Urbanalysis | Devpost

The problem we are solving
AI generated suggestion
Web app implementation showing air pollution hotspot results
Data Pipeline

I. Inspirations

The Urbanalysis project was inspired by growing evidence that rapid urbanization is amplifying heat stress, air pollution, and the loss of accessible green spaces, particularly in cities of the Global South. Contemporary reports and case studies from cities such as Dhaka, Madrid, and Pune highlight how vulnerable communities disproportionately experience these environmental burdens. Despite the availability of rich satellite data from NASA and ESA missions, there remains a significant gap between data availability and its effective use in urban planning practice. Urbanalysis was conceived to bridge this gap by transforming complex Earth-observation data into actionable, equitable, and policy-relevant urban insights.

Rapid urbanization is exacerbating environmental stressors in cities, with implications for public health and sustainability. One prominent issue is the urban heat island (UHI) effect, wherein dense built-up areas trap heat and experience higher temperatures than their rural surroundings. This phenomenon is driven by the loss of vegetation and the prevalence of impervious surfaces that absorb and re-emit solar radiation.

Recent advances in satellite remote sensing enable detailed UHI monitoring. Thermal sensors on platforms like MODIS and the ECOSTRESS mission provide land surface temperature (LST) data that reveal fine scale hot spots and diurnal heating dynamics in urban agglomerations.

Another critical challenge is urban air quality. Cities often suffer from elevated levels of nitrogen dioxide (NO₂) and fine particulate matter (PM₂.₅). The Sentinel-5P satellite’s TROPOMI instrument allows high-resolution mapping of NO₂ and other trace gases over metropolitan areas.

Urban green spaces mitigate heat, improve air quality, and provide recreational and health benefits. However, access to parks and green amenities is often distributed unequally across socio-economic groups.

For this we present Urbanalysis, a smart web application that automatically downloads and processes satellite data, identifies environmental hotspots using density-based clustering and normalised indices, and summarizes these findings via a large language model (LLM) to deliver site-specific recommendations. Our system fuses MODIS and ECOSTRESS land surface temperature products with Sentinel-2 vegetation imagery and Sentinel-5P air-pollution measurements, uses density-based spatial clustering (DBSCAN) to delineate hot zones and converts the results into human-readable planning guidance.

An intuitive interface allows planners to select an area of interest, run analyses on demand and explore the results in interactive maps. Demonstrations on districts in Dhaka show that Urbanalysis can expose heat hotspots, air pollution exceedances and green-space gaps, while the AI assistant recommends targeted interventions such as cool roofs, micro parks or traffic diversion. The platform bridges the gap between satellite intelligence and actionable urban-planning decisions.

II. Methods

A. Data Sources

Urbanalysis integrates multi-mission Earth Observation (EO) and ancillary datasets to capture thermal stress, air pollution, and green-space accessibility at city scale.

Dataset	Description
Landsat	Historical reference, NDVI fallback
SMAP	Soil moisture
MODIS	Land surface temperature
ECOSTRESS	High-resolution thermal
Sentinel-2	NDVI vegetation
Sentinel-5P	Air quality
JRC GSW	Surface water
SoilGrids	Soil properties
OpenStreetMap	Roads and land use

B. Data Processing and Analysis Pipeline

AOI selection and ingestion
Harmonization and quality control
Feature engineering and normalization
Density-based clustering (DBSCAN)

Figure 1: sample line graph using colors which contrast well both on screen and on a black-and-white hardcopy

III. Results and Discussion

A. Air Quality Hotspots

Sentinel-5P NO₂ data are analyzed to identify severe and elevated pollution zones.

Figure 2: Air Quality Hotspots

B. Urban Heat Island Clusters

Normalized LST from MODIS and ECOSTRESS identifies the hottest 10% of pixels.

Figure 3: Urban Heat Islands

C. Green-Space Access Gaps

NDVI and network service-area analysis reveal underserved neighborhoods.

Figure 4: Green-Space Access Gaps

D. AI Guidelines and Recommendations

A large language model translates analytical outputs into planning guidance.

Figure 5: AI Recommendations

E. Web Implementation

Urbanalysis provides a user-friendly web interface.

Figure 6: Web Interface

F. Financial Overview

Item	Cost (USD/month)
Data acquisition	0
Infrastructure	100
LLM inference	150
Marketing	50

IV. Case Study

Study Area: Narayanganj, Bangladesh

Challenges We Ran Into

Several challenges were encountered during the development of Urbanalysis. Such as:

Data-quality limitations, including cloud contamination in optical imagery and the coarse spatial resolution of some air-quality products, required careful preprocessing and multi-date compositing.
Harmonizing datasets with different spatial resolutions, temporal frequencies, and uncertainty characteristics posed additional technical complexity.
Algorithmic challenges included selecting robust normalization thresholds and DBSCAN clustering parameters that generalize across diverse urban contexts.
Infrastructural constraints related to large raster processing and considerations around user trust and adoption of AI-generated recommendations required deliberate design choices.

V. What We Learned

Through the development of Urbanalysis, we learned that technical sophistication must be balanced with interpretability and usability to achieve real-world impact.

We have learnt how to process big data and how to manage algorithms when the computational resources become crutial in the case of heavy data processing.

Integrating multi-source satellite data highlighted the importance of transparent preprocessing and normalization choices.

We also gained insight into the strengths and limitations of density-based spatial clustering for urban hotspot detection and the careful role that large language models can play in translating quantitative analytics into decision-support narratives. Overall, the project reinforced the value of interdisciplinary approaches that combine remote sensing, data science, and urban planning expertise.

VI. Conclusion

Urbanalysis demonstrates how automated satellite data pipelines, density-based clustering, and AI-powered summarization can transform Earth observation data into actionable urban planning insights. The platform enables evidence-based decision-making to support greener, cooler, and healthier cities.