| His Coordinates | |
“The mantra for Geospatial AI must be only Trust, but verify”
says Dr. Mukund Kadursrinivas Rao in an interview with Coordinates magazine, as geospatial systems transition from mapping the world to autonomously reasoning about it. |
 |
|
What has AI fundamentally changed in how geospatial data is processed, analysed, and interpreted today?
Geospatial technology (or GIS or Geomatics) emerged from research in late 1960s in digital concepts (though in analogue form they have been existed as knowledge systems for much longer). As a result, there is a strong knowledge foundation for GIS. Over the decades, tt has triggered considerable advances in Geospatial education and research, business and societal development activities. On other hand AI, in its present “intelligence” form is emerging into mainstream society in recent years (though in its “earlier Avatars” can be traced through Digitalisation of data and coding, pattern recognition Software/Applications that do a specific job, and Expert Systems solutions that anybody could use). In a way, Geomatics through computerisation – and Geospatial data and AI, have been “fusing” together into a single stream. Now we are in a situation where AI seems to be “encompassing” all technology streams of digital activity, including Geospatial technology and applications. Geospatial activities are seemingly subsumed into “AI” – through spatial data content, locational knowledge and timeline-spatial records – as embedded pillars of “intelligence” of human reasoning.
In the yesteryears Geomatics or Geospatial project workflows, humans acted as the primary “intelli-translators” – manually digitizing maps or digitally analysing features from imagery; defining rigid, rule-based “integration” (as models) to understand spatial relationships; and then provide limited but different, perspective of various natural, social and economic phenomenon.
Today, AI (specifically GeoAI) has flipped that earlier human dynamics – we are moving away from manually/semi-automated Descriptive GIS (what is where?) towards fully Autonomous and Predictive GIS (why is it happening and what will happen next?). This is a also a major shift from the late 1990s fixed-model Deterministic GIS (static layers, manual queries, fixed algorithms) to Probabilistic Cognitive Engines with automated reasoning and predictive insights. AI is rapidly transforming Geospatial methods and processes from the earliest Display of Maps through Layer Overlays to a Query system to an Information System to a Decision Support System to a Reasoning system and now into an autonomous Geospatial AI that Detects, Assesses, and Decides autonomously.
In this AI advancement, there are two underpinnings of GIS fundamentals that are powering AI: first, the digital spatial data; and second, the location (xyz). Together, these two have provided the root to the advanced concepts (or data concepts) of “intelligence” which AI is taking great advantage of. Over the past five decades or so, the availability of digital spatial data with location in a data-bin has revolutionised “intelligence activity” in a significant manner. So, today just any data is characterised by location – be it images, maps, financial records, addresses, landuse, infrastructure, weather, flights, census data, social development, economic spatialisation – anything and everything that is digital on this Earth is now having a location tag – and that makes every bit of data on this Earth amenable to location analytics – brining newer and newer meaning and understanding of Earth and every human activity.
Traditional “silos” (Shapefiles, isolated Rasters) have undergone drastic changes and adapted to scale. The AI era relies on CloudNative and spherical tessellation into Discrete Global Grid Systems (DGGS) to normalize disparate sensors and spatial datasets. Thus, satellite pixels, IoT sensor readings, VGI, tabulated data, text reports and any bit of data on this Earth are hashed to a common index on spatial location. A “flood” tweet and a “soil moisture” vector layer and a “Radar image” now share a common mathematical key (of location), enabling real-time fusion of petabyte-scale datasets on a spatial frame. This brings tremendous advantage to AI models as the model not only undertakes location analytics but streams only relevant byte-ranges without downloading full files or tiles or yesteryears.
AI is now about “Live Maps” (unlike earlier time-static maps or displays) – fusing historical, current and streaming satellite/ drone images and geospatial layers (elevation, land use etc) with dynamic data streams (live traffic, hourly weather sensors data, and instant device GPS pings). AI acts as the realtime “dual glue” – one, “fixing” and “holding” the location (in xyz domain) of any data AND two, automatically reconciling different data formats and temporal scales and data volumes to provide real-time LIVE MAPS fusion and data spatialising.
There are now Foundation Models bespoke for every object (e.g., “find cars”; “what is that object” or “hotspots temperature”). AI now uses Geospatial Foundation Models (GeoFMs) pre-trained on Earth Science/Geodesy and DGGS as Vision Transformers (ViT) and Masked Autoencoders that reconstruct the Earth from sparse data, gaining a latent understanding of terrain, cities, rivers, clouds, rain, atmosphere, climate and even outer Space. Geo-SAM (Segment Anything Model) tools instantly convert unstructured pixels (Raster) into topological vectors (Polygons), automating the analysis/digitizing tasks in minutes.
Earlier, Overlay was the foundation of a GIS but now MultiModal Logic provides more deeper and truer insights by correlating orthogonal datasets (Optical + Radar + Text + Video + Voice+…) to discover hidden and unknown relationships and insights. Virtually, the world is now modelled in Geospatial AI as a nested graph, not a map – though it is ultimately visualised by humans as Maps in a Spatial Knowledge Graph (SKGs) or Graph Neural Networks (GNNs).
The final mile of delivery is Agentic AI (Agentic GIS) – GIS Apps and GIS Dashboards and GIS Portals are getting obsolete now. Instead of just displaying an analysed heatmap or a GUI menu or a fixed-Portal, the AI system uses Large Language Models (LLMs) to reason, fuse voluminous data, use pre-trained “intelligence” logic, eliminate improbabilities and provide a more deeper and new insight and action-sets as to why “that heat pattern” or “road alignment” is the best option with human reasoning and logic. The future is also of Semantic Alerting based on behavioural anomalies – predicting and alerting about events and phenomenon in advance.
Any human being now can “own” the Agentic AI and through a set of Generative Language Interface or Conversation AI, run LLMs of millions of parameters OR re-train the Agentic AI with advanced prompts and obtain the best of INSIGHTS – a ready solution for action. In fact, the Agentic AI can also embed testing and validation of results to score the “best” from a range of possibilities.
Agentic GIS will make hitherto complex GIS operations and command into a geo-conversation…just talk/question/ask/choose and obtain results or outcomes – in ready form for action!!!
Which areas of geomatics – remote sensing, GIS, photogrammetry, positioning and navigation, are being most transformed by AI right now, and why?
The wide range of Geospatial technology are feeling the “AI effect” and the transformation isn’t uniform. The impact is based on data streaming to automating tedious tasks to enabling entirely new ways of perceiving the physical world.
Images – from Satellites or Drones or by humans, are the “Scale” Leader and are transforming because we have so much more and more pixels than human eyes can perceive or an Analyst can ever process. The Geo Foundational Models are Vision-Foundation Models that understand Earth’s surface, satellite images, Drone captures, Device-camera images etc, not just on Earth but also Earth’s interiors, atmosphere and even outer space. These models are already pre-trained on massive global datasets, allowing them to detect, from any IMAGE, floods, fires, crop health, under-canopy activities, vehicles, aircrafts, troop movements, storms and any human and natural activity with minimal additional processing, training and human intervention. The advancement is so intense that IMAGE processing is moving from computers, ground servers to the satellites themselves OR to the user’s hand-held devices – Edge Processing. The processing can be anywhere for GeoFM data that is everywhere (if Data Centres move to Space then compute would happen there)!
High-resolution image analysis and Photogrammetry is shifting from measuring geometry to synthesizing reality and provide the Fidelity to AI GeoFM. Traditional photogrammetry relied on matching millions of “feature points” to create meshes. New AI techniques like Gaussian Splatting and Neural Radiance Fields or Auto Cleaning allow for the creation of photorealistic 3D environments from a handful of images. This is revolutionizing “Digital Twins” for cities or creating “real” infrastructure management remote and safe.
Positioning and Navigation is providing the Resilience of Location to AI models – ensuring location-certainty and precision. The problem of Multipath Mitigation in “urban canyons” (tall buildings) or even within Building-GIS, is fused into AI models that look at past data records and predict and cancel out these reflected signals in real-time from a fast and real-time back-andforth switch between orbiting GPS satellites and Ground PNT systems, allowing for sub-cm accuracy even in dense city centers or within buildings. Similarly, Dead Reckoning when GNSS signals are lost (e.g., in tunnels or due to jamming) is way gone as AI models interpret data from IMU sensors (accelerometers/gyros) or streaming time-signals much more accurately than traditional calculus-based filters, “bridging” the gap until the signal returns.
Finally, GIS is enabled as the supreme “Decision” Agentic Partner. Natural Language Mapping has broken the GUI barrier and Geo-GPTs have come to stay. So now, what is relevant is: “Identify all residential buildings within 500m of this proposed metro line and calculate the potential land acquisition costs and property tax loss.” Agentic GIS is making Predictive Simulation a reality – Agentic GIS is no longer just a “map” – in fact, it is now a LIVE model of the world!
Are we moving from traditional map-making towards real-time spatial intelligence? How central is AI to this shift?
Yes, the shift is absolute. We are moving away from cartography (a static picture of the past) towards spatial intelligence (a real-time nervous system for the planet).
Traditional map-making is retrospective or “What Was” but AI is making it a current or future perspective or “what is” or “what can be”. GIS is now to be only LIVE – moving away from Traditional Mapping of Collect data → Process in office → Validate → Publish map which used to take months to weeks; now Spatial Intelligence GeoFMs start from the IoT/ Satellite stream → Edge AI Processing → Autonomous Fusion → Agentic GIS → Options → Actions – all in few minutes.
Without AI, the sheer volume of real-time spatial data would be a humongous “madness on human analysts”. AI performs three critical roles – Pattern Recognition from the “LIVE” spatial GeoFM (that could be billions of pings from GPS, thousands of images from cameras, and live streams from various sensors and voluminous historical past data) applying a primary filter, identifying only the meaningful data and changes and ignoring the millions of “unneeded” data points; Predictive Modelling which is the real-time intelligence of analysing the present, looking at past and more so about anticipating the future event to occur; and Agentic Automation for autonomously monitoring spatial feeds and trigger alert actions.
Are we dealing with a scenario of abundance of geospatial data than we can meaningfully use? Does AI help or complicate this situation?
Globally, we are undeniably in an era of data hyper-abundance, while in India it is not yet so (discussed in later sections). The “data gap” in GIS has shifted from a lack of information to a lack of “meaningful processing” bandwidth – something of a “Data-to-Insight Lag.” While we can capture the entire Earth in high resolution every day, the majority of that data remains “dark” – stored in servers but never actually looked at or used for bettering human lives. AI is the primary force attempting to solve this, but its role is a double-edged sword: it both clears the backlog into meaningful data but creates a new layer of complexity by adding fresh data insights.
Without AI, the abundance of Geospatial data could be just “noise” but AI helps by acting as an automated triaging system of Dimensionality Reduction, to compress massive “data cubes” (layers of time, space, and spectrum) into compact summaries for an Agent; Feature Extraction at Scale, from a Foundation Model and Exception-Based Mapping, detecting when something changes—a new road appearing, a lake drying up, or a ship entering a restricted zone and so on.
But at the same time, I must also underscore that AI must not be treated as a “Black-Box” and requires sound principles and processes for the GeoFMS and Predictive models. Else there can be three complications – The Trust Gap (Explainability) because the Deep Learning model is hidden in millions of neural weights making it difficult for legal or justification for high-stake activities; Synthetic Data Loops because AI generates more spatial data (e.g., auto-completed maps), there is a risk of “model collapse” where future Agents are trained on data produced by past AI Agents, potentially amplifying small errors into massive “geographical hallucinations”; and Infrastructure Tax because managing the AI that manages the GeoFM requires its own massive infrastructure and calls for huge “data plumbing” efforts to ensure the AI pipelines don’t break and keep learning and running.
So while it looks that we have solved the “data” problems or “analysis” problems, but we might be creating a “decision fatigue” problem. AI can generate thousands of “insights” per unit time and humans can get overwhelmed by the sheer number of insights, alerts and predictions. Further, AI may also tend to tilt towards and be biased on what Decision is being looked for and may provide a complacency bias.
In fact, at present in my experience, it is prudent to say the quality of your Agentic GIS is a function of the quality/depth/ intensity of your Generative prompts – varying prompts on same Foundation Models give different outcomes and pose a huge Trust Gap! So a word of caution…it is the Generative prompts (Human Intelligence – so critical!) that are key for an Agentic GIS to be successfully meaningful – else the old GIGO!
So, is it a net Positive? That is what the world is saying and one can hope – the ability to monitor the planet and human activities in near-real-time allows us to respond to climate change, urban growth, reach logistics, undertake profitable business, and manage disasters in ways that were physically impossible even ten years ago. Involuntarily, we may be building a Planetary Geo AI System and maybe even a Space Geo AI System of Moon and all planets, galaxies put together!
I also wonder – earlier in 1970s we used to envisage a “GIS for Bengaluru” and later “Natural Resources Information System” and later “National GIS” and later Google Earth or Sentinel and so on…so our ability has been in expanding the data horizon (from a city to the state to the nation to the Earth)…today AI is pushing that horizon limit – both in depth of data and widening the spread to every inch of Earth, Moon planets, space…the Planetary Geo AI system!! But the AI data has been and even now is just one aspect…the ability to put to use all that AI outcomes to bettering human life, our knowledge, society, business, harmony, peace, environment, climate, managing strife…and working for a bettering life on Earth…we need to work more intensely and need many examples to establish that Human-AI interface enterprise!
How is AI reshaping professional roles in geomatics? What new skills should surveyors, GIS professionals, and geospatial analysts prioritise?
We are seeing a shift from technical production (making the map) to strategic curation (validating the intelligence). Traditional surveying is moving away from hours of field-based manual point survey to rapid Drone Survey and Field Data Streamers and Satellite constellation Images – becoming more realistic, legal sound and also advisory in nature. Surveyors and Analysts are now the ultimate authority of accuracy, verifying that Geo AI outputs meet rigorous legal and engineering standards, maintaining the cross-consistency in the GeoFM.
Geo AI Professionals are no more “Digitizers” but are “Agent Orchestrators” managing AI Agents that monitor data feeds and trigger analyses automatically. One has to be not only a “software user” but also a “system architect” and also a “logic expert” and have sectoral knowledge. Analysts must act as the bridge between raw AI predictions and real-world policy, interpreting complex patterns for non-technical decision-makers.
To stay competitive in the AI “value chain” would require knowledge of AI Literacy; GeoAI models; Prompt Engineering in GIS; Python/R for Automation; libraries like GeoPandas, TensorFlow, and PyTorch; Data Curation & Ethics; Spatial Reasoning & Logic; Data Storytelling; Critical Thinking & Validation; Explainable AI; Ambiguity Management; Legality of AI; Creative Problem Solving etc and many other knowledge integration.
The above knowledge has to be ‘twinned” with knowledge of Geodesy, Geography, Spatial Science, Mapping and GIS concepts. So what we need is a twin-skill of AI+Geospatial in one person – something hard to find, at least in India. We must work on this to build a cadre of AI+Geospatial experts in coming years.
In my view, at the core level – two important soft-skills are fundamental – thinking and imagining, both of which are basic and critical skills/capabilities for the AI era – where a professional must be able to think across any subject and find “model prompts” and imagine wide-ranging scenarios to buttress the “model prompts”. In my view, it is these that can make the prompts, LLMs and Agentic GIS so solid, robust, reliable and successful. Any person that can have these two core skills and gain above mentioned AI knowledge – I think they will make the best future Geo AI professionals!
In AI-generated spatial insights, how do we ensure trust, accuracy, and accountability in geospatial decision-making?
The mantra for Geospatial AI must be only “Trust, but Verify.” As we shift toward automated decision-making (say, where AI might trigger flood alerts or approve land-use permits or advance security actions etc), AI “magic box” must be a highly audited pipeline. There are four important structural pillars – Explainable AI (XAI) & Semantic Clarity to explain what and which data (pixels or spatial features or text data) influenced the AI model’s decision; Semantic Layering as a “Guardrail” that ensures the AI isn’t just matching patterns, but is following human-like logic (e.g., “A building would not exist in the middle of a permanent water body” – verify such outcomes!) and AI Model Cards, which are like a nutrition label, or Metadata documenting its Geo FM data, Training data, known biases, math, reasoning and logic and recommended use cases.
As synthetic imagery or information (AI-generated) becomes harder to distinguish from reality, the Open Geospatial Consortium (OGC) initiative to prioritize IPT (Integrity, Provenance, and Trust) is commendable – where the concept is that every data point carries a digital “passport” that records its source (which satellite/drone), what AI models touched it, and what edits were made and by whom. For high-stake decisions, audit trails are often stored on immutable ledgers to prove that spatial evidence hasn’t been tampered with in Blockchain Verifiable Logs.
Ultimately, accountability remains a human dependency and global standards that mandate professional oversight are getting defined. Exception-Based Reviews by AI flags “lowconfidence” areas where it is unsure, and a human expert (like a licensed surveyor or a Troop Analyst or an Environment Officer) performs a manual audit of only those specific cases. AI based Ground-Truthing Apps which are Real-time field apps (like Fulcrum or ArcGIS Field Maps) allow community groups or field crews to verify AI predictions on-site, feeding “ground truth” back into the model to correct errors instantly.
It is obvious that legal accountability in Geospatial AI is called for. The EU AI Act calls for mandatory third-party audits and risk assessments. Algorithmic Impact Assessments (AIA) being defined by many nations now require an AIA before deploying AI for public-facing decisions. The India AI Mission is also working details in these lines.
Quite a lot needs to be done in building the trust!
From an Indian perspective, what opportunities does AI-driven geomatics open up, and what structural challenges still need to be addressed?
India has made many attempts in past for nation-wide Geospatial data and applications; it must now develop a National “testbed” for AI-driven Geomatics.
The National Geospatial Policy 2022 claims to democratize data access and allow Indian startups to build “Bharattested” models that can be exported globally. A classic surge in private Map enterprise in India have become successful with private investments. Private EO satellites have been announced. AI is purported to analyze fragmented land holdings (typical of Indian farms) to provide hyper-local crop insurance and yield predictions in a business environment by some industries in India. Under the National Geospatial Mission, high-resolution (5-10cm) 3D “Digital Twins” for urban cities are aiming to be built and unique Operation Dronagiri project are on ground. With the IndiaAI Mission, India is envisaging indigenous foundation models like BharatGen for real-time monitoring of carbon sequestration and climate adaptation, helping India meet its ESG goals.
Recently, in her budget speech, Union Finance Minister Nirmala Sitharaman announced Bharat Vistar, a new AIdriven initiative for India’s agriculture sector. Elaborating further, she described Bharat Vistar – the Virtually Integrated System to access Agricultural Resources -as a multilingual AI platform that will integrate India’s agri-stack portals with the Indian Council of Agricultural Research (ICAR) package of agricultural practices, powered by AI systems.
However, in my view, real-time AI spatial intelligence in the Indian context is still staring at the scattered and many long-standing “potholes” of many years and also the wide gaps of deep-seated structural and organisational turfs in India. National alignment and mission approach has been earlier elusive – Geospatial AI will need this alignment to be a successful Indian homegrown initiative – else foreign AI technology and Geo FM data will power Indian Geospatial AI!
Despite the visible optimism of recent years, I still see some longstanding “bottlenecks” remain in the Indian Geospatial ecosystem:
• The Data Availability “Hurdle” – which has been a bane in India from 1990s – through NSDI, NNRMS, Bhuvan, NRDMS, National GIS, State-GIS etc. We still face a DATA drought! However, while recent policies have liberalized the business intent for GIS data, the Data-to-Insight Lag persists in various ways – especially in form of Patchy Data of both historical and current character. Geo AI models require clean, consistent historical and most current data to learn and improve pattern models and robust Agents. In many parts of India, land records are still being digitized, and historical satellite imagery for specific regions are unavailable or difficult or expensive to procure. EO Image Continuity from Indian satellites is even now a major problem – most users have to depend on foreign EO data sources – public-domain or commercial – which becomes a bottle-neck and also speaks on national capability! Seamless access to a Geospatial Cloud Archive of national level geospatial data is still not available – even a basic Spatial Foundation Dataset is yet to be adopted nationally – users create their own foundations and their own datasets which brings in immense non-standards and “silos”. Dronagiri project aims to create localised datasets – but will it be fitting into one National Geo FM is still un-demonstrated. Bhuvan has lots of GIS data and EO content – but access and ingest to Geo FM Models poses great difficulties. Statewide and Nation-wide compilations of standard GIS datasets are hitherto still unavailable. So, with this patchiness in data and multitude, localised data models – the base of a Geo Foundational Model and Agentic GIS across large regions within the country poses a great challenge. However, the spot-light can be on localised Geo FMs and LLMs for small project – just akin to and replicating the large number of GIS Project of yesteryears! To me, this is not the way to go – a massive effort at overcoming this Data “hurdle” is called for!
• Spatial Heterogeneity: India’s diverse geography (from desert to tropical rain-fed farms to coastal to glacier regions and so on) means a LLM trained in Rajasthan won’t work in Karnataka and even in another corner of Rajasthan itself. This “Domain Shift” requires massive amounts of local “Ground Truth” and “pre-training” knowledge for a robust LLM. This can be a challenging task as it labor-intensive and time/ resource intensive to collect and organsie. The model and LLM intensity has to be very high to cater to the wide ranging spatial heterogeneity of India’s natural, social, economic and security problems. Indian capability in AI Models and LLMs seem more oriented towards “million logics” at local scales (akin to earlier distributed architecture) – though I personally feel that is not the way to go now. We may have tens of thousands of these LLMS in operation in the country at lcalised spread. Like the “kirana stores”, the localised LLMs may appear best way to go locally – but managing the thousands of sectoral and localised LLMs (with millions of parameters within them) may not only be inelegant but we may never build a AI Model at scale for the nation and its business ecosystem. Such localised GeoFM and LLMs may also not be most relevant for the large variation of our states, natural resources, social fabric, economic inequity, security perspectives and will, once again in my view, turn out to be a paradox of “part but not the whole” and be a humungous management challenge! However, if we can build a marketplace of these thousands of Agentic GIS, maybe the business system may thrive to some degree – not just from domestic and localised market but may also be, at some level, exportable as customised AI models to many other “smaller” nations/regions of the world. Architecturing all of this will be a real innovation!
• Indigenous AI technology Stack – dependency outside of India for AI technology stack, GIS Engines, Highres satellite images, advanced Drones, Cloud Systems etc is also a hurdle that needs to be overcome. Until and unless India can create its own AI Technology Stack, the ability of Indian users to use and benefit from AI will always be dependent on outside resources – which is like “repeating” the same music again of the past!
I remember that in 1990s, the debate for building a GIS system was to go “bottoms-up” – build district systems and stitch them into state and national systems, as against “top-down” – building from nation expanding down to villages; India went the “bottoms-up” and demonstrated quite many district/city systems but never could organise a full National or Global system!! Google went “top-down” bit by bit and is now covering the whole Earth! In recent, MapMyIndia is another example that has gone top-down and has covered whole nation. Similarly, in mid 1990s, same debate happened for GIS Engines – indigenous GIS software efforts across the country BUT dependency on more robust foreign COTS GIS Engines fuelled many GIS projects – we lost the GIS Engine software capability!
In fact, when I use some of the AI tools – like Google Earth Engine, ArcGIS AI, Planetary Computer, Navi, Sagemaker EO, Carto, TorchGeo – I am amazed at the diversity of data models and capability of large instruction LLMs that distinguish in image analysis, climate analysis, Sustainability models, ML pipelines etc at scale, and in Conversation Mode, makes open data and custom data registries and geo-spatial processing and modelling so rich, easy and versatile. The world of Geospatial is certainly undergoing so much of advancements in the AI domain.
We must have an India AI Technology Stack that powers the Geospatial AI!!
• The AI Skill Gap: There is a massive need for professionals who understand both Geospatial and AI and deep learning, as explained earlier. Most Indian GIS professionals are trained in traditional software (ArcGIS/QGIS) but lack the MLOps and AI skills needed to deploy AI agents. Conversely, AI engineers often lack the “Spatial Reasoning” and GIS knowledge to understand map projections, coordinate systems, Laws of Geography, Terrain slopes, Landslides, urban taxation etc leading to geographically “hallucinated” results and dearth in Agentic GIS that are technically sound and robust and geographically adaptive. Just the traditional Geomatics skillset which emphasises on mapping and GIS is less of a requirement and relevance today in the Geospatial AI market. Our education and training systems must orient towards twinning “Geospatial+AI” skillset in the country.
The AI Skill Gap must get bridged on a campaign mode – either by creating a new cadre Geospatial AI over time and/ or by spotting and encouraging and empowering such existing Geospatial AI individuals within our system immediately.
• Security vs. Innovation which is an important case for India. As GeoAI becomes central to national defense and security infrastructure, at the same time central to development needs, the paradigm of constant tension between “Open Data” and “National Security” should get bridged and eliminated. Even Security would require the innovations of Geospatial AI – so an open and inclusive secure working model is called for. The Digital Personal Data Protection (DPDP) Act must add a layer of complexity to geospatial data that includes personal information (like land ownership), requiring organisations/ companies to build “Privacy-by-Design” into their AI models.
Looking ahead, do you see geomatics becoming an invisible but critical infrastructure for society, and what role will AI play in shaping that future?
That has always the goal even earlier – make EO and GIS embedded in society and governance – make it a part of governance workflow. It should also develop as a commercial enterprise (subscriptions – just look at how many people subscribe to ChatGPT, Gemini, Clause.AI, GEE, ArcGIS AI, CoPilot etc). So we should hope for that and work towards creating an invisible, ubiquitous fabric that governs the nation and world around us. Geospatial data will always be a critical data element in the AI business – society will only notice when Geospatial content fails. AI can be the catalyst that makes this paradigm shift for Geospatial technology both as a possibility and also invisibility. Concepts like Ambient Navigation; Predictive Cities; Massive Digital-Twins and Context-aware services (for insurance, tax, penalties, damage etc) and many many others – all as Agentic GIS will dictate the economy of national development and many business enterprise. I have no doubt on that – but the main issue I grapple is whether it will be built an Indian AI Technology Stack or on a foreign and global commercial AI stack!
Geospatial technology is moving from the “back office” of mapping to the “front line” AI for effective actions. The value is shifting from spatial data to spatial wisdom. Government agencies still have overlapping jurisdictions, and “Security” is still used, at times, as an excuse to block data. The mission mode is missing – so we make slow progress. Frigidity in Bureaucratic Inertia must not remain – not just at organisational levels but also at deep-down local levels. Local municipal and panchayat bodies still don’t know how to use the highres Geospatial data that the Geospatial Policy has technically “freed”. So how will they develop/manage the millions of localised Agentic GIS and develop the Million Parametric Models across the country?
However, I do see very positive developments and improvements over the past. At this stage, I see the net progress is in Policy Liberalization – removal of prior security clearances and permission for Indian entities to generate Geospatial data and applications; Private Pivot – companies announcing to launching private EO constellations and MapMyIndia’s nation-wide GIS maps and services; and Infrastructure Integration – emergency of shared geospatial backbone for digital transparency in development and government programmes, like PM Gati Shakti, MGNREGS, Pragati, Bengaluru Challenge 2026, Karnataka-GIS and many others. These are considerable steps ahead – we need to see the outcomes of these and many other initiatives that will set another action-roll ahead to Agentic GIS! The India AI Mission must align, integrate and mission-ise in the coming years.
India spent 200 odd years to make its first topographic maps and in last 50 years trying and trying to build the largearea GIS data content/applications. In 2026, it is essential to imagine and build the spatial intelligence using AI as a Unified Geospatial Interface – speeding Geo FM, private satellite EO images, private sector real-time and Live AI Maps of India, thousands of localised LLMs and Agentic GIS and the Business Ecosystem for a National Geospatial AI.
At a more global level, in next few decades, Geospatial AI will have transitioned from multiple Agentic GIS into the autonomous central nervous system of a hyper-connected Planet Earth or a “Cognitive Earth,” a state where a planetary-scale digital twin to monitor environmental shifts but pre-emptively negotiate global resource flows, stabilizes climate micro-systems through automated interventions, and manages “Human Development “ of scale. I am also pretty confident that AI will pervade into Planetary Science and Outer Space Eco-system that will be foundation for Moon, Mars, planets, outer space human endeavour – much needed!!
In such an era, India can emerge as the Global South’s Geo-AI Sovereign an contribute to an Outer Space.AI, leveraging its Geospatial AI foundation to leapfrog ahead.
(The thoughts and articulations are shaped from years of professional experience, complemented by curated generative intelligence, to present forwardlooking perspectives of Geospatial AI.)
(No Ratings Yet)
Loading...
Leave your response!