UKZN PhD Graduate Pioneers AI Analysis of Nighttime Lights for South African Cities

Dr. Zandile Mncube, a trailblazing researcher from the University of KwaZulu-Natal (UKZN), has made significant strides in urban studies through her doctoral research. Graduating this week with a PhD in geography, she becomes the first in her family to achieve postgraduate success. Hailing from Mnambithi in Ladysmith, northern KwaZulu-Natal, Mncube's journey began in 2016 with limited resources but unwavering family support. Her thesis, titled A Geo-Temporal Analysis and Forecasting of Nighttime Light Intensities over Three Largest Municipalities of South Africa, leverages artificial intelligence to decode patterns in satellite-captured nighttime lights, offering fresh perspectives on urban dynamics in Cape Town, Durban, and Johannesburg.

This interdisciplinary work fuses geography, Geographic Information Systems (GIS), remote sensing, and data science. By examining how artificial lights reflect human activity—from households and streetlights to commercial hubs—Mncube provides tools for policymakers to monitor growth, economic vitality, and environmental shifts in real time. Her achievement underscores UKZN's role in fostering innovative research addressing South Africa's pressing urban challenges.

Understanding Nighttime Lights Data: A Proxy for Urban Activity

Nighttime lights data, captured by satellites like the Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB), measures radiance emitted from Earth's surface after dark. This low-light imaging sensor detects lights from cities, towns, and infrastructure, serving as a reliable indicator of human presence and economic function. Unlike traditional census data released decennially, VIIRS provides monthly observations, enabling granular tracking of urban expansion.

In South Africa, where rapid urbanization sees metros like Johannesburg housing over 5.5 million residents, Cape Town around 4.8 million, and Durban 3.9 million (Stats SA 2022), such data reveals settlement sprawl, business proliferation, and even urban heat islands—areas where concrete traps heat, exacerbating temperatures. Mncube's study highlights how brighter radiance correlates with higher activity, contrasting starkly with dimmer rural zones.

Advanced Methodology: Hybrid Deep Learning Meets Satellite Imagery

Mncube employed hybrid deep learning models on VIIRS data from 2014 to 2023, accessed via Google Earth Engine—a cloud platform for planetary-scale analysis. She compared baseline Long Short-Term Memory (LSTM) networks with enhanced versions: Wavelet Denoise-LSTM (WD-LSTM), Empirical Mode Decomposition-LSTM (EMD-LSTM), and Ensemble Empirical Mode Decomposition-LSTM (EEMD-LSTM).

These models decompose non-stationary time series into intrinsic mode functions (IMFs), denoising signals to capture long- and short-term dependencies. Performance was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), with hybrid models outperforming baselines in interpretability and accuracy for forecasting urban trends. For instance, EEMD-LSTM excelled in Cape Town and Johannesburg, while EMD-LSTM led in Durban.

Mann-Kendall tests confirmed trends, validating the approach for South African contexts. Detailed methodology appears in her publications, including Evaluating hybrid deep learning models in Frontiers in Remote Sensing.

City-Specific Trends: Durban Shines While Others Dim

Analysis revealed divergent paths. Durban exhibited a strong upward NTL trajectory (z-score 11.748), signaling robust urban expansion and economic buzz, possibly from port activities and tourism recovery post-COVID.

• Cape Town: Significant decline (z-score -5.464), linked to energy constraints despite tourism-driven growth.
• Johannesburg: Sharpest drop (z-score -9.252), reflecting industrial slowdowns and infrastructure strains.

These patterns, visualized through radiance trajectories, highlight uneven development across metros, informing targeted interventions. Mncube's work in Evolving Earth details comparative projections.

Load Shedding's Shadow: Disconnect Between Lights and Economy

South Africa's power crisis profoundly distorts NTL data. During Stage 6 load shedding peaks (2022-2023), satellite radiance plummeted, even as GDP rose 0.6% in Q4 2023 (Stats SA). Mncube noted: “Nighttime light was going down and then GDP was going up,” underscoring blackouts' masking effect on true activity.

Her models adjust for these anomalies, revealing ground realities like persistent commerce via generators. This insight is crucial for Eskom planning and urban resilience, as load shedding cost the economy R900 billion since 2008 (Eskom reports).

Photo by PJ Gal-Szabo on Unsplash

Projections to 2027: Forecasting Urban Futures

Extrapolating trends, Mncube forecasts continued divergence: Durban's lights may brighten further, while Cape Town and Johannesburg stabilize or dip without interventions. These predictions aid in anticipating heat islands—urban areas 2-5°C warmer—and sprawl pressuring water resources (e.g., Cape Town's Day Zero legacy).

Step-by-step: data preprocessing denoises via EEMD; LSTM forecasts IMFs; reconstruction yields 2027 radiance maps for planning.

Implications for Sustainable Urban Planning in South Africa

Mncube's tools enable monthly monitoring, bridging census gaps. For Johannesburg (Gauteng GDP hub), dimming signals efficiency needs; Durban's growth demands infrastructure scaling; Cape Town's trends highlight renewable pushes like wind farms.

• Resource allocation: Prioritize lighting upgrades in high-activity zones.
• Climate action: Track heat islands for green corridors.
• Equity: Rural-urban light disparities reveal development gaps.

Integrated with SA's National Development Plan 2030, this supports smart cities. See coverage in Independent on Saturday.

UKZN's Role in Cutting-Edge geospatial Research

UKZN's Discipline of Geography leads in remote sensing, with Mncube's work exemplifying AI integration. The university's Centre for Geospatial Research supports such theses, producing alumni tackling local issues like eThekwini Municipality's urban planning.

Similar efforts at Wits and UCT use NTL for economics (Codera report mismatches in density vs. lights).

Challenges: Data Limitations and Power Crises

Despite advances, blooming (overglow), gas flares, and load shedding bias data. Mncube's denoising mitigates, but ground validation via drones/GPS is needed. SA's 340+ days of shedding (2023) demands hybrid models accounting for outages.

Mncube's Inspirational Journey and Future Outlook

From humble beginnings, Mncube's perseverance—backed by parents and late grandmother—inspires first-generation students. “Nighttime light data... for timely decision-making,” she emphasizes.

Future: Expand to all metros, integrate climate models. SA universities like UKZN position her for policy roles, advancing AI-urban research amid 2% annual urbanization (UN Habitat).

Photo by Ömer Haktan Bulut on Unsplash

Broader Impacts: Shaping Policy and Academia

This study bolsters evidence-based planning, aligning with SDG 11 (Sustainable Cities). UKZN's output enhances SA's research footprint, with NTL informing R1 trillion infrastructure spend (MTBPS 2026).

For students: Pursue interdisciplinary PhDs; tools like Google Earth Engine democratize analysis.