Does Deep Learning Enhance the Estimation for Spatially Explicit Built Environment Stocks through Nighttime Light...

Area of Science:

• Environmental Science
• Urban Planning
• Geospatial Analysis

Background:

• Built environment stocks are crucial for understanding material and energy flows and environmental impacts.
• Accurate estimation of building stocks supports urban mining and resource circularity strategies.
• Nighttime light (NTL) data offers high-resolution insights but suffers from blooming/saturation effects impacting building stock estimation.

Purpose of the Study:

• To propose and train a Convolutional Neural Network (CNN)-based building stock estimation (CBuiSE) model.
• To apply the CBuiSE model to estimate building stocks in major Japanese metropolitan areas using NTL data.
• To evaluate the model's ability to mitigate NTL blooming effects and improve spatial accuracy.

Main Methods:

• Development and training of a custom CNN model (CBuiSE).
• Application of the CBuiSE model using Nighttime Light (NTL) data.
• Spatial analysis and validation of building stock estimations in Japanese metropolitan areas.

Main Results:

• The CBuiSE model successfully estimated building stocks at a high resolution (approx. 830 m).
• The model effectively captured spatial distribution patterns of building stocks.
• The CBuiSE model mitigated overestimation issues caused by NTL blooming effects.

Conclusions:

• The CBuiSE model demonstrates the potential of NTL data for refined building stock estimation.
• This approach offers a new direction for anthropogenic stock studies in sustainability and industrial ecology.
• Further improvements in accuracy are needed to enhance overall model performance.