*Editorial* **Artificial Intelligence for Multisource Geospatial Information**

**Gloria Bordogna \* and Cristiano Fugazza**

CNR–IREA, Via A. Corti 12, 20133 Milano, Italy

**\*** Correspondence: bordogna.g@irea.cnr.it; Tel.: +39-02-23699-299

#### **1. Introduction**

The term Geospatial Artificial Intelligence (GeoAI) is quite cumbersome, and it has no single, shared definition.

An initial, narrow definition characterizes GeoAI as the application of machine learning toolkits to the context of Geographic Information Systems (GISs) in order to simulate future scenarios via data classification and smart predictive analysis with respect to several events and phenomena, such as the occurrence of disasters, human health epidemiology, and the evolution of ecosystems and biodiversity, which, in turn, is undertaken in order to respond to communities and support community resilience by processing traditional kinds of geographic information represented in digital cartography [1].

Another wider definition considers GeoAI as the processing of Geospatial Big Data (GBD) of heterogeneous forms and sources, including both traditional digital cartography managed by GISs, remote-sensing-based multidimensional data including images and image time series, georeferenced unstructured and semi-structured texts, and complex geo databases, with a focus on the geographic dimension [2].

Thus, the application of techniques from AI and data science to GBD, via the exploitation of high-performance-computing platforms, are merged into GeoAI in order to understand natural and social phenomena.

A general definition characterizes GeoAI as the use of artificial intelligence methods, including machine learning and deep learning, to produce knowledge through the analysis of spatial data and imagery [3]. In this sense, GeoAI is regarded as an emergent spatial analytical framework for data-intensive geographic information science, facilitating both environmental sensing and so-called "social sensing" by exploiting both the digital traces people leave behind as they interact with the IoT and the user-generated digital content created on social networks to understand the dynamics related to human mobility patterns and social phenomena.

Moreover, the specificities and importance of the geospatial dimension; its heterogeneity in terms of both conceptualization based on either "place" or "space"; the varied formats of spatial information; the different scales; the need for representing distinct geosemantics, i.e., the semantics of locations; and the different needs of analysis dictated by the goals of the applications, which often necessitate geospatial and temporal reasoning, pose new challenges and opportunities with respect to AI.

The current research topics include multiresolution and multisource GBD fusion; the multiscale geosummarization of information to improve the quality of GBD; multisource, heterogeneous GBD integration for data reuse; and experimentation in deep learning applied to multispectral remote-sensing images, such as CNN, RCNN, LSTM, and GANs generally used for RGB pictures. Finally, GeoAI must bridge the gap between opaque technologies, such as deep learning, which are generally regarded as black-boxes, and more traditional and transparent machine learning approaches to knowledge management, such as decision trees; KNN; clustering algorithms; data mining; soft computing, including genetic algorithms and fuzzy logic; ensemble approaches; and semantic representation and analysis. This can facilitate the advancement of the features of explainable AI, which

**Citation:** Bordogna, G.; Fugazza, C. Artificial Intelligence for Multisource Geospatial Information. *ISPRS Int. J. Geo-Inf.* **2023**, *12*, 10. https:// doi.org/10.3390/ijgi12010010

Received: 22 December 2022 Revised: 23 December 2022 Accepted: 27 December 2022 Published: 30 December 2022

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

constitutes a mandatory characteristic of software when used for critical tasks impacting people's safety and security, such as in the health and law enforcement domains.

Our motivation to organize this Special Issue stemmed from an observation of the increasing number of academic papers focused on the application of GeoAI and on the evaluation of its potential to analyse natural, environmental, human-driven, and social changes and events.

Nevertheless, the Special Issues published at the launch date of our proposal mostly conceived of GeoAI in the strict sense, and not in the broader view we have already addressed in this Special Issue, wherein we welcomed approaches that merged multisource and heterogeneous GBD.

This Special Issue has received a total of 20 submitted papers; 10 of these papers have been accepted.

The authors' affiliations correspond to the following countries: Italy, Egypt, the United Arab Emirates, South Korea, Turkey, Kazakhstan, China, and the US.

The contributions can be grouped into three main topics:

(1) Social sensing by mining geotagged, user-generated content and traces in the form of either semi-structured textual data or photos;

(2) Environmental monitoring and analysis by employing remote-sensing spatial temporal data;

(3) Methodological approaches to integrating, mining, representing, and interpreting multisource and multidimensional spatial-temporal data.

#### **2. GeoAI for Mining Geotagged, User-Generated Content and Traces**

Within this section, we consider the descriptions of original approaches developed to classify and mine geotagged user-generated content and traces, which are created either purposefully or unknowingly by users of social networks:


require the creation of a classification category in advance; moreover, it is capable of flexibly extracting categories for each tourist destination and improving classification performance even with rather small data volumes.

(iii) *"Detecting People on the Street and the Streetscape Physical Environment from Baidu Street View Images and Their Effects on Community-Level Street Crime in a Chinese City"* by Han Yue et al. [6] is another example of a social-sensing application, which, in this case, is used to assess street crime via traces unknowingly left by Baidu users. This study is the first to combine Street View images (Baidu Street View), deep learning algorithms, and spatial statistical regression models to retrieve the number of people on a given street and the features of the visual streetscape environment to understand street crime. Finally, this study determines the quantitative measurement of people on a given street and the set of streetscape features that has potential influences on crime by combining the outputs of two deep learning networks. Specifically, they found that the number of people on the street had a significantly positive impact on the total street crime assessment.

#### **3. Remote-Sensing Spatial-Temporal Data for Environmental Monitoring**

This section groups three articles that provide novel approaches to the application of GeoAI methods to interpret remote-sensing spatial-temporal data, either acquired from LiDAR or from sensors on satellites. They apply a range of different machine learning and deep learning techniques for distinct environmental applications and tasks, and assess the accuracy of the results by running experiments on real data:


#### **4. Methodological Approaches to Dealing with Multisource and Multidimensional Numeric and Alphanumeric Spatial-Temporal Data**

This section considers approaches whose focus is primarily on the methods for the integration, management, querying, and mining of multisource and multidimensional numeric and alphanumeric spatial-temporal data. Specific attention is paid to the inherent inconsistencies and uncertainty of the spatial temporal information:


functions and fuzzy classes; then, it generates fuzzy association rules. Therefore, FSOLAP does not require predefined sets of fuzzy inputs. This approach is applied to handle non-spatial and fuzzy spatial queries, as well as spatiotemporal fuzzy query types. Additionally, FSOLAP is not only used to query and analyse historical data but also to handle predictive fuzzy spatiotemporal queries, which typically require an inference mechanism.

(x) "*Implicit, Formal, and Powerful Semantics in Geoinformation*" by the authors herein and our colleagues Paolo Tagliolato Acquaviva D'Aragona and Paola Carrara [13] addresses the need to identify suitable methodologies and frameworks in order to represent and mine GBD depending on their geosemantics—whose classification is often ill-defined. A meta-review of the state of the art in geosemantics is performed to pinpoint relevant "keywords" representing key concepts, challenges, methods, and technologies of the domain. Then, real case studies dealing with geoinformation are first categorized based on three forms of semantics, defined as implicit, formal, and powerful (i.e., soft) depending on the kind of the input data they use; consequently, they are successively associated with the previously identified relevant keywords for the domain of geosemantics. Finally, the similarities between each pair of analysed case studies in the space of the keywords are computed in order to ascertain whether distinguishing methodologies, techniques, and challenges can be related to the three distinct categories of implicit, formal, and powerful. The outcomes of the analysis identified the methods and technologies that are more suited to modelling and processing specific forms of geosemantics categorised into implicit, formal, and explicit categories.

#### **5. Conclusions**

The contributions published in this Special Issue offer a panoply of techniques and approaches used to deal with GBDs by means of a variety of GeoAI methods. The approaches are varied with respect to the objectives of their studies, which include both social and environmental sensing, as well as with respect to the kind of GBD sources, genre, and formats. While remote-sensing data from satellites and sensors are used mainly for environmental applications, social media-georeferenced data, both textual and pictorial, are mainly used for social applications. In both domains, a current trend is to apply deep learning methods and to compare the results achieved with baselines or with more traditional machine learning algorithms.

Besides the mainstream deep learning methods, some bucking methods were also proposed by some of the papers, such as the use of transparent machine learning algorithms based on soft computing and fuzzy logic. This was motivated by the need for the analyst to have greater control over the automatic process in order to be able to understand the phenomenon and to explain it to stakeholders.

Some methodological proposals outlined the need to tackle new challenges with respect to GBD management, including the need for novel means for multisource GBD integration and transformation as well as uncertainty and imprecision management. Finally, from a meta-review of approaches, a synthesis is proposed in order to outline the most suitable GeoAI methods for managing GBD depending on their geosemantics.

We are also aware that the collected contributions and their topics do not exhaustively cover all of the challenges related to GeoAI. For example, other, unincluded challenging topics of GeoAI concern spatial-temporal and thematic solutions, which entails the ability to answer user questions regarding the retrieval of relevant information from heterogeneous, multisource GBD, thus satisfying user needs related to specific geographic areas and to a desired time range, such as "find a well-reputed pizza restaurant close to Milano railway station which is open on Monday evening". Another issue in the perspective concerning the reproducibility and replicability of experiments is the need for high-quality, labelled GBD benchmark collections that are freely available and allow the research community

to compare the proposed methods. While this practice is well-established in the textual information retrieval field, it is still at its infancy in the geographic research community [2].

Finally, we believe we are still in the early stages of integrating and analysing multisource and multimodal GBD using GeoAI methods, including geotagged voice and audio files, remote-sensing images and their derived products, and geotagged text annotations, which have been collected as natural and environmental observations in many citizen science projects. The application of GeoAI methods based on embedding representations may constitute a quantum leap in multimodal GBD integration.

**Author Contributions:** Both authors contributed equally to the conceptualization and writing of the editorial. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We want to express our congratulations to the authors of the papers for their interesting works; to the anonymous referees whose key help made it possible to improve the contents of the papers; and finally to the editorial staff of IJGI for their excellent assistance in producing this Special Issue.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
