FusionLightSR Multimodal Data Extraction Pipeline: VIIRS NTL, WorldPop & ESA WorldCover Retrieval Scripts

For Bangalore, three complementary geospatial datasets were extracted using Google Earth Engine to support multimodal super‑resolution. First, monthly nighttime light (NTL) radiance was obtained from the VIIRS DNB collection, filtered for the year 2020 and clipped to a polygon defining the Bangalore metropolitan area. The median composite of all monthly images was converted to 32‑bit float to preserve radiometric precision, then exported at multiple spatial resolutions (30 m through 500 m) for analysis of scale effects. Second, population density data from the WorldPop 100 m product were averaged over the same time period and region, yielding a continuous surface of estimated inhabitants per grid cell. This layer was similarly exported at the set of target scales, providing an auxiliary modality that reflects human settlement patterns. Third, land‑cover information was drawn from the ESA WorldCover 2020 map, clipped to Bangalore, and exported at 10 m resolution; its categorical classes supply contextual information on urban, vegetation, and other land‑use types.

To apply the identical workflow in Delhi, one only need to redefine the study area polygon. In place of the Bangalore coordinates, the following polygon may be used to clip all three datasets for the Delhi region:

var delhiPolygon = ee.Geometry.Polygon([
[76.7, 28.2],
[76.7, 29.0],
[77.9, 29.0],
[77.9, 28.2],
[76.7, 28.2]
]);

With this single change, the VIIRS NTL, WorldPop population, and ESA WorldCover layers will be extracted and exported for Delhi in exactly the same manner as for Bangalore.