The geospatial industry in South Africa stands at a pivotal moment. We must wholeheartedly embrace the artificial intelligence (AI) revolution unfolding globally, or risk being left behind by professionals from outside our field who will. AI already possesses the remarkable ability to automate nearly every aspect of our work as geospatial scientists, from data extraction and database development to map production, data analysis, metadata creation, and even contributing to policy and strategy development.

We must also not to be naive about the power of AI in the geospatial industry, yet we must also exercise caution in its application. AI does not resolve all issues and fulfil all geospatial functions. GIS professionals will remain essential for validating and interpreting of AI outputs, testing them against established theoretical frameworks and the reality we understand.

AI’s journey has been around since the 1940s where it was designed to emulate the human brain. A significant breakthrough occurred in the 1980s and 90s with backward propagation algorithms, leading to more detailed use of neural networks in real-world applications. The 2000s and 2010s saw the rise of deep learning, offering multi-layered networks capable of processing increasingly complex spatial data and imagery. Now, in the 2020s, we are witnessing the introduction of large language models (LLMs) and AI agents, which has brought a quantum leap in allow us to leverage natural language in combination with AI models to achieve spatial reasoning and intuitive interaction with GIS analysis.

AI: Bridging the Gap in Geospatial Data Complexity

A critical role for AI in the geospatial industry is its capacity to bridge the gap between the inherent complexity of analysing geospatial data and its practical application. International research indicates that only about 10% of the global population can perform spatial problem-solving, although this might have changed with wider exposure to spatial data via tools like Google Maps. Still, many, including data analysts, find it challenging to use geospatial data and GIS software for analysis and interpretation. LLMs are making this process significantly easier, exposing our geospatial ddata and analytics to a much broader global audience.

The challenges we face with geospatial data are considerable:

• Complexity: We deal with high-dimensional data from multiple sources.
• Non-linearity: Multiple layers of data are often nonlinear, complicating traditional analysis methods.
• Unstructured and Textual Data: A large amount of data exists in unstructured or textual formats.

This is where AI’s incredible strength lies. It can model nonlinear data, automate processes, and integrate multiple diverse sources to analyse and identify hidden patterns that traditional methods, often univariate or bivariate analysis, might miss. AI allows for a multivariate approach with various data types and sources.

AI in Action: Common Geospatial Applications

Some of the most common applications of AI in geospatial analysis include:

• Classifying satellite imagery: Using Artificial Neural Networks (ANNs) satellite imagery can be classified to identifying land cover. By integrating other information layers, ANNs can go further, analysing land cover and predicting land uses with much higher accuracy.
• Geodemographic segmentation: Using Self-Organizing Maps (SOMs), a type of neural network, population characteristics, human behaviour, and consumer behaviour can be clustered to predict lifestyle groups.
• Object Detection and Extraction: An increasingly powerful area where AI uses symbols and features to identify and extract information like roads and buildings from aerial photography.

The critical advantage of AI in the geospatial environment is its ability to model nonlinear data and integrate data from a multitude of intuitive sources, enabling a better understanding of observed patterns.

Real-World Impact: Pioneering Case Studies

GeoScope since its establishment 18 years ago has used AI to develop geospatial datasets. Below are several case studies demonstrating AI’s tangible capabilities:

• Lifestyle Segmentation : Many years ago, Kohonen self-organizing maps (SOM) was used to cluster the 2001 census data at a granular level. Variables included population size, household income, house style, ownership, employment levels, sex ratios, population group, age profiles, home language, education, and area type. The lifestyle segmentation, resulting in 20 distinct categories being identified, such as “highbrow” (high-income suburban residents), “Jongens” (younger populations with moderate income in formal housing), “Eastern Nkosi” (rural isiZulu-speaking people with lower service access), and “Mjondolas” (densely populated informal settlements). This work showed the diversity of informal settlements and townships across South Africa and underscored the importance of understanding these groupings’ characteristics for targeted service delivery by government and commercial opportunities.
• Living Standards Measures (LSM) and Socio-economic Measures (SEM): More recently neural networks have been employed to model and map South Africa’s LSMs and SEMs. These proxy indicators offer crucial longitudinal trends on poverty and wealth. Using the General Household Survey (GHS) initially and later the Marketing All Product Survey (the largest consumer survey in SA with 20,000 annual interviews), we integrated household survey data with secondary datasets.

A general regression neural network is used to impute data from interviewed areas to over 103,000 spatial enumeration areas across the country. This invaluable information is used by companies for target market definition and by entities like Eskom for demand forecasting related to electricity provision. AI is one of the most effective mechanisms for this ongoing, regular data update.

• MSMEs in Africa: For several years, we’ve mapped the concentration and distribution of micro, small, and medium enterprises (MSMEs) in select African countries. In Malawi, a survey was conducted, and that data was integrated with about 14 universally available geospatial datasets covering the continent. By creating a model between the survey data and these geospatial datasets, we could indicate MSME concentration and distribution across Malawi. This model was then successfully applied in South Africa, Lesotho, Eswatini, and recently Tanzania, with adjustments made using other secondary datasets.

This information serves as a crucial sampling frame for organizations to conduct more detailed analyses of MSMEs, especially regarding their financial access. Key factors in defining MSME concentration and character include proximity of interviewed people to roads and nightlights (reflecting economic activity and residence), along with terrain, slope, and population data. This process involved providing integrated data to a combination of machine learning algorithms, which analysed and identified the most accurate predictive models for data imputation.

Glimpsing the Future: Emerging Trends in AI Geospatial Applications

The future of AI in the geospatial industry holds exciting prospects. Some of the applications are described below.

• Real-Time Spatial Information Use: Spatial information will increasingly be used in real-time for decision-making, such as in disaster and traffic management. In the Western Cape, AI agents are already used in traffic management services to deploy vehicles where needed, with operators interacting directly with the AI. This also supports the development of smart city infrastructure.
• Automated Feature Recognition: Significant work is ongoing in automated feature recognition for housing types, including the complex task of differentiating informal settlements in dense areas. AI will improve classification and enable higher resolution predictions of dwelling concentration and types at a local level, as well as road extraction and land use classification.
• Digital Twin Cities: Neural networks will play a critical role in defining aspects like predictive maintenance for buildings and roads, and understanding urban growth, which is a dynamic, non-linear process requiring substantial information.
• Climate Models and Demographics: AI is increasingly integrated into climatic models, improving prediction accuracy. In demographics, artificial neural networks have been used to update population estimates for South Africa to 2024 levels at an enumeration area (EA) level, uusing multiple layers of information.
• Ethical Considerations: As AI is integrated more effectively into the geospatial industry ethical aspects of AI use will need to be addressed, ensuring transparency and fairness in information precision. GIS professionals will play a vital role in ensuring the use of ethical and true practices.

Two applications that reflect the potential of AI in the geospatial industry are described below.

• AI GeoNavigator: An AI GeoNavigator application is being developed, based on Ireland’s GeoDirectory Address database (considered by Google to be the most sophisticated address database globally) with 2.3 million records that is updated annually to enable a broader audience of users to analyse the data. Despite its extensive information, only a small percentage of the population with geospatial skills can analyse this data. By integrating LLMs, more people will be able to analyse this data, leading to more sustainable and profitable development of products like retail facilities, offices, distribution centres, or service facilities.
• AI Swarms: A significant movement is the development of AI swarms – combinations of AI agents working collaboratively. Their development in the geospatial environment will be cutting-edge and facilitate the automation of many geospatial tasks from data extraction and analysis to map production, visualization, and metadata compliance. All through natural language models and API creation for integration into GIS software.

Africa’s AI-Powered Geospatial Renaissance

The availability of geospatial information in Africa is notably limited. However, with AI technologies and universally available geospatial datasets, it is anticipated that there will be a significant increase in granular data on the African continent. This is critically needed for development, not just in South Africa but across Africa. To realize this, expertise in using AI technologies within the geospatial industry must be developed to facilitate these development. Partnerships between government and businesses will be crucial to develop these geospatial datasets that are needed for targeted interventions.

For example, AfricaScope has recently modelled and mapped the income distribution for Nigeria and Kenya. This type of data is generally not publicly available because it is too complex to model using traditional methods. However, AI technologies and machine learning algorithms have enabled AfricaScope to model this data at a granular level. This data is critically needed for infrastructure planning, economic development, supporting enterprises, allocating resources to schools and health facilities, locating retail facilities and government services as well as contributing to the development of smart cities.

The future of geospatial intelligence will not only be powered by data, but AI technologies will drive the development of geospatial data itself. This data will increasingly be used by AI to learn more about what is happening in South Africa and across the entire African continent.

For a more indepth perspective on AI in the geospatial industry view our YouTube video on Mapping the future – AI in the geospatial industry