Forest disturbances are increasing in frequency and intensity due to global change. Effective conservation requires monitoring these events at scale. Current methods often rely on coarse-resolution data or simple temporal averages that ignore local ecological context. To detect browning accurately, we need an expected baseline for every location by accounting for its local environmental conditions.
Why?
Interactive Anomaly Map: High-resolution (10m) mapping of forest browning across Switzerland based on NDVI observations on 22/08/2023 and 24/08/2023. Forest pixels with a negative anomaly score or cloud cover are shaded. Other forest pixels fell within their expected NDVI ranges and are not highlighted.
Predicted Seasonal Cycles: Per-pixel expected NDVI ranges (shaded) versus actual observations (points).
Abstract
Natural and anthropogenic disturbances are impacting the health of forests worldwide. Monitoring forest disturbances at scale is important to inform conservation efforts. Here, we present a scalable approach for country-wide mapping of forest greenness anomalies at the 10 m resolution of Sentinel-2. Using relevant ecological and topographical context and an established representation of the vegetation cycle, we learn a predictive quantile model of the normalised difference vegetation index (NDVI) derived from Sentinel-2 data. The resulting expected seasonal cycles are used to detect NDVI anomalies across Switzerland between April 2017 and August 2025. Goodness-of-fit evaluations show that the conditional model explains 65% of the observed variations in the median seasonal cycle. The model consistently benefits from the local context information, particularly during the green-up period. The approach produces coherent spatial anomaly patterns and enables country-wide quantification of forest browning. Case studies with independent reference data from known events illustrate that the model reliably detects different types of disturbances.
Context-Aware Phenology Modeling
Our method moves beyond simple historical baselines by using predictive quantile modeling. We train a Multilayer Perceptron (MLP) to predict the parameters of a double logistic function. This function represents the vegetation cycle through six key phenological parameters: minimum and maximum NDVI, start of season, duration of green-up, start of senescence, and duration until dormancy.
The model incorporates high-resolution environmental context to adjust for local conditions:
- Topographical Context: Information such as elevation, slope, aspect, and terrain wetness index.
- Ecological Context: Information such as forest height, tree species, and habitat type.
By minimizing pinball loss, the network estimates the 0.25 and 0.75 quantiles. This creates a robust, pixel-specific interquartile range that serves as the "normal" reference. Observations falling significantly below this range are identified as browning anomalies.
Detecting Real-World Disturbances
The framework was validated against five known disturbance types in Switzerland:
Case Studies: Aggregated anomalies for five known disturbance events, showing clear temporal patterns of browning that align with the timing and severity of each event.
- Forest Fire (Bitsch, 2023): Over 75% of pixels in the burn area were identified as anomalous immediately after the fire, with median scores falling below -4.
- Summer Drought (2018): Captured a rapid increase in browning on July 29, 2018, with the anomalous fraction jumping from 0% to 37% in just three days.
- Windthrow (Storm Burglind, 2018): Correctly identified localized damage, with the anomalous fraction reaching near 100% in confirmed reference areas.
- Clear-cut (Airolo, 2021): Detected a sharp, persistent drop to anomaly scores of -3 during the growing season following the harvest.
- Beetle Outbreak (Lake Zurich, 2020): Identified browning signatures peaking near 100% anomalous pixels, with the signal persisting into the following year.