COLISEUM Project: Heritage Weathering Forecast

• COLISEUM Project is a research initiative that integrates synchronized sensor arrays and comprehensive diagnostics to model and forecast heritage material weathering under climate change.
• It employs standardized instrumentation and multimodal data fusion to capture microclimatic and structural degradation parameters with high precision at distinct French sites.
• The project leverages advanced AI forecasting, achieving promising results (e.g., RMSE ≈ 0.12, R² ≈ 0.85) for predicting material deterioration under IPCC scenario projections.

The COLISEUM Project is a research initiative focused on quantifying, modeling, and forecasting the weathering of heritage materials under climate change through multimodal data fusion and artificial intelligence methods. By establishing tightly synchronized sensor arrays and comprehensive diagnostic protocols at three climatically distinct French heritage sites—Strasbourg Cathedral, Bibracte archaeological site, and Saint-Pierre chapel—the project aims to correlate microclimatic forcing with material degradation, ultimately enabling robust, site-specific predictions of heritage vulnerability under Intergovernmental Panel on Climate Change (IPCC) scenario projections (Cormier et al., 17 Nov 2025).

1. Experimental Setup and Instrumentation

The COLISEUM experimental design employs an identical instrumentation scheme at all target sites to ensure methodological consistency and reliable cross-site comparison. Each site instruments at least two contrasting exposures (e.g., sunlit South-West vs. shaded North-East) and utilizes control blocks (5 × 5 × 5 cm stone cubes) to provide unweathered material references. The sensor suite is designed to capture the leading microclimatic and mechanical drivers of weathering:

Sensor	Parameter(s)	Resolution / Interval
Thermo-hygrometer	Air temperature T (°C), RH (%)	±0.1 °C/±1.5%; 20 min
PT100 surface probe	Surface temperature Ts (°C)	±0.1 °C; 20 min
TDR moisture probe	Masonry moisture θ (%) at 10 cm	±1.0 %; 20 min
Fissurometer	Crack opening Δw (mm)	±0.1 mm; 60 min

All time series are time-synchronized to UTC and downloaded quarterly for analysis. Sensor arrays typically deploy 2–4 units in proximity on each exposure, enabling joint calibration and redundancy (Cormier et al., 17 Nov 2025).

2. Multimodal Data Acquisition and Diagnostics

Weathering is monitored through a four-pronged protocol, ensuring both surface and bulk processes are captured and quantified:

• In Situ Analyses: Colorimetry (CIELab), surface moisture mapping, high-resolution crack photography (visible/UV/IR), and geometric registration via total station and photogrammetry.
• Laboratory Analyses: Ion chromatography quantifies soluble anions; inductively coupled plasma optical emission spectroscopy (ICP-OES) determines cation content; gravimetric moisture, X-ray diffraction (XRD) characterizes crystalline phases; scanning electron microscopy (SEM) provides high-resolution microfabric images.
• Scientific Imaging: Hyperspectral imagery (400–1000 nm) maps mineralogical changes; structured light 3D scanning quantifies geomorphology and crack volume; multispectral photogrammetry campaigns are performed every 3–6 months.
• Textual and Cartographic Context: Archival repair records, historical climate logs, geospatial information system (GIS) maps (block ID, type, exposure), and ICOMOS deterioration glossaries annotate spatial and typological data modalities.

Roughly 70 sub-zones are delineated per site, and each data modality is co-registered in space and time for downstream integration (Cormier et al., 17 Nov 2025).

3. Weathering Index Formalism

Quantitative modeling leverages the "weathering index" $i \in [0,5]$ for each sub-zone, integrating structural configuration, discrete alteration types, and scalar parameters. The index is composed as:

• $i_{\mathrm{structure}}$ (fixed, based on stone, elevation, orientation)
• $i_{\mathrm{alteration}, k}$ (alteration sub-index per damage family $k$, normalized by area-coverage rating)
• $i_{\mathrm{param}, p}$ (normalized scalar parameter $p$ from in situ or lab measurements)

The global sub-zone index is computed as:

$$i_j = \alpha \cdot i_{\mathrm{structure},j} + \beta \cdot i_{\mathrm{alteration},j} + \gamma \cdot i_{\mathrm{param},j}, \quad \text{with} \ \alpha + \beta + \gamma = 1$$

Typically, $\alpha = \beta = \gamma = \frac{1}{3}$ initially. The weathering matrix $W = [i_j(t)]_{j=1}^{N}$ for $N$ sub-zones across monitoring epochs $t$ becomes the central data structure for machine learning and forecasting (Cormier et al., 17 Nov 2025).

4. AI-Powered Forecasting Methodology

COLISEUM employs advanced AI architectures to predict future deterioration as a function of both historical weathering and projected environmental forcing:

• Model Architectures: Transformer encoder–decoder networks leverage self-attention over multimodal input sequences; regression heads predict $\Delta i_j$ over user-specified horizons. Benchmark models include Random Forest regressors and LSTM-based sequences.
• Input Features: Past $W(t)$ vectors, pre-processed climate variables (T, RH, Ts, θ, Δw), and IPCC scenario projections form the multimodal feature set.
• Optimization Protocols: The core loss function is mean squared error (MSE) on $i_j$. Early stopping regulates overfitting. Validation is performed with train/test splits that partition by both sub-zone and epoch (e.g., Apr 2024–Mar 2025 for training; Apr–Sep 2025 for testing).
• Performance Metrics: Root mean square error (RMSE) and $R^2$ (per exposure) are reported. Early experiments at Strasbourg show RMSE ≈ 0.12 and $R^2$ ≈ 0.85 for 3-month horizons.
• Scenario Forecasting: Trained models, seeded with present-day $W(t_0)$, can be rolled forward recursively under downscaled IPCC scenario vectors through 2050, supporting block-, face-, or site-level risk aggregation (Cormier et al., 17 Nov 2025).

5. Initial Site Diagnostics and Results

The first operational test occurred at Strasbourg Cathedral’s spire from April to September 2024, providing a comprehensive case study:

• Microclimatic Extremes: The spire experienced daily temperature cycles up to 39.6 °C and RH > 90% on 94 of 118 days; surface dew-point was exceeded >200 times in 4 months. The South-West face was consistently warmer by 1 °C compared to the North-East.
• Cracking Dynamics: Crack width $\Delta w$ increased from 5.44 mm to 5.56 mm, with RH–crack correlation $r \approx +0.72$ ($p<0.01$).
• Moisture Gradients: Surface TDR probes identified a mean $\theta \approx 5.52\%$ at the upper position and 5.48% at the lower, with net summer drying of –0.5% over 3 months, consistent with top-down water ingress.
• Salt and Water Content: Drill sampling revealed SO₄²⁻ up to 5.51% at shallow depths (regulatory threshold <0.1%), and hygroscopic water $w_h$ up to 13.2%, indicating advanced salt contamination.
• Model Performance: Early transformer-based models yielded RMSE ≈ 0.12 and $R^2$ ≈ 0.85 for short-term blocks, demonstrating the feasibility of multimodal prediction at this temporal and spatial resolution (Cormier et al., 17 Nov 2025).

6. Scalability, Generalization, and Future-Proofing

The methodological architecture of COLISEUM is designed for broad deployment:

• Modular sensor arrays and standardized "batch block" units facilitate adaptation to other masonry heritage sites.
• The weathering matrix formalism supports extensible multimodal data structures, unifying disparate measurement types (time series, spectral images, maps, text).
• Transformer models naturally accommodate missingness and irregular time steps, and their multimodal embedding capability enhances robustness.
• Climate projections downscaled to site level can be directly ingested, enabling automated risk-mapping under any IPCC scenario (e.g., RCP2.6–RCP8.5).
• The data-driven approach establishes a basis for decision-support tools targeted at conservators and policymakers to anticipate material degradation and plan interventions under plausible future climates (Cormier et al., 17 Nov 2025).

7. Adaptation of Distributed Particle Sensing for Muography (CosmoLink Integration)

The CosmoLink system, detailed in (Elangovan, 13 Feb 2025), is leveraged as the scalable, portable array for on-site muon flux monitoring within the COLISEUM framework, with technical adaptations to meet the project’s muography and space-weather objectives:

• Mechanical Scalability: Enlargement of scintillators (up to 200 × 100 × 10 mm³) and rack-based deployment of 50–100 modular detectors.
• Electronic Upgrades: Multi-channel ADCs and FPGA-based TDCs enable sub-5 ns coincidence resolution for high-throughput stacked arrays.
• Environmental Sensors: Integration of humidity, ambient light, and radon monitors allows co-analysis with microclimatic drivers.
• Network Infrastructure: LoRa mesh or hybrid LTE backhaul ensures robust data aggregation across the spatial grid.
• Autonomy: Solar charging, battery management, and remote firmware updates facilitate sustained, large-area operation.

A plausible implication is that high-density cosmic-ray tomography will become a routine correlative measurement within climate–heritage studies, as distributed muon flux variations can be linked with environmental indices tracked elsewhere in the COLISEUM protocol (Elangovan, 13 Feb 2025).

---

The COLISEUM Project represents a comprehensive, modular, and data-driven strategy for monitoring, modeling, and forecasting the impacts of climatic forcing on heritage materials, harnessing synchronized multimodal sensing and advanced AI, and incorporating scalable muography for unprecedented integration of environmental and structural diagnostics (Cormier et al., 17 Nov 2025, Elangovan, 13 Feb 2025).