Incorporating the ECOSTRESS diurnal thermal infrared data in urban land cover classification

Understanding accurate urban land cover is essential for thermal environment research. Previous studies, however, have often overlooked the distinctive thermal features among urban built-up areas when conducting land cover classification.

 

To address this gap, this study proposes and evaluates a land cover classification scheme based on the thermal characteristic, thermal inertia. The study employs a multi-step methodology to quantify thermal properties, compare classification methods, and analyze challenges in the proposed approach.

(1) Land surface temperature and thermal inertia were quantified using ECOSTRESS (Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station) thermal infrared (TIR) data and Pleiades visible near-infrared (VNIR) data acquired during the summer and autumn of 2022 in Beijing;

(2) Conventional classification was conducted using VNIR data alone, while the proposed classification method incorporated both VNIR and TIR data. Their performance was evaluated using thermal inertia, surface albedo metrics, and confusion matrices;

(3) The challenges of the proposed classification were examined through statistical analysis by comparing the thermal features of pixels classified as wooden structures with those identified through visual interpretation;

(4) The underlying causes of classification challenges were explored and evaluated through thermal environment simulations using microscale urban canyon models.

 

The results indicate that while classifying wooden structures in urban built-up areas remains challenging, the proposed method was particularly effective in clustered wooden structure areas, where distinct thermal feature differences between wooden structures and other land covers were more pronounced. However, shaded wooden structures with high thermal inertia and low albedo were often misclassified as other impervious areas.

 

This study underscores the necessity of distinguishing land covers with unique thermal features in urban built-up areas and enhances the understanding of classification methods that integrate TIR and VNIR data.