Mapping Urban Agglomerations via Google Earth Engine

Diagram: pixel-based search for urban growth mapping, using nightlight data.

Google Earth Engine provides an easy access to NASA's global nightlight data in its developer console, which enables a wide variety of possible raster data treatment operations, including but not limited to pixel-level arithmetic, zonal statistical analysis and aggregation, as well as regression & classification. In this particular project, I'm interested to find the pixels that: 1) had undergone the most amount of light intensity change; 2) with an increasing trend of light intensity; which, if combined, would serve as a proxy indicator to measure the magnitude of urban growth. For example, below is a map and the code showing the 2013 night light data rendered in the East Asia region.

A linear regression model was run on the global pixels with the nightlight data from 1999 to 2013, trying to predict the future nightlight intensity. Following is a map showing the predicted results - bright green shows a robustly increasing trend of light intensity, while the blue shows relatively unchanged or declining trend.

A more zoomed-in view of the nightlight trend data in the City of Changsha, Hunan Province, China. Note that how the central core of the City is shown in blue, since the light intensity of the old town area largely remain consistent, as compared to the significant increase found in the surrounding newly urbanizaed areas.

Finally, a Reduce Neighborhood operation was used to find the large contiguous area of pixels that meet the conditions of consistent, significant light intensity levels and a positive trend. They are highlighted in blue on the map below, with red and pink colors for those that only have smaller connected areas.

As you can see. many metropolitan urban conurbations in East Asia were highlighted: Tokyo-Yokohama-shi, Seoul, Hong Kong - Macau - Guangzhou region, Beijing-Tianjin, and the west urban belt of Taiwan, etc, all of which had experienced significant expansion and urban development in the past several decades.

Some zoomed-in views of the identification result in other regions: