Ground-Truth Survey for Forest Monitoring in Kumaon

The forestwatch viewer maps forest cover, change, burned area, canopy height and candidate planting parcels for eight study boxes in Kumaon from free satellite data. Every area it reports is a count of map pixels, labelled unassessed, because no reference sample has been collected. Four questions cannot be answered from orbit:

Question	Why it is open
How accurate are the forest, other-class and forest-loss areas?	Areas are pixel counts with no reference-sample accuracy assessment.
What is actually on the candidate parcels?	Lidar from the GEDI instrument shows woody vegetation (median 4.5–7.9 m across the boxes) in parcels the screen accepts at up to 40 % canopy cover; tenure, use, grazing and fire history are unknown.
Do satellite canopy heights match tree heights on the ground?	Both GEDI and the Meta/WRI canopy-height model are samples or predictions, with no local calibration.
Does the satellite burn-severity index match what burned?	In every box 85–100 % of pixels inside mapped burned cells show no change in the burn index, and almost none reach moderate-high. That has not been tested against the ground.

We follow the standard good-practice design for estimating area and accuracy together (Olofsson et al. 2014). The map is split into strata (land-cover classes, plus candidate parcels and recent forest loss as extra strata so rare classes are not starved), and points are drawn at random within each, in homogeneous 3×3 blocks at least 90 m apart. Each point carries a weight so the estimators can recover areas with standard errors.

A target standard error of 0.02 on overall accuracy needs about 300–380 reference points per box. Visiting them all on foot is not affordable, so a trained interpreter first labels every point from licensed very-high-resolution imagery (phase 1), and the teams then visit a random 60 per box (phase 2) to measure how far the interpretation can be trusted on the points imagery cannot resolve: tree against scrub, Lantana, canopy cover and scorch.

Box	Phase-1 points for SE 0.02
Kosi	380
Gananath	309
Jageshwar	351
Gagas–Chanthria	359
Ramgarh–Mukteshwar	317
Sattal–Bhimtal–Naukuchiatal	306

A single visit sees today’s state only. Whether land was forest in 2019 or cleared in 2023 comes from the imagery history and from informants, so accuracy for change will be weaker than for state, and the results will say so.

Component	What	Per box (plan)
A. Map-verification point	30 m plot: cover fractions, canopy cover, tallest tree, disturbance, photos, GNSS position and its own accuracy	60 field points
B. Inventory plot	0.1 ha circle centred on a usable GEDI footprint: species, diameter at 1.37 m, laser heights, regeneration	about 12 plots across forest strata
C. Burn plot	Composite Burn Index with matched controls just outside the burned cells	a handful per burn-severity class
D. Candidate parcel	Boundary check, cover transects, and a consented interview on use, burning, planting and tenure as stated by the community	16 parcels

The planned equipment is deliberately simple: a dual-frequency GNSS receiver (expect 3–10 m under forest on steep ground, so the receiver’s accuracy is logged at every point), laser rangefinder, diameter tape, clinometer, densiometer and cameras. No drones are used.

Interview answers are kept only where the informant has consented, and survey records are private working data; only aggregate and de-identified results are published. Boundaries from the Forest Department and Van Panchayats are used only with written permission. Every returned record is checked automatically against a published schema (ranges, vocabularies, GNSS accuracy, duplicate and out-of-box coordinates) before it is accepted.

The sample-design, validation, accuracy-estimation and field-pack tools are written and tested, and trial draws exist for two boxes. A pilot is planned for the February–March 2027 dry-season window in Ramgarh–Mukteshwar, subject to permissions. Nothing here has been agreed with the Uttarakhand Forest Department, the wildlife authorities, any institution or any village. Fieldwork inside a sanctuary or reserved forest needs permission first, and the plan changes as those answers arrive. The results will be reported with their uncertainty, and the screening maps remain a research preview in the meantime: not field-verified, and never an eligibility decision.

Open the forestwatch viewer