BORDER OPTICS : Satellite Verification of India's Vibrant Villages Programme: A Compositing-Window Robustness Assessment of Border-Village Development

SAKSHI D. MASKE 

Ask the Government of India if things have worked out for them with their flagship scheme for the Border Villages, the honest answer (on parliamentary record) is No, it has never been verified anywhere. The Vibrant Villages Programme (VVP-I) has approved around ₹4,800 crore in support of 2,967 villages in five Himalayan border states/union territories including 662 villages marked 'priority' for Phase-I of the programme since 2022–23. There is no impact assessment in any manner. This study utilizes two satellite proxies: Sentinel-2 built-up-index (NDBI) and VIIRS night-lights at 258 individually geocoded villages across three core states (Arunachal Pradesh, Sikkim, Uttarakhand), an illustrative sample of 7 villages in Himachal Pradesh, and compares them with a matched non-VVP control group of 753 villages across the same 14 districts. To determine if the answer depends on the way the set of 2021-2025 satellite data is synthesized, everything was calculated twice, with two defensible methods of compositing the same satellite record. It does. A full year composite doesn't show any significant difference in built-up area (Wilcoxon signed ranked, p = 1.000), and the same villages, but shifted to one year in the June-September window shows a difference (Wilcoxon signed ranked, p < 0.000001); the data is contaminated by snow and the summer monsoon cloud wipes out Sikkim's data. At that point, before any robustness testing is done, that is the headline. What holds up, then, is narrower than either raw number on its own. The summer-window signal holds (did coefficient = +0.0377, cluster-robust p = 0.0021), but the full-year gap does not (p = 0.215) indicating evidence for an effect specific to VVP-I villages and not the region. Sweep the extraction buffer from 500m down to 250m or out to 1km, and the result is stable for all distances at which samples were extracted (all p < 0.002). Extend the gap to 3 years (2021, 2023, 2025) and a new complication comes to light: 2021 is reported as the gap compared to 2025, but that's really a story of recovery between 2023 and 2025; you'll see it recover nicely from 2023 to 2025, and the reported gap isn't a steady growth. Night-lights, tested the same way the rest of the series were ran, do not show a meaningful trend through either the full year (p = 0.050) or just the summer months (p = 0.9999), making it the more believable of the two proxies, simply because it doesn't change with the seasons. Even the nature of the budget tells its own story, as the amount of sanction in Arunachal Pradesh is roughly ten times higher than Uttarakhand (2,749.74 cr INR vs 270.58 cr INR) but the change is almost similar (measured, +0.0284 vs +0.0293). No significant effect is produced for built-up change from border proximity (H3), though a much more unstable effect is produced for night-lights, strict to the full-year time period (ρ = 0.043–0.037, both not significant). None of this leads to a clear verdict — and that is not the verdict that VVP-I would like to report — although the one signal that does make it past every test thrown at it, and that is the beauty of the business: a control group, three years of data, and three buffer radii, that one signal clears.