The Cotton Line: Testing the Institutional Persistence of an Erased Colonial Land-Tenure Boundary Across Vidarbha's Agrarian and Vegetation Geography

SAKSHI D. MASKE 

Can a provincial boundary abolished in 1956 still be detected in today's farmland? In the nineteenth century Berar, in western Vidarbha, was settled under the Ryotwari land-revenue system. The Central Provinces districts to the east were settled under Malguzari. On district lineage the old line matches the present boundary between Amravati and Nagpur Divisions, which I rebuild as a 384.7 km line from 120 talukas.

I test two hypotheses at that line: a level step (H1) and a change of slope (H2). The tests use four present-day outcomes:
- a satellite cropping-intensity index (the fall in Landsat NDVI from November to March, measured in two years);
- land surface temperature (LST) on one June date;
- the 2011 ratio of agricultural labourers to cultivators;
- the 2011 share of net sown area.

The model is a spatial-discontinuity regression with rainfall and elevation controls. It is run on a 32-taluka corridor, on the talukas the Landsat scenes observe (98 for NDVI, 56 for LST), and on all 119 census talukas. NDVI is computed at 30 m with cloud, shadow and cropland masks. Inference uses the most conservative of heteroskedasticity-robust and Conley spatial standard errors, with Holm correction over 20 tests.

**Findings.**
- **No robust level step (H1).** No outcome shows a step at the line in the corridor, in the Landsat sample, or in local-linear estimates at the line. For NDVI this depends on how the Landsat paths are calibrated against each other: side and path almost coincide, and removing the estimated path offsets produces a step in year 2 (−0.052, p = 0.007) but not in year 1.
- **One exception, which is not local.** In the 119-taluka census model, labour structure differs across the line (+0.96 labourers per cultivator on the Berar side; Holm p = 0.029). The difference is absent within 60 km of the line and in local-linear estimates, so it is regional rather than a step at the boundary.
- **The NDVI pattern is a ridge across the line.** The cropping-intensity index is highest within 50 km of the line on both sides (0.29) and falls away towards Buldhana in the west (0.21) and the paddy districts in the east (0.13). The raw gap between the sides is small and unstable: 0.009 in year 1, 0.057 in year 2.
- **Slopes do differ (H2).** The gradient changes across the line for NDVI decline in both years (Holm p < 0.001) and, before correction, for LST and labour. For NDVI this is the crest of the ridge.
- **Fake boundaries do the same.** Placebo cutoffs inside each side reproduce the NDVI and labour slope changes (12 of 36 placebo tests significant). Only the LST pattern passes that test, and it is a single pre-monsoon date.
- **Irrigation explains the eastern fall-off.** The Census 2011 irrigated share (0.4–77 % by taluka; about 11 % on the Berar side and 34 % on the Central Provinces side) correlates with the NDVI decline at r = −0.57 and −0.72. With irrigation as a control the NDVI slope changes remain (p = 0.004, < 0.001), the LST slope change does not (p = 0.075), and no step appears.

**Validation and conclusion.** At district level the NDVI index tracks the official cotton share (r = 0.86 for 10 districts; 0.77 without the two paddy districts), so it measures a real agricultural pattern. The cotton core of Vidarbha straddles the old line rather than stopping at it. The evidence shows no discontinuity at the line, and it cannot attribute anything to land tenure.

**Keywords:** spatial discontinuity, boundary persistence, historical GIS, Landsat, NDVI, land surface temperature, Conley standard errors, Vidarbha, Ryotwari, Malguzari