Overview

Geospatial data science and artificial intelligence holds the promise to transform how UNICEF approaches environmental and climate hazards estimation and resilience in areas  emergency preparedness and response, humanitarian planning and programming in areas affecting children.From improvements in geospatial data workflows and data platforms to advances in artificial intelligence models, 2025 may be the year where pilots and proof-of-concepts from the past several years become common production-ready approaches. 

Geospatial data infrastructure modernization and workflow enhancements over the last several years has led to advances in how geospatial data formats are ingested, processed and integrated into downstream analytical activities. This is critically important as data is getting bigger, faster and more complex and user needs also increasing in complexity. Spatial data scientists are now able to move away from bespoke, proprietary GIS tooling towards open, collaborative and cloud-native environments (for instance, see the guide from IMPACT, Development Seed and Cloud-Native Geospatial Foundation). This year we should see opportunities for increased scalability, versatility in data handling and interoperability across geodata platforms. 

Additionally, advances in the low-code tooling and AI integration in coding platforms, presents large progress in the accessibility and capacity of geospatial data scientists and analysts to perform complex workflow tasks. This opens the possibility for analysis to be performed by people with wider skillsets with more analytical and interaction capabilities for end users.

As these innovations take shape, the following are emerging trends shaping the future of geospatial AI. From expanding access to geospatial data at scale to leveraging AI-powered insights and embracing open science principles, these developments are driving transformative change across the field.

Expanding Access to Geospatial Data at Scale

There is an increasing trend towards making geospatial data more accessible and at large scale (for instance, open source catalogs such as SpatioTemporal Asset Catalogs, eoAPI, OPeNDAP). Recent collaborations between NASA and Microsoft toward the development of NASA Earth Copilot holds the promise for how people interact with Earth data by integrating conversational AI with NASA’s open source data analysis platform VEDA. This could make geospatial data search, discoverability and analysis easier. This has real-world implications as it can facilitate the retrieval and analysis of historical data to improve predictive models and, in turn, anticipatory disaster readiness and response.

AI-Powered Insights from Geospatial Foundation Models

Following the release of the HLS Geospatial Foundation Model (HLS Geospatial FM) through a public/private partnership between NASA and IBM Research in 2023, geospatial foundation models drawing on large inventories of space agencies’ satellite imagery have increased their capacity to produce insights on Earth’s processes. Specific AI/ML applications have been developed for classification, object detection, time-series segmentation and similarity search, which can be useful for flood and drought monitoring and mapping, land use and land cover mapping, natural disaster damage assessments and fire management and recovery. Geospatial foundation models using earth observation data developed by non-profits have started flourishing, such as those from Clay (backed by Radiant Earth) and Earth Genome’s Earth Index. 

The Open Science Revolution in Geospatial AI

Through NASA’s Open-Source Science Initiative (OSSI), a commitment to build the open science community over the next decade, advances in model development mean that current and future models will be able to train on smaller datasets, which saves on cost and time to production. It also reduces efforts to build downstream applications using AI models to perform tasks such as monitoring natural disasters and anticipatory damages of child-centric infrastructure.

Addressing Challenges Ahead

While these advancements hold great promise and continue to evolve, they also present challenges that must be addressed to ensure vulnerable populations can fully leverage these technologies in ways that maximize benefit and minimize risks.

The Infrastructure Gap: Who Can Build Geospatial AI? 
While progress in developing such geospatial AI models will continue, there are challenges and risks. The infrastructural requirements for geospatial AI models remains high - only well-resourced public science agencies and private sector companies using supercomputers can develop such models. For example, the IBM WatsonX Foundation model stack is running on NASA’s Science Managed Cloud Environment. 

The Cost of Scaling Geospatial Data
Another challenge pertaining to geospatial data applications is the need to standup and maintain cloud data warehouses. Geospatial data is characterized by frequent data updates and constant need for spatial relationship-based joins and operations. These necessitate warehouses to efficiently store and query data at scale by separating compute resources from storage to enable fast retrieval and parallel processing. There are drawbacks including variation in spatial functionalities being offered across cloud services, inconsistency with Open Geospatial Consortium standards, and scaling costs without control. The setup of data warehouses with cost control methods in place requires thoughtful architects that understand the analytical user needs.

Risks of Inaccurate AI Predictions in Critical Contexts 
There are risks that geospatial AI models get deployed without sufficient evaluation in diverse environments resulting in incorrect predictions. Poor performance can mean flawed monitoring and incorrect mapping of spatial areas. If emergency response partners depend on such models to inform early warning alert systems, then wrong model outputs could mean unnecessary alerts are sent creating alert fatigue or confusing populations or, even worse, no alerts are sent out and people bear the consequences that they might otherwise not have had to. In both cases, wrong model outputs can result in populations losing trust in emergency response system and these underlying geospatial AI models.

Skewed Access in Geospatial AI Development 
Another risk is that these AI initiatives, whether it be Earth Copilot or the foundation models, are developed by large technology companies with certain license to access requirements limiting use to certain users. A commitment to accessibility, open data and open science principles will be important to the impact it can have.

Workforce Readiness
Finally, a more practical risk is that the workforce broadly speaking is not yet ready to meaningfully take up and use geospatial technologies. While some geospatial data scientists and analysts have the training and expertise, they remain a small minority within the organization. Moreover, current geospatial analyst staff tend not to be sufficiently equipped to evaluate the effectiveness and use of geospatial data technologies. Investments in upskilling and educating analysts should be made to realize the potential of these geospatial data technologies.

Amid these challenges, the continued evolution of geospatial AI offers a path toward greater access and more effective solutions for those who need it most. By addressing these issues thoughtfully and implementing the necessary safeguards, we can ensure that these technologies not only reach vulnerable populations but do so in ways that are equitable, sustainable, and impactful.

How Geospatial Data and AI Can Enhance UNICEF’s Capabilities to Deliver Impact

Geospatial data infrastructure modernization and workflow enhancements can have transformative effects on UNICEF programming. Moreover, geospatial AI applications can enable the realization of use cases that were not possible very recently.  Some examples are:

Anticipatory action
Geospatial data and AI applications can be integrated into early warning systems to detect signs of natural disasters, such as the immanence of floods and other extreme weather events, and other catastrophes.  Geospatial data workflows can be used to overlay predictive model outputs with child-centric infrastructure and population-dense areas. This can inform emergency preparedness and response planning protecting vulnerable youth populations.

Humanitarian emergency response 
Geospatial data and AI applications can provide real-time insights needed to support emergency response once disaster strikes. Having accessible and interactive data at the disposal of humanitarian actors at the onset of disaster events is critical to prioritize aid delivery and mitigate harm to children. Conversational AI applications can allow such actors to quickly identify and prioritize responses as disasters unfold and changes quickly unfold.

Climate hazards estimation
Estimation of climate hazard risks often requires analysts and scientists integrating geospatial data layers. This work has been traditionally challenging due to complexities of the size, formats and latency of the data from geospatial data sources. Modernized data workflows can enable analysts to develop pipelines and integrate AI components to better quantify uncertainties related to climate hazards and their impacts on children. Moreover, conversational AI application can allow specialists to quickly understand climate events and their impact on children vulnerability measures over spatial dimensions.