Multi-temporal gridded panel dataset of NDVI, NDBI, MNDWI, and Land Surface Temperature for the Dhaka Metropolitan Region (1990–2025)

Description

This dataset provides a multi-temporal gridded panel of Landsat-derived spectral indices and Land Surface Temperature (LST) for the Dhaka Metropolitan Region, Bangladesh, covering five decadal epochs (1990, 2000, 2010, 2020, 2025). It supports analysis of urban densification and its thermal consequences in one of the world's most rapidly urbanising megacities. Spatial coverage: Detailed Area Plan (DAP) jurisdiction administered by Rajdhani Unnayan Kartripakkha (RAJUK), approximately 39 km × 60 km. Resolution: 6,427 grid cells of 500 m × 500 m (~250,000 m² each), with mean values computed via zonal statistics. CRS: WGS 84 / UTM Zone 46N (EPSG: 32646). Variables per cell per epoch: NDVI — Normalised Difference Vegetation Index NDBI — Normalised Difference Built-up Index MNDWI — Modified Normalised Difference Water Index LST — Land Surface Temperature (°C) LST_relative — year-demeaned LST (controls for interannual climatic variability) Centroid coordinates (Easting, Northing) Source imagery: Landsat 5 TM (1990), Landsat 7 ETM+ (2000, 2010), Landsat 8 OLI/TIRS (2020, 2025), Collection 2 Level-2 Surface Reflectance with LaSRC atmospheric correction. Annual median composites were generated in Google Earth Engine with a 20% scene-level cloud filter; reprojection and zonal statistics were performed in QGIS. Intended use: Spatial econometric and machine learning analyses of vegetation–built-up–temperature relationships, urban heat island characterisation, cooling threshold estimation, and heat-mitigation feasibility assessment. The data are structured as a balanced panel (6,427 cells × 5 epochs = 32,135 observations). Notable patterns: Mean LST peaked at 31.1 °C in 2025; a 2020 anomaly (25.5 °C) reflects reduced anthropogenic heat emissions during COVID-19 restrictions. Mean NDBI increased from −0.166 (1990) to −0.134 (2025). Mean NDVI declined from 0.392 (1990) to 0.358 (2010) before partial recovery to 0.436 (2025), reflecting urban-core vegetation loss alongside peri-urban greening. Companion publication: Accompanies the manuscript "Density-Dependent Limits of Vegetation Cooling under Urban Densification: A Spatial Econometric and Machine Learning Analysis of Dhaka" (under review, Sustainable Cities and Society). Quality control: Cells with excessive cloud contamination or insufficient valid pixels in any epoch were removed to ensure a balanced panel. Limitations: Annual median composites represent yearly structural averages rather than seasonal extremes; LST represents radiative surface temperature, not near-surface air temperature.

Steps to reproduce

1. Acquire Landsat imagery (Google Earth Engine) Use Collection 2 Level-2 Surface Reflectance: LANDSAT/LT05/C02/T1_L2 (1990), LANDSAT/LE07/C02/T1_L2 (2000, 2010), LANDSAT/LC08/C02/T1_L2 (2020, 2025). Filter by AOI (DAP/RAJUK boundary) and target year. Apply CLOUD_COVER < 20%. Generate annual median composites with ee.ImageCollection.median(). 2. Compute spectral indices (GEE) NDVI = (NIR − Red) / (NIR + Red) NDBI = (SWIR − NIR) / (SWIR + NIR) MNDWI = (Green − SWIR) / (Green + SWIR) Bands: Landsat 5/7 use SR_B2/B3/B4/B5 (Green/Red/NIR/SWIR); Landsat 8 uses SR_B3/B4/B5/B6. 3. Extract LST (GEE) Use ST_B6 (Landsat 5/7) or ST_B10 (Landsat 8). Convert scaled DN to Kelvin: LST_K = DN × 0.00341802 + 149.0; then to Celsius. Export all rasters at 30 m in WGS 84 (EPSG:4326). 4. Reproject and grid (QGIS) Reproject to WGS 84 / UTM Zone 46N (EPSG:32646). Generate a 500 m × 500 m fishnet covering the DAP boundary using Vector → Research Tools → Create Grid. 5. Zonal statistics (QGIS) Compute mean, SD, min, max of NDVI, NDBI, MNDWI, and LST per grid cell. Means are used as primary variables. 6. Water mask and distance to water (QGIS) Threshold MNDWI > 0; apply a ≥5 connected-pixel filter to remove noise. Compute Euclidean distance to water using Proximity (Raster Distance). 7. Relative LST (Python) LST_relative(i,t) = LST(i,t) − mean(LST_t). This removes interannual climatic shifts (including the 2020 COVID-19 anomaly) while preserving within-year spatial structure. 8. Quality control Drop cells with excessive cloud contamination or insufficient valid pixels in any epoch. Final balanced panel: 6,427 cells × 5 epochs = 32,135 observations. 9. Clustering (optional, Python) Standardise NDVI, NDBI, MNDWI, LST_relative for 2025; apply K-means (k = 3, random_state = 42) to derive three urban morphological clusters. Software: Google Earth Engine, QGIS 3.28+, Python 3.10 (pandas, numpy, scikit-learn, libpysal, esda, spreg).

Institutions

• Jahangirnagar University

  Dhaka Division, Dhaka

Categories

Licence