Remote Sensing for Heritage Sites: Satellite Imagery, InSAR and Change Detection

Why Remote Sensing Matters for Heritage

Ground-based surveys are thorough but slow, expensive to repeat, and physically limited to the area a team can cover in a field season. Satellite remote sensing can provide continuous temporal coverage across thousands of sites simultaneously at a fraction of the cost — allowing heritage agencies to detect change quickly and deploy ground resources where they are most needed.

Three problems in Indian heritage management make remote sensing particularly valuable: First, encroachment monitoring at scale — India has over 3,600 centrally protected monuments and tens of thousands of state-protected sites that cannot all be physically inspected on a quarterly basis. Second, structural deformation detection — subsidence, tilt, and settlement in foundations are often invisible until cracks appear, but InSAR can detect them years earlier. Third, buried site discovery — Tamil Nadu and other regions have thousands of unexcavated archaeological sites visible only as spectral anomalies in agricultural land.

Sensor Types and Their Applications

Different sensors capture different physical properties of the landscape and structures. Choosing the wrong sensor for a task produces misleading results — or no useful result at all.

Which Sensor for Which Heritage Task

This decision table is ordered by task — because the task is what you know first, not the sensor. Use it to select the most cost-effective approach before allocating budget.

Satellite Change Detection for Encroachment Monitoring

Change detection compares satellite images of the same area at two or more points in time to identify what has changed. For heritage buffer zones, the primary use is detecting illegal construction, vegetation removal, or land use conversion.

Sentinel-2 revisits most Indian sites every five days (combined Sentinel-2A and 2B), providing excellent temporal density. At 10-metre resolution, the sensor reliably detects building footprints from approximately 50 m² upward — adequate for detecting construction encroachments.

The standard approach is image differencing: subtract a baseline classified map from a current image and threshold the difference. Changes above the threshold are flagged for ground verification. A two-year baseline is necessary before operationalising an alert system — otherwise seasonal vegetation change triggers constant false positives.

For buffer zones where illegal construction can proceed rapidly, Planet Labs' daily 3-metre imagery closes the temporal gap at higher cost. For court-admissible evidence, sub-metre imagery from Cartosat-3 (via NRSC) or WorldView is required — 10-metre data has insufficient geometric precision for boundary dispute cases.

InSAR: Monitoring Structural Deformation

Interferometric SAR (InSAR) measures surface deformation by comparing the phase of radar signals acquired on different dates. Because radar wavelengths are measured in centimetres, phase differences correspond to movement at millimetre precision. This makes it uniquely capable of detecting structural problems before they become visually apparent.

Sentinel-1 provides free C-band SAR data with approximately 6-day repeat at Indian latitudes. A stack of 20–30 images processed with Persistent Scatterer InSAR (PS-InSAR) or Small Baseline Subset (SBAS) methods can reveal deformation trends on monument surfaces and surrounding ground.

The critical interpretive step is separating seasonal thermal expansion (reversible, cyclical signal) from genuine structural movement (trend signal, potentially irreversible). Always plot deformation time series, not just cumulative deformation maps. A monument that shows ±3 mm seasonal variation but no annual trend is behaving normally. One showing a consistent downward trend of 4 mm per year requires investigation.

• Differential foundation settlement between zones of a large monument complex
• Ground subsidence linked to groundwater extraction or tunnelling near a site
• Seasonal clay soil swelling/shrinkage under historic foundations
• Post-earthquake or post-cyclone structural movement at vulnerable sites
• Long-term consolidation in reclaimed coastal land bearing historic port structures

Multispectral Analysis for Buried Site Discovery

Buried archaeological features — ditches, walls, floors, pits — affect the moisture retention and rooting depth of overlying vegetation. These differences create subtle spectral anomalies detectable in near-infrared (NIR) and short-wave infrared (SWIR) bands.

Crop marks appear when cereal or grass crops grow differentially over buried features: faster and greener over ditches (more moisture, deeper roots), stunted over buried walls (shallower soil, water stress during dry spells). The contrast is maximised during periods of water stress. In Tamil Nadu's agricultural areas, the late dry season (March–May) after the northeast monsoon harvest is the optimal window — fields are bare or in early cultivation, and water stress differentials are greatest.

Soil marks appear on bare ploughed fields where subsoil features are exposed: darker soils over organic-rich pits and ditches, lighter soils over masonry remains. These are more common in northern India where extensive wheat and rice agriculture creates predictable bare-field periods. Sentinel-2 SWIR bands (Band 11 at 1610 nm and Band 12 at 2190 nm) are particularly sensitive to soil texture and moisture differences.

Indian and International Data Sources with Costs

India has a mature domestic Earth observation programme through ISRO and NRSC, providing data pathways not available through European or American platforms. For heritage applications, knowing which portal to use for which resolution requirement saves significant time and money.

Processing Workflow: Raw Image to Heritage Decision

Raw satellite imagery cannot be used directly for heritage analysis. A standard pre-processing chain is required before deriving any conclusions. Skipping steps — particularly atmospheric correction — is the most common cause of unreliable results.

1. 1Scene selection: identify acquisitions with <10% cloud cover over the site area for the target season
2. 2Atmospheric correction: convert top-of-atmosphere reflectance to surface reflectance (Sen2Cor for Sentinel-2; LaSRC for Landsat). This step is mandatory for multi-date comparison. Skip only for InSAR.
3. 3Co-registration: align multi-date images to sub-pixel accuracy (<0.5 pixel RMSE) before any change detection
4. 4Index calculation: compute NDVI, NDWI, NBR, or custom indices for the analysis task
5. 5Change detection: apply threshold, supervised classification, or ML-based change algorithm
6. 6Ground truth validation: field-verify at least 10–20% of flagged change polygons before reporting findings
7. 7Documentation: record the full processing chain — software, version, parameters, dates — for reproducibility and legal admissibility

Cost, Time and Accuracy Trade-offs

The following table provides realistic estimates for common heritage remote sensing tasks in the Indian context. Times assume a single analyst with moderate GIS and remote sensing experience.

Field Notes from Practice

**Coastal temple monitoring during northeast monsoon:** Several temple complexes along the Tamil Nadu Coromandel Coast are vulnerable to cyclone storm surge and long-term coastal erosion. During the northeast monsoon (October–December), optical satellites cannot acquire cloud-free imagery — precisely when damage risk is highest. Sentinel-1 SAR (which penetrates cloud and operates at night) provides the only viable monitoring window. A time series of Sentinel-1 VV/VH backscatter through three monsoon seasons reveals inundation extent, duration, and recession timing — data that was previously unavailable and that now directly informs conservation planning for these sites.

**Urban subsidence under old town heritage quarters:** Old town areas in Coimbatore, Madurai, and parts of Chennai with colonial-era and pre-colonial fabric sit on heterogeneous fill and alluvial soils. Municipal groundwater extraction creates spatially variable subsidence patterns. PS-InSAR from Sentinel-1 reveals these patterns at block level — identifying which clusters of historic buildings are moving and at what rate — years before visible cracking manifests. This makes proactive foundation investigation feasible rather than reactive repair after failure.

**Cauvery delta crop mark survey:** The Cauvery delta has extensive Chola-period settlement archaeology including irrigation tank bunds, village boundaries, and channel networks. March–April Sentinel-2 NDVI anomaly mapping over dry paddy fields has revealed linear features consistent with buried tank embankments and field divisions. Ground verification confirmed archaeological significance in several cases. This is reproducible reconnaissance using publicly available data — no specialist survey budget required beyond analyst time.

Common Mistakes Made by Professionals

These mistakes recur across projects and produce either misleading conclusions or wasted effort.

• Using raw digital number (DN) values instead of atmospherically corrected surface reflectance: raw DNs are not comparable between scenes from different dates or sensors. All spectral indices (NDVI, NDWI, NBR) require surface reflectance inputs.
• Ignoring terrain shadow in hilly sites: shadowed pixels are misclassified as water or dark soil. Sites in the Western Ghats or on hill forts require topographic correction using a DEM before any spectral analysis.
• Reporting InSAR line-of-sight displacement as vertical: InSAR measures movement along the radar look direction (line-of-sight), not vertically. Separating vertical from horizontal displacement requires ascending and descending pass data combined.
• Overinterpreting multispectral anomalies as archaeological features: soil texture variation, drainage patterns, past cultivation, and recent disturbance create anomalies identical to buried features in spectral appearance. All anomalies require ground truth before any archaeological claim.
• Using 10-metre Sentinel data for buffer zone boundary disputes: 10-metre data has insufficient precision for legal boundary definition. Sub-metre imagery is mandatory for boundary evidence submitted to courts or enforcement bodies.
• Failing to account for image acquisition geometry: off-nadir commercial imagery produces building lean and geometric distortion that can offset structure outlines by several metres. Use only orthorectified products for any positional measurement.
• Not establishing a baseline before raising alerts: a change detection system without a two-season baseline will flag every monsoon vegetation cycle as an encroachment event, destroying credibility with field teams.

Frequently Overlooked Considerations

These points rarely appear in standard remote sensing courses but are critical in heritage applications:

• Data licensing for legal use: Sentinel data is freely licensed for all uses including legal proceedings. Cartosat and WorldView commercial data has redistribution restrictions. Confirm licensing before submitting satellite evidence to courts or enforcement bodies.
• Coordinate system mismatch: Survey of India heritage site boundary records are in the Everest 1830 datum. Satellite data is referenced to WGS84. The offset is 100–200 m in Tamil Nadu — potentially the difference between a construction being inside or outside a protected buffer zone. Transform all layers to a common datum before overlay.
• Temporal baseline as the core asset: a single high-resolution image tells you the current state. A 10-year time series of moderate-resolution images tells you the history. For heritage monitoring, the archive is the most valuable product.
• Night-time light data for urbanisation pressure: VIIRS night-time lights (free, NOAA) track urbanisation pressure around heritage buffer zones across years — useful for long-term strategic conservation planning.
• Distinguishing thermal expansion from structural movement in InSAR: without plotting the full time series, seasonal thermal expansion on stone monuments (which can be 5–10 mm amplitude) can be misread as settlement. Always plot time series and identify the seasonal component before reporting a trend.
• Monsoon imagery gaps: India's two annual monsoon systems create multi-month cloud cover gaps in optical data over different regions. Plan analysis windows around these gaps — or use SAR to fill them.

Pre-Project Checklist

1. 1Define the specific question (encroachment / deformation / discovery / damage assessment) before selecting any sensor
2. 2Select sensor and resolution appropriate to the detection limit required — use the decision table above
3. 3Identify the optimal acquisition season for the site type and region (March–May for dry Tamil Nadu crop marks; October–December use SAR for monsoon monitoring)
4. 4Download and verify data licensing terms for the intended use (research / enforcement / legal proceedings)
5. 5Check coordinate system of existing site boundary records — transform to WGS84 if using ESA or commercial satellite data
6. 6Perform atmospheric correction on all optical imagery before deriving spectral indices
7. 7Establish a minimum two-season cloud-free baseline before operationalising any change alert system
8. 8Set change alert thresholds based on observed seasonal variation in the study area, not default software values
9. 9Plan the ground verification protocol: who verifies, what they record, what constitutes confirmation
10. 10Document the full processing chain (software, version, parameters, dates) before reporting findings