Climate AI Nordics

Climate AI Nordics

Nordic network for collaborations around tackling climate change using AI and machine learning.

Featured project: DEPRIMAP - AI-driven Mapping of Urban Deprivation and Climate Vulnerability at Global Scale

Project actors: Sai Ganesh Veeravalli, Stefanos Georganos, et al.

Project website: https://sola.kau.se/deprimap/

1. Context: Urban Deprivation as a Climate Risk Multiplier

Deprived Urban Areas (DUAs) are neighborhoods where overcrowding, inadequate infrastructure, and limited access to services undermine daily life. While slums are one form of DUA, deprivation also occurs in other urban settings such as poorly serviced housing estates and rapidly expanding peri-urban areas. These communities are often located in high-risk zones and lack protective infrastructure, making them disproportionately exposed to climate hazards such as floods and heatwaves.

Despite their scale, projected to exceed 2 billion people by 2030, DUAs remain poorly represented in global datasets and urban planning systems. Many censuses are outdated or exclude informal settlements / slums, and existing global population products capture only a fraction of these populations. This data gap limits our ability to accurately assess climate risk and design targeted adaptation strategies.

2. What DEPRIMAP Does

DEPRIMAP is structured as a three-phase framework to systematically analyze urban deprivation and its climate implications.

Phase I – Mapping: Identify deprived urban areas using geospatial indicators derived from satellite data and other spatial data sources, capturing patterns of building density, layout, and infrastructure.

Phase II – Population: Estimate how many people live in these areas by integrating mapped extents with bottom-up spatial population modelling approaches.

Phase III – Vulnerability: Assess how exposed these populations are to climate hazards such as floods and heatwaves by combining deprivation maps with spatial hazard data.

3. Technical Approach

DEPRIMAP combines Earth Observation data with machine learning to analyze urban environments at neighborhood scale. The framework integrates globally available datasets on buildings, roads, land cover, and derives spatial indicators that capture urban form and infrastructure conditions. These features are used within supervised learning models to identify patterns associated with urban deprivation. Designed for scalability, the approach can be applied consistently across thousands of cities, enabling comparable assessments in data-scarce regions.

4. Why This Matters for Climate AI

Climate risk is not evenly distributed within cities. Populations living in deprived urban areas often face a combination of high exposure to hazards and limited adaptive capacity due to inadequate infrastructure, insecure housing, and restricted access to services. Yet most climate risk assessments and urban datasets treat cities as homogeneous units, overlooking these intra-urban inequalities.

DEPRIMAP addresses this gap by providing spatially explicit information on where vulnerable populations are located and how they intersect with environmental hazards. This enables more targeted climate adaptation planning, improves risk modelling, and supports monitoring of global development frameworks such as SDG 11.1.1. By making hidden and underserved populations visible in data, the framework contributes to more equitable and effective climate decision-making.

5. Example Outcomes of Project

(A) Understanding Informal Settlement Transformation through Google’s 2.5D Dataset and Street View based Validation: Using Google’s Open Buildings 2.5D temporal dataset (2016-2023), this study analyses how informal settlements evolve over time. A rule-based classification on 100-meter grid cells identifies four change types: vertical densification, horizontal densification, combined growth, and decline. Validation using Google Earth Pro and Google Street View achieved 96.7% accuracy. Results show that horizontal densification dominates within informal settlements, while vertical growth is more common in surrounding formal areas, highlighting distinct development patterns within cities.

Example pairs of earlier and recent Google Street View imagery used for validation across different slum locations.

(B) Towards a Spatial Measure of SDG 11.1.1: Open Data for Urban Deprivation Mapping: This work introduces a scalable, open-data-driven framework for mapping urban deprivation using six indicators related to building morphology, vegetation, and infrastructure. Applied first in Nairobi and then across seven global cities, the framework demonstrates strong spatial alignment with known deprived areas. The analysis highlights globally robust indicators, such as building size and tree presence, while also revealing context-specific limitations, underscoring both the potential and challenges of globally transferable deprivation mapping.

You May Also Enjoy

Featured member, March 2026: Armi Tiihonen

Published:

Armi Tiihonen is an Assistant Professor at Chalmers University of Technology, in the Department of Mechanical Engineering. Her research focuses on the development of materials for more robust and stable solar cells, using approaches accelerated by machine learning and laboratory automation.

Machine learning based classification of tree crops of Syrian Arab Republic

Event date:

Webinar with Purnendu Sardar, Lund University. Accurate mapping of tree crops is vital for regional resource management, ecosystem service assessment, and the support of local livelihoods within the Syrian Arab Republic. Despite their socio-economic importance, tree crops are frequently misclassified or omitted in global and regional cropland products due to their complex spectral signatures and structural similarities to natural vegetation. This study proposes an integrated machine learning framework that combines the computational power of Google Earth Engine (GEE) with Python to enhance classification precision of tree crops across Syria. The methodology evaluates the integration of Global Ecosystem Dynamics Investigation (GEDI) LiDAR data with Sentinel-2 multi-spectral imagery to facilitate robust tree crop mapping. By utilizing GEE for the large-scale preprocessing of Sentinel-2 time-series data, the workflow generates high-dimensional, machine-learning-ready datasets that incorporate both structural and phenological variables. A Convolutional Neural Network (CNN) is subsequently trained in Python, chosen for its proficiency in processing time-series remote sensing data where temporal spectral patterns are more diagnostic than spatial textures. This approach allows the model to capture the distinct phenological cycles of various tree species, overcoming the limitations of traditional pixel-based or purely spatial classifiers. The findings underscore the efficacy of the CNN in distinguishing tree crop cover with high efficiency, demonstrating that the fusion of LiDAR-derived structural metrics with multi-temporal satellite data significantly reduces classification errors. The resulting high-resolution tree crop map provides an essential tool for sustainable agricultural planning and resource allocation in Syria.

AI in the wild: How Neural Networks Help Us Understand Our World Through Sound

Event date:

Webinar with Jeppe Rasmussen, University of Copenhagen. Bioacoustics, the study of nature’s sounds, has long been a powerful tool for studying wildlife. With the rise of artificial intelligence, particularly deep learning, the potential of this field has expanded dramatically. By applying advanced AI algorithms to bioacoustic data, researchers can now identify and monitor species with greater accuracy, even in environments where visual observation is difficult, such as dense forests or deep oceans. This capability is especially critical as we face the sixth mass extinction. AI-enhanced monitoring offers new hope for conservation by providing deeper insights into the presence, behavior, and well-being of endangered species. Beyond detection, AI also opens doors to understanding animal communication and emotional states, thanks to its ability to autonomously identify and prioritize key acoustic features. In this talk, I will present a series of case studies spanning multiple species and ecosystems to illustrate how cutting‑edge AI research can meaningfully advance our understanding of the living, complex world around us—and how these methods can help mitigate the biodiversity crisis we face and discover the surprisingly rich inner life of the animals surrounding us.

Bridging data gaps in the Global South: Harnessing AI and Earth observation in the humanitarian sector

Event date:

Webinar with Isabelle Tingzon, RISE. This talk highlights how AI and Earth Observation (EO) are transforming data availability for two global challenges: climate resilience and universal school connectivity. In the Caribbean, where extreme climatic hazards like hurricanes, floods, and landslides can devastate over 90% of built infrastructure across Small Island Developing States, comprehensive and up-to-date housing stock information is critical to enable resilient housing operations but is often incomplete, inaccessible, or completely non-existent. The “Digital Earth for Resilient Caribbean” initiative addresses these gaps by integrating EO data such as drone, LiDAR, satellite, and street-view imagery with AI workflows to classify roof types, materials, and building conditions. These efforts aim not only to enhance disaster response and recovery but also strengthen long-term resilience through government capacity building and knowledge exchange. In parallel, the presentation showcases UNICEF Giga’s mission to connect every school to the Internet by 2030. Reliable school location data is critical for planning telecommunications infrastructure, yet often inaccurate or incomplete in many low- and middle-income countries. By leveraging ML and satellite imagery, we’ve developed scalable pipelines to map schools across Africa on a nationwide scale, helping to ensure that no child is left behind in accessing digital learning opportunities. Together, these projects demonstrate how Geospatial AI can help bridge critical data gaps in the Global South, supporting both climate adaptation and inclusive education.