Eliades et al. (2026) Reference-anchored envelope modelling for predicting tree mortality hotspots driven by drought

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2026
• Date: 2026-09-22
• Authors: Filippos Eliades, Dimitrios Sarris, Felix Bachofer, Silas Chr. Michaelides, Chris Danezis, Diofantos Gl. Hadjimitsis
• DOI: 10.1016/j.jag.2026.105609

Research Groups

• Department of Civil Engineering and Geomatics, Remote Sensing and GeoEnvironment Lab, Cyprus University of Technology, Limassol, Cyprus
• Eratosthenes Centre of Excellence, Limassol, Cyprus
• KES Research Centre, Nicosia, Cyprus
• KES College, Nicosia, Cyprus
• German Aerospace Center (DLR), Earth Observation Center (EOC), Wessling, Germany

Short Summary

This study proposes a novel reference-anchored, quantile-envelope susceptibility model to predict drought-driven tree mortality hotspots by combining physiographic similarity, drought severity, and Landsat spectral collapse. The model, calibrated on historical events in Cyprus, successfully identified independent mortality sites in 2025, demonstrating its transferability and utility for proactive forest management.

Objective

• To develop and test a reference-anchored, envelope-based susceptibility model implemented in Google Earth Engine (GEE) that delineates mortality-like “high-risk areas” by integrating AOI-derived physiographic similarity constraints, Landsat-based event-year spectral decline, and a post-event persistence requirement.
• To assess the extent to which signatures from confirmed mortality events can predict near-future high-risk areas beyond the Areas of Interest (AOIs).
• To determine the optimal tradeoff between sensitivity and specificity for the majority rule in the model.
• To identify which predictors contribute most to performance and which are weak or site-dependent.

Study Configuration

• Spatial Scale: National forest domain of Cyprus (country-wide screening layers); 30 meter spatial resolution. Analysis focused on three specific forested areas (Akamas, Stavrovouni, Machairas) as Areas of Interest (AOIs).
• Temporal Scale: Annual time series from 1984 to 2022 for remote sensing data and drought index. Reference mortality events from 2008 and 2016. Independent validation using 2025 mortality sites. Pre-event baseline of 3 years.

Methodology and Data

• Models used: Reference-anchored, quantile-envelope susceptibility model; implemented through a reproducible workflow in Google Earth Engine (GEE). Kernel Density Estimation (KDE) for heatmap generation. Binomial test for statistical significance.
• Data sources:
  • Remote Sensing: Annual time series (1984–2022) of Landsat 5 TM, Landsat 7 ETM+, Landsat 8 OLI, and Landsat 9 OLI-2 Collection 2 Level-2 surface reflectance data (30 m resolution). Derived Normalized Difference Vegetation Index (NDVI) and Normalized Burn Ratio (NBR).
  • Drought Index: Standardized Precipitation-Evapotranspiration Index-12 (SPEI-12) (CSIC/SPEI/2_10) for 1984–2022, derived from Cyprus Department of Meteorology climate data.
  • Topographic Data: COPERNICUS/DEM/GLO30 (Copernicus Data Space Ecosystem) for elevation, slope, aspect, Topographic Position Index (TPI at 300 m and 1000 m), and curvature (curv8).
  • Geological Data: National geological map from the Geological Survey Department of Cyprus for substrate (lithology) codes.
  • Reference Data: Three confirmed tree-mortality footprints (training AOIs) in Cyprus (Akamas, Stavrovouni, Machairas) for 2008 and 2016 events. Two independent mortality sites in 2025 (Orkonta-Kambos tis Tsakistras, Pyrgos-Kambos tis Tsakistras) for validation.

Main Results

• The within-AOI evaluation (forest-only) showed substantial representativeness (spatial coverage 70.49%–85.35%) and high recovery of most reference mortality pixels (recall 67.62%–100%) under the relaxed envelope configuration.
• Two independent mortality sites identified in 2025 in Cyprus (OKT and PKT) coincided with exceptional drought (SPEI < -2) and a sharp decline in NBR, and were fully contained within the high-risk zones mapped by the model calibrated on 2008/2016 events.
• For the 2025 sites, 83.50% (OKT) and 82.69% (PKT) of their forested area exceeded a high-intensity heatmap threshold (HEAT ≥0.60), with mean risk scores of 84.91% (OKT) and 75.55% (PKT).
• With 18.88% of Cyprus’s total forest flagged as high-risk, the probability of the 2025 co-occurrence being random is approximately 0.036 (p < 0.05).
• Single-feature diagnostics indicated that slope, multi-scale topographic position (TPI300, TPI1000), and curvature (curv8) exerted the strongest and most consistent control on the relaxed classification, while aspect was secondary and site-dependent.
• Leave-one-AOI-out cross-validation demonstrated transferability of spectral decline thresholds, with coverage and recall remaining above the 60% reliability gate for held-out AOIs (e.g., Area1: 85.35% coverage, 100.00% recall; Area2: 61.89% coverage, 70.01% recall).

Contributions

• Proposes a novel reference-anchored, quantile-envelope susceptibility model for predicting drought-driven tree mortality hotspots, addressing the challenge of strong influence from climate, landform, and substrate.
• Develops an operational framework that simultaneously anchors mortality signatures to confirmed footprints, accounts for physiographic landscape controls, and transfers predictions across contrasting species and terrain settings within a reproducible cloud-based workflow (Google Earth Engine).
• Demonstrates the transferability of the model by successfully predicting independent, near-future mortality events (2025) using signatures derived from past events (2008, 2016), providing an out-of-sample plausibility check.
• Provides a practical, transferable screening product for prioritizing surveillance and ground checks in high-risk environments, shifting from broad surveillance to targeted intervention for forest management.
• Leverages globally available Earth observation covariates, making the framework readily transferable to other semi-arid regions with limited documented mortality episodes.

Funding

• ‘EXCELSIOR’ project (European Union’s Horizon 2020 Research and Innovation Programme), grant number “857510”.
• Government of the Republic of Cyprus through the Directorate General for the European Programmes, Coordination and Development.
• Cyprus University of Technology.

Citation

@article{Eliades2026Referenceanchored,
  author = {Eliades, Filippos and Sarris, Dimitrios and Bachofer, Felix and Michaelides, Silas Chr. and Danezis, Chris and Hadjimitsis, Diofantos Gl.},
  title = {Reference-anchored envelope modelling for predicting tree mortality hotspots driven by drought},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2026},
  doi = {10.1016/j.jag.2026.105609},
  url = {https://doi.org/10.1016/j.jag.2026.105609}
}