Understanding the basics of AI within the geospatial domain (Part 1- Sensors)

As a GIS strategy consultant, I’ve know that any robust technology strategy must account for both current and emerging advances in artificial intelligence (AI). I am talking with many organisations about what they need to do to adapt to advances in AI and how it relates to geospatial technology. However, many people that I work with are unsure about AI and it's implications, and this is often not helped by the current marketing hype around this topic. It seems nowadays that anything and everything remotely related to technology is magically "POWERED BY AI !".

To aid with this situation I have written a series of articles providing a non-technical explanation of some of the basics of AI and how these concepts relate to the domain of geospatial technology.

There are three overlapping and increasingly interwoven key strands of what is happening with AI:

1. Automatically recognising objects or occurrences from sensors.
2. Advanced processing techniques, including attempts to mimic human thought processes.
3. Enhanced human to machine (and vice versa) communications, including Natural Language Processing (NLP) and Large Language Models (LLMs) such as ChatGPT.

It is also important that organisations understand the potential benefits, risks and ethical considerations around the use of AI.

Below Part 1 of this series of blog posts focusses on the first of these concepts.

Expanding and automatically recognising objects or occurrences from a huge variety of sensors

The 'Internet of Things' (IoT) is in some ways a distinct concept from AI, however the huge variety and numbers of sensors that are being deployed is one of the major drivers of the rise of AI. The vast amounts of data being generated from the Internet of Things is too large to be monitored or analysed by humans, which has contributed to many new approaches to automatic analysis of those data sources, in particular approaches that isolate and identify specific elements of the data that are the most important. Those important elements are then summarised to be passed on to humans, or those elements can be used to trigger automated actions through some form of technology (for example an obstacle identified in the planned path of a robot will trigger a change in the robots movement).

There is a lot of debate about what qualifies as "AI" and indeed many elements of the technology discussed in this article has been around for a long time in rudimentary forms and been called other names in the past. When reading through the following sections it is important to note that it is not the sensors or data collection concepts themselves that exhibit characteristics of AI, instead it is the fact that the sensory data being collected is increasingly being automatically recognised, classified, processed, analysed, and used to trigger automated responses (which then does qualify as a type of AI). For example a camera taking a picture is clearly not AI, but if that picture is then automatically analysed to determine what it contains, and then decisions or actions are taken as a result of that determination, then that overall process does qualify as AI.

To understand the breadth and capabilities of these sensors I think it is useful to relate them to the five human senses.

Electromagnetic Spectrum Sensors and Computer Vision

Computer vision is a field within artificial intelligence (AI) that enables computers to "see" and understand images and videos, much like humans do—but through using mathematical models and algorithms. It is used in many everyday technologies, such as facial recognition to unlock your phone, car licence plate recognition in car-parks, and security systems that detect movement. Computer vision is perhaps the most useful 'sensory' capability of AI with respect to GIS because mapping is predominantly a visual medium.

Computer vision works by teaching machines to recognise patterns in visual data (such as images or video). For example, when you show a Computer Vision application thousands of pictures of cats, it learns what features—like ears, whiskers, and eyes—are common across those images. It also applies measurements between those features to be able to distinguish things like the typical distance between an average cat's eyes and it's nose, to help distinguish a picture of a cat from a picture of a dog.

Once trained, these systems can perform tasks like identifying objects in a photo, detecting movement in a video, or reading handwritten text. They do this by breaking down images into pixels (tiny dots within a grid), analysing the patterns and shapes, and comparing them to what they’ve learned. Computer Vision also involves techniques like edge detection and image segmentation (dividing an image into parts), object detection (finding and labelling things in an image), and classification (deciding what category something belongs to). These techniques help machines make sense of complex visual scenes and respond intelligently.

Computer Vision can be significantly enhanced by incorporating data from sensors that capture non-visible bands of the electromagnetic spectrum, and from combining those datasets with LiDAR data (Light Detection and Ranging). Standard cameras operate within the visible spectrum—the range of light that human eyes can see. However, many important features in the environment are better revealed through other parts of the spectrum, such as infrared (IR), ultraviolet (UV), and thermal bands. Infrared imaging, for instance, is useful for detecting heat signatures.

In the context of GIS, these non-visible electromagnetic bands are often applied to agriculture mapping: this can help monitor plant health by identifying water stress or disease before it becomes visible to the naked eye. Plants with those conditions will show as a different 'colour' in a derived image to the surrounding plants.

Other examples include surveillance, where non-visible electromagnetic bands can detect people or animals in low-light conditions. Thermal imaging, a type of infrared, shows temperature differences and is used in building inspections to find heat leaks or in search and rescue operations to locate people in darkness or smoke. Ultraviolet imaging can highlight materials or substances that fluoresce under UV light, which is useful in forensic analysis or detecting surface contamination. All of these data elements are locatable and therefore are ideal for recording and analysing with a GIS. These techniques can be applied to geographic features, such as combining various forms of electromagnetic spectrum data from satellites of aircraft with LiDAR from aircraft or drones to provide detailed 3D data of buildings. These learning processes are usually powered by neural networks, which are systems designed to mimic how the human brain processes information.

LiDAR uses laser pulses to measure distances to objects, creating highly accurate 3D maps of environments. It is especially valuable in applications such as autonomous vehicles, where it helps detect obstacles, measure road surfaces, and understand the shape of surroundings in real time. In forestry and environmental monitoring, by comparing the LiDAR returns from the ground to returns from the trees LiDAR can measure tree height, canopy density, and terrain elevation with great precision. It is also used in urban planning and archaeology to reveal structures hidden under vegetation or to map complex urban landscapes. When combined with traditional image data, LiDAR adds depth and spatial context to Computer Vision. For example, a system can not only recognise a pedestrian in an image but also estimate their exact distance and movement path. This fusion of data from visible or non-visible bands and LiDAR with standard imagery creates multi-modal vision systems. These systems are more robust, especially in challenging conditions like fog, darkness, or cluttered environments. They enable smarter decision-making in fields such as autonomous navigation, robotics, precision agriculture, disaster response, and smart cities.

The integration of Computer Vision into GIS has evolved significantly over time. Initially, during the 1960s to 1980s, GIS and image processing developed as related disciplines, with analysis interpreting aerial and satellite imagery using basic techniques like pixel classification and edge detection. These early methods were limited but laid the foundation for future automation. In the 1990s, digital remote sensing became more widespread, and GIS began to adopt more advanced image analysis tools. Automated classification methods allowed for more efficient mapping of land use and environmental features, although these techniques were still relatively simple compared to modern standards.

The 2000s introduced to GIS workflows the concept of machine learning, which is essentially a branch of artificial intelligence where computers learn patterns from data to make predictions or decisions without being explicitly programmed for each task, through encountering new data or new processes and adding that new data or processes to the system without human intervention. During this period algorithms such as decision trees and clustering improved classification accuracy, and object-based image analysis (OBIA) allowed for more context-aware interpretation of spatial data. From the 2010s onward, deep learning revolutionised Computer Vision in GIS. Convolutional Neural Networks (CNNs) enabled highly accurate object detection and image segmentation, supporting applications like urban mapping, disaster response, and precision agriculture.

Today, Computer Vision in GIS is characterised by multi-modal analysis, combining imagery with LiDAR and spectral data. The focus is shifting toward real-time automated processing. For example: consider a situation where there is a fire in a large and complex building such as an airport terminal. A common problem with these types of buildings is that a small issue, such as a small fire in a trash receptacle can cause a large scale evacuation and disruption. Many large buildings such as this have a Digital Twin utilising Indoor GIS capabilities, and many of those Digital Twins record the location and status of smoke detectors that are being monitored in real-time. If a fire triggers a smoke detector then the location of the detector will be immediately highlighted within the GIS. It is feasible that the GIS using AI will then trigger cameras or heat sensing sensors in that local area of the smoke alarm to begin automatically isolating the precise location of the fire. With agentic AI it is also increasingly common that an automatic evacuation process will be activated, with people being directed by signs and automated speaker announcements which paths to follow to avoid the risk zone. Distant zones within the large terminal will be on standby, not necessarily evacuated immediately. Fire responders will also be automatically notified, not just that there is a fire, but precisely the location of the fire, and the best route to travel to reach that location.

Audio Sensors

Audio sensors play a role in artificial intelligence by serving as the ears of intelligent systems. These sensors capture sound waves from the environment—such as speech, music, ambient noise, or mechanical vibrations—and convert them into digital signals that AI models can process and interpret. This capability enables a wide range of applications across industries.

In consumer technology, audio sensors are central to voice-activated assistants like Siri, Alexa, or Google Assistant. They detect spoken commands, and AI models use natural language processing (NLP) to understand and respond to user requests. In healthcare, audio sensors combined with AI can monitor breathing patterns or detect signs of distress in patients. In industrial settings, AI systems analyse audio data from machinery to identify anomalies, helping predict failures before they occur—a technique known as acoustic predictive maintenance.

Examples from the domain of GIS include:

• Environmental monitoring, where audio sensors can detect wildlife sounds, enabling AI combined with GIS to track species presence or behaviour without visual contact.
• Security systems, where AI can analyse audio for signs of intrusion, aggression, or emergencies, even in low-visibility conditions, and areas of interest can be presented via GIS.
• Noise regulated areas: for example some airports or concert venues are bound by regulations on how much noise is permitted late at night. It is also possible to record the diffusion of sound across different directions to build up a 3D picture of where sound-waves are travelling.

Vibration, Pressure or Temperature Sensors

AI can "feel" through sensors that detect pressure, texture, or vibration—technologies that simulate a sense of touch, often referred to as tactile sensing. These sensors allow machines, such as robots or drones, to gather physical data from their environment and respond intelligently, much like how humans use their skin to sense contact and texture.

Pressure sensors measure the force applied to a surface, helping robots determine how firmly they are gripping an object. This is crucial in tasks like picking up fragile items or performing delicate assembly work. Texture sensors, often based on arrays of tiny sensors or specialised materials, help AI systems distinguish between surfaces—like rough versus smooth—enabling more nuanced interaction with objects. Vibration sensors detect subtle movements or feedback, which can be used to identify material properties or detect changes in the environment.

In robotics, these tactile sensors are used to enhance dexterity and safety. For example, a robotic hand equipped with pressure and texture sensors can adjust its grip in real time to avoid dropping or damaging an item.

Examples from the domain of GIS include:

• Road maintenance assessment: where tactile sensing can detect of surface texture and irregularities, such as cracks, potholes, and wear, by physically measuring the road’s roughness and smoothness. This is similar to how a person might run their hand over a surface to detect bumps or gaps. These sensors can be mounted on vehicles or robots to scan roads in real time. Beyond surface texture, tactile sensors can also assess the material properties of the road, such as hardness and elasticity. This helps identify signs of degradation, like softening or brittleness, which are early indicators of structural issues. These measurements provide insights that visual inspections alone might miss. Robotic systems equipped with tactile sensors can perform automated inspections, especially in areas that are difficult or dangerous for humans to access. Tactile sensing is often used to complement other technologies like cameras and LiDAR by adding depth and material feedback, making road assessments more accurate and comprehensive.
• Seismology: where AI is playing an increasingly important role in seismology by enhancing how scientists detect, analyze, and respond to earthquakes. One of its key contributions is in real-time earthquake detection and early warning. AI models can quickly analyze seismic signals and distinguish between actual earthquakes and background noise, allowing for faster and more accurate alerts to the right locations. AI is particularly good at identifying patterns—such as foreshocks, aftershocks, or subtle signals that might precede larger seismic events.

Gas or Airborne Particle Analysers

AI is being increasingly integrated with gas and airborne particle analysers to enhance how we monitor and respond to environmental and industrial conditions. One of its key roles is in real-time data processing, where AI can quickly analyze streams of sensor data to detect unusual patterns or dangerous spikes in pollutants. This allows for faster alerts and responses to potential hazards like gas leaks or air quality deterioration.

Examples that are used in conjunction with GIS include:

• Air quality monitoring across a city: AI can combine with GIS to analyse vast amounts of spatial and temporal data collected from networks of air quality sensors and mobile monitoring units. Machine learning algorithms can identify pollution hotspots, detect patterns in pollutant dispersion, and predict future air quality levels based on variables like traffic flow, industrial activity, and weather conditions. These predictions help city planners and environmental agencies take proactive measures, such as adjusting traffic routes or issuing health advisories. GIS provides the spatial framework to visualize and map this data, showing where pollution is concentrated and how it moves across different neighborhoods. AI can also help classify pollution sources—distinguishing between vehicle emissions, industrial outputs, or natural contributors like dust or pollen—by analyzing sensor data in context with GIS layers such as land use, road networks, and wind patterns. Together, AI and GIS support real-time monitoring and dynamic response strategies. For example, AI can trigger alerts when pollution exceeds safe thresholds in specific areas, and GIS can guide response teams or inform the public through location-based notifications. This combination also aids long-term planning by simulating the impact of proposed infrastructure changes or environmental policies on air quality.

Liquid or solid analysers

AI is being used alongside liquid and solid analysers to collect and interpret complex chemical and physical data, which is then spatially represented within Geographic Information Systems (GIS). These analysers measure properties such as contamination levels, nutrient content, or material composition in water, soil, or industrial samples. AI processes this data to identify patterns, detect anomalies, and classify substances with high accuracy. When integrated with GIS, the AI-enhanced data is mapped across geographic locations, allowing researchers and decision-makers to visualize environmental conditions, track pollution sources, and monitor changes over time.

• Precision agriculture: AI is being used in conjunction with liquid and solid analysers to support agriculture by collecting detailed information about soil composition, moisture levels, nutrient availability, and water quality. AI processes this data to detect patterns, classify soil types, and predict crop performance or potential issues like contamination or nutrient deficiencies. When integrated into GIS platforms, the AI-enhanced data is mapped across fields and regions, allowing farmers and agronomists to visualize spatial variations and make informed decisions about irrigation, fertilization, and planting strategies. This leads to more precise, sustainable, and efficient agricultural practices.
• Water quality management: These analysers measure parameters such as pH, turbidity, nutrient levels, and contaminant concentrations in water bodies and sediments. AI processes this data to detect trends, identify pollution sources, and predict changes in water quality over time. When integrated with GIS, the AI-enhanced data is located precisely across catchments, reservoirs, and urban water networks, enabling smarter, more sustainable water resource management across cities and regions.

Keep an eye out for Part 2 of this series where we will discuss advanced data processing techniques used in AI.

Terminology and Definitions

Classification (in computer vision)

Classification is when AI looks at an image and decides what it is. For example, it might look at a photo and say, “This is a tree” or “This is a car.” It’s like teaching a computer to recognize and label things in pictures.

Clustering (in machine learning classification)

Clustering is when AI groups things that are similar without being told what they are. For example, it might group photos of animals into clusters—one for birds, one for dogs—based on patterns it finds, even if it doesn’t know the names of the animals.

Computer Vision

Computer vision is a field of AI that helps computers “see” and understand images and videos. It’s used in things like facial recognition, self-driving cars, and analyzing satellite images.

Convolutional Neural Networks (CNNs)

CNNs are a type of AI model that’s especially good at looking at images. They work by scanning small parts of an image, finding patterns like edges or shapes, and using those to understand what the image shows.

Decision Trees

A decision tree is a simple way for AI to make decisions. It asks a series of yes/no questions to figure something out—like a flowchart. For example, “Is the object round?” → “Is it red?” → “It’s likely an apple.” Decision trees are popular because they are easy to understand and interpret. They can be used for both classification (e.g., identifying categories like types of plants) and regression (e.g., predicting values like temperature or price). In practice, AI systems often use many decision trees together in a method called random forests, which improves accuracy by averaging the results of multiple trees.

Digital Twin (in the context of geospatial data)

A digital twin is a virtual copy of a real-world place or system. In geospatial data, it might be a digital model of a city that updates in real time using sensor data, helping planners monitor traffic, pollution, or infrastructure.

Edge Detection (in computer vision)

Edge detection is when AI finds the outlines of objects in an image. It helps the computer understand where one object ends and another begins—like tracing the edges of buildings in a satellite photo.

Electromagnetic Spectrum (in computer vision)

The electromagnetic spectrum includes all types of light, not just what we can see. AI uses data from parts of the spectrum like infrared or ultraviolet to analyze things like heat patterns or vegetation health in images.

Image Segmentation (in computer vision)

Image segmentation is when AI divides an image into parts, like separating roads, buildings, and trees in a satellite photo. It helps computers understand what each part of the image represents.

Internet of Things (IoT)

IoT refers to everyday devices—like sensors, cameras, or thermostats—that are connected to the internet and share data. AI uses this data to monitor environments, automate systems, and make smart decisions.

Large Language Models (LLMs)

LLMs are AI systems trained to understand and generate human language. They can answer questions, write text, translate languages, and more.

LiDAR

LiDAR is a technology that uses laser light to measure distances. It creates detailed 3D maps of surfaces, like terrain or buildings. AI uses LiDAR data to analyze landscapes, detect changes, or guide autonomous vehicles.

Machine Learning (ML)

Machine learning is a way for computers to learn from data. Instead of being programmed with exact instructions, they find patterns and improve their performance over time—like learning to recognize different types of trees from photos.

Multi-modal Analysis (in computer vision)

Multi-modal analysis means AI looks at different types of data together—like combining images, sound, and text—to get a better understanding. For example, it might use both satellite images and sensor readings to assess air quality.

Natural Language Processing (NLP)

NLP is a branch of AI that helps computers understand and use human language. It’s used in chatbots, translation tools, and systems that read and summarize documents.

Neural Networks

Neural networks are AI models inspired by the human brain. They’re made of layers of “neurons” that process information and learn patterns, helping with tasks like image recognition or speech understanding.

Object Detection (in computer vision)

Object detection is when AI finds and labels specific things in an image—like identifying and marking all the cars in a traffic photo. It’s used in surveillance, mapping, and automation.

Object-Based Image Analysis (OBIA) (in spatial analysis)

OBIA is a method where AI looks at groups of pixels (objects) in an image instead of individual pixels. It helps in analyzing satellite images by identifying features like buildings, fields, or forests more accurately.

Pressure Sensors

Pressure sensors measure how much force is being applied to a surface. AI uses data from these sensors to monitor things like water flow, structural stress, or touch in robotics.

Robotics

Robotics involves machines that can move and perform tasks. AI gives robots the ability to sense their environment, make decisions, and act—like navigating a warehouse or inspecting infrastructure.

Sensors

Sensors are devices that collect data from the environment—like temperature, light, motion, or chemicals. AI uses sensor data to understand what’s happening and make smart decisions.

Tactile Sensing

Tactile sensing is when a device can “feel” physical contact, like pressure or texture. AI uses this in robotics to help machines handle objects gently or detect surface conditions.

Texture Sensors

Texture sensors detect the feel or pattern of a surface—like smoothness or roughness. AI uses this data to identify materials or assess wear and tear in surfaces like roads or machinery.

Vibration Sensors

Vibration sensors measure movement or shaking. AI uses them to detect problems in machines, monitor earthquakes, or assess structural health in buildings and bridges.

The views outlined in this article are solely the opinion of the author and do not necessarily represent the views of my employer nor any other party.