Overview

This case study looks at how industrial areas in the Delhi National Capital Territory (NCT) affect the land around them. Using Sentinel-2 satellite imagery, I mapped three things close to industrial clusters: green cover, surface water and moisture, and built-up surfaces. The short answer from the work is that vegetation tends to thin out and built-up area tends to rise as you move closer to industrial zones, and these patterns show up clearly in the spectral indices once you separate the land into distance bands around each site.

I carried out this study as a researcher in 2025, working with open Sentinel-2 data over Delhi NCT, India. The aim was practical: build a repeatable, low-cost workflow that planners, students, and environmental groups can run for any industrial area in India without paid imagery.

Quick facts

• Study area: Delhi NCT, India
• Sensor: Sentinel-2 (10 m and 20 m bands)
• Indices: NDVI (vegetation), NDWI (water and moisture), NDBI (built-up)
• Spatial method: Multi-ring buffer zones plus zonal statistics
• Software: QGIS, Google Earth Engine, Python, ArcGIS Pro

Approach

The core idea was simple. If industrial activity changes the surrounding environment, then the change should vary with distance. Land right next to a factory cluster should behave differently from land a few kilometres away. So instead of treating the whole NCT as one block, I treated distance from industry as the main variable.

To do that, I combined two methods that work well together:

• Spectral indices to describe what the surface is made of (green, water, or built-up).
• Buffer zones around industrial locations to slice the surrounding land into distance bands.

Then I used zonal statistics to summarise each index inside each band. This turns a continuous satellite image into a clear table: for every distance ring, what is the average vegetation, water, and built-up signal. That table is where the story becomes readable.

I kept the workflow open source first. Sentinel-2 is free, QGIS is free, and Google Earth Engine handles the heavy cloud-side processing. This matters for India, where many local bodies and university teams cannot rely on commercial imagery budgets.

Methods

Data and preprocessing

I worked with Sentinel-2 surface reflectance imagery over Delhi NCT, choosing scenes with low cloud cover so the indices would not be corrupted by haze. Standard steps applied here:

• Selecting clear-sky scenes and masking clouds.
• Clipping the imagery to the Delhi NCT boundary.
• Stacking the bands needed for each index (red, near-infrared, green, and shortwave infrared).

Spectral indices

Three normalised indices carried most of the analysis:

• NDVI (Normalized Difference Vegetation Index): uses near-infrared and red bands to measure healthy green vegetation. Higher values mean denser, healthier plant cover.
• NDWI (Normalized Difference Water Index): uses green and near-infrared bands to highlight water bodies and surface moisture.
• NDBI (Normalized Difference Built-up Index): uses shortwave infrared and near-infrared bands to pick out concrete, rooftops, and other built surfaces.

Reading the three together is important. Built-up land and vegetation usually move in opposite directions, so NDBI rising while NDVI falls is a strong signal of land being converted from green to grey.

Buffer analysis

Around the industrial locations I generated multi-ring buffer zones, a set of concentric distance bands moving outward from each cluster. These rings are the spatial frame for everything that follows. Each ring represents a distance class, for example near, intermediate, and far from industry.

Zonal statistics

For each index raster I ran zonal statistics against the buffer rings. This produced average index values per ring per index. With NDVI, NDWI, and NDBI summarised this way, I could compare the inner rings against the outer rings directly and see whether the environmental signal changes with distance from industrial land.

I ran the spatial processing across QGIS and Google Earth Engine, used Python for tidying and charting the zonal outputs, and used ArcGIS Pro for cartographic checks and final map layouts.

Outcome

The buffer-and-zonal approach made the relationship between industry and its surroundings easy to read. Across the distance rings, the general pattern was consistent with what you would expect from intense industrial land use:

• Vegetation (NDVI) tended to be lower in the rings closest to industrial clusters and stronger further out.
• Built-up signal (NDBI) tended to be higher near industry, matching denser construction and sealed surfaces.
• Water and moisture (NDWI) added context on where surface water and moist ground sat relative to the industrial footprint.

Taken together, the indices point to greener, less built-up conditions as you move away from industrial zones, which is the kind of gradient planners care about when they think about buffers, green belts, and where to protect remaining open land.

Just as useful as the maps is the method itself. The workflow is:

• Repeatable. The same buffer plus zonal statistics chain can be pointed at any industrial area in India.
• Low cost. It runs entirely on free Sentinel-2 imagery and open tools.
• Scalable. Google Earth Engine lets the same logic run over larger regions without local hardware limits.

For Delhi NCT specifically, this gives a fast screening tool. Before commissioning expensive ground surveys, a team can use this remote sensing baseline to spot which industrial clusters sit next to the sharpest drops in green cover, and prioritise those areas for closer study. That is the value I wanted from the project: turning open satellite data into a clear, distance-based picture of industrial pressure on the land around Delhi.