Publications

Liu, YX; Hu, CM; Zhan, WF; Sun, C; Murch, B; Ma, L (2018). Identifying industrial heat sources using time-series of the VIIRS Nightfire product with an object-oriented approach. REMOTE SENSING OF ENVIRONMENT, 204, 347-365.

Abstract
Carbon-based fuels burned at industrial facilities account for a large proportion of greenhouse gas emissions, and an up-to-date spatiotemporally detailed inventory is essential for a better understanding of global carbon emission patterns. The Visible Infrared Imaging Radiometer Suite (VIIRS) Nightfire product offers a quantitative estimation of the temperatures of sub-pixel heat sources, providing the potential for detecting thermal anomalies from industrial sectors across the globe. However, identifying subcategories of various industrial heat sources is challenging because there are scarcely any stable and typical characteristics for their classification at a single thermal anomaly scale. Specifically, these nighttime thermal anomalies exhibit a strong spatiotemporal heterogeneity (e.g., fluctuations in retrieved temperature, spatial shifts in position, and presence of false positives), even in industrial heat sources that do not vary through time. Here, we demonstrate an object-oriented approach to robustly segment and accurately classify various industrial heat sources from a time-series of the VIIRS Nightfire product. The approach operates from the cluster level of spatially adjacent nighttime thermal anomalies (i.e., nighttime-heat-source objects rather than individual thermal anomalies) to generate fingerprint like characteristics and to address the challenge of spatiotemporal heterogeneity. Specifically, the spatial-aggregation characteristic of nighttime thermal anomalies from continuously operating industrial heat sources and the temporal-aggregation characteristics of biomass burnings were incorporated to differentiate industrial nighttime-heat-source objects from ubiquitous biomass burnings. Subsequently, the similarity of the thermal signals of nighttime thermal anomalies from identical industrial heat sources was used to generate highly recognizable characteristics for their identification. A spatiotemporally detailed inventory of industrial heat sources across the globe was then established from this object-oriented classification. The inventory included a total of 15,199 industrial heat sources, representing 49.52% of all higher confidence nighttime thermal anomalies in the VIIRS Nightfire product. Validation of the results showed that only 218 objects (1.43%) were biomass burnings or active volcanoes that were misclassified as industrial heat sources. Further validation of sub-categories indicated an overall classification accuracy of 77%. Our findings suggest that the VIIRS Nightfire product has great potential for monitoring the global distribution and dynamics of industrial heat sources, and combined with the object-oriented approach developed here the methodology is simple, robust, and cost-effective.

DOI:
10.1016/j.rse.2017.10.019

ISSN:
0034-4257