspired by the theoretical debates in the social turn in geography, particularly the debate around the concept of space, we critically reflect on existing studies of spatially integrated social sciences. Following that, we propose a geosocial analytical framework for a more comprehensive knowledge of our lived society. The geosocial analytical framework should lay geographical research pathways at its center, keeps open to both quantitative and qualitative methods as well as computational technologies, remains interested in any topics relevant to human societies and potentially engages with conventional social theories. Some challenges possibly faced by the implication of geosocial analytics have been identified as well, namely data sources, ethical concerns, and the difficulties in the combination of different research approaches. We take the chapter as an initiative to introduce the geosocial analytics and we encourage more researchers to further work on it.

Uncovering the multi-dimensional interplay between computation and urban life’s spatial-social aspects has both theoretical and practical implications for urban planning and public health science. Many analytical methods have been implemented and applied to deal with high-dimensional, heterogeneous, and unstructured location-based social data drawn from urban locales. Computational urban science has four interdependent layers: human dynamics-centered, platform-based, action-oriented, and convergence-driven. As a research paradigm based on computational thinking and spatiotemporal synthesis, computational urban science can provide a needed framework for addressing many pressing urban sustainability challenges from a systematic perspective.

“Deviation” is common in scientific research, referring to the phenomenon that the output of a process is different from the expected. Deviation may possess various appearances and definitions, e.g., deviation of an observation from the truth, the general trend, or the theoretical value under assumptions, etc. Although in many cases it is perceived by the researcher as unwanted, it may be an inspirer and facilitator, leading to new discoveries and insights from innovative pathways. This chapter initiates a discussion on what and how we can learn from deviations, particularly in urban science. We use several application examples featuring big data and deep learning to illustrate our points.

This chapter presents some variants of the location-allocation problems (LAPs) with additional criteria for service planning such as partial coverage of service demand, contiguous service areas, and equal service areas. The variants arise in applications such as the selection of facility sites for the “15-minute city”, the delineation of public service areas, and the provision of some emergency services in the COVID-19 pandemic. The criteria are formulated as linear inequalities and thus can be added to the classical LAP models. It is challenging to solve those variants, since LAPs are known to be nondeterministic polynomial time hard (NP-hard), and the new criteria may impose further obstacles to the analytical solution. At the end of the chapter, I discuss possible methods to solve the variants.

The transportation system, particularly the surface transportation system, has been evolving, albeit slowly. But that evolution has been exacerbated recently toward a smarter, more sustainable, and more resilient system. Connected and automated technologies enable vehicles smarter; electrical vehicles, and shared mobility, particularly shared micromobility and microtransit system make the transportation systems more sustainable; adaptive and resilient infrastructure planning and design makes the transportation infrastructure system more resilient. These changes represent the future of transportation system, a smarter, more sustainable, and more resilient system, with mobility on demand.

This chapter is dedicated to the concept, characteristics, components, and structures of high-definition (HD) maps and their state of development for autonomous driving. The self-driving vehicle is essentially a rolling supercomputer, controlled more by its software than its hardware. HD maps assume a decisive control role in guiding such a vehicle safely and efficiently through a dynamic environment. Compared to standard maps, HD maps are fundamentally different in terms of generation procedure, map content, map scale and target users. HD mapping is analytically composed of three elements—the “here” mapping, the “now” mapping and the integrated “here” and “now” mapping. The main tasks associated with each element are demonstrated with best practice examples. Key research challenges include extraction of meaningful driving scenarios, edge-case modeling in the absence of training data, predicting contextual human behavior, and safety-first decision making in moral dilemmas.

Land teleconnections refer to the supply–demand relationship of land between distant countries/regions and its socio-environmental impacts. Modeling land teleconnections is critical for understanding the environmental and social consequences arising from land use and consumption. In this chapter, we first explain a widely used analytical tool for the quantification of land teleconnections, and briefly describe several data sources of global/national trade. We then discuss three potential research themes of land teleconnections for future work, such as identifying the functional characteristics of land use, evaluating the land-related impacts of changing consumption patterns, and associating land teleconnections with sustainability.

Crime Geography and spatial analysis of crime has gained great momentum lately, coupled with the advancement of geographic information science (GIScience) and big data in human mobility. According to (Liu in Oxford Bibliographies in Geogr 2021), crime geography and crime analysis normally cover spatio-temporal crime pattern detection, crime explanation, crime prediction, crime prevention and crime intervention assessment. The acronym of DEPPA captures these five elements. Pattern detection uncovers spatio-temporal patterns of crime distribution, such as crime hotspots. Crime explanation aims to discern major contributing factors based on multivariate regression modeling and machine learning. Crime prediction forecasts future crime patterns using machine learning and other predictive methods. Crime prevention devises targeted intervention strategies such as hot spot policing, based on historical and future crime patterns. Assessment examines the effectiveness of crime prevention, to find out if crime is reduced in the targeted area and whether the nearby areas are affected by the intervention. This chapter summarizes some of the latest progresses and challenges of crime geography and crime analysis along the issues of the unit of analysis and spatial scale, comparison analysis, new data and new variables, crime prevention and assessment, and the spatio-temporal mismatch problem.

The chapter looks at the contribution Geographical Information Systems (GIS) have made to research into the spatial and temporal patterns of urban crime and criminality, indicating areas of possible growth in the future and sketching the policy relevance of these developments for practical policing and research. It discusses four main GIS empowered crime research elements using a “3W1H” framework to help organize ideas: “Who (2P: Police and Public)”, “What (2C: Crime and Context)”, “Why (2W: When and Where)” and “How (2D: Data and Design)”. We highlight how interactive data visualisation leads to better public communication, essential for successful policing, as well as contributing to research agendas.

This chapter recounts personal anecdotes of augmenting public health infrastructure using GIS. I start with what motivated me to work primarily on real world problems in public health data integration and disease surveillance. In the process of enhancing spatial data capacity and GIS solutions, a key to success is being flexible while working with public health programs for their other needs. Enabling visualization of multiple disease maps for multiple programs is a good starting point from maps to mechanisms. We must continuously build other community health datasets to snowball programs’ inquiries. Expanding GIS capabilities in disease cluster detection is technologically easy, but organizationally challenging, which requires domain knowledge and design thinking. In the Big Data era, GIS bridges precision neighborhood health with precision medicine to improve population health at both individual and neighborhood levels.

GIS has been widely used in geochemistry-related environmental health studies and practices to map and analyze sampling locations and spatial distribution of geochemical features and health information. In the big data era, the focus is shifting towards revealing the hidden patterns and features. This chapter explores the challenging issues of spatial analysis, machine learning, and uncertainty in such studies and practices. Spatial analysis needs to focus more on hot spot analysis and identification of spatial outliers, as well as exploration of spatially varying relationships. Machine learning can be adopted to conduct deep learning with a focus on non-linear features and their links with causal effects. The field and laboratory uncertainty of environmental geochemistry should be incorporated in GIS analysis. The analyses of the association between environment and health need to be more intelligent and accurate. GIS continues to provide useful tools to make novel findings in environmental geochemistry and health from the spatial aspect.