DIMPEO

Motivation

In recent years, the damage inflicted on forests by increasingly dry and hot summers has emerged as a critical concern for diverse stakeholders across Switzerland and neighboring regions. Understanding when and where climate extremes impact forest health and vitality is essential for developing effective forecasting and early warning systems. Satellite-based remote sensing offers a powerful means of monitoring forests, allowing the detection of anomalous "browning"—an unexpected decline in plant growth or a sudden reduction in plant productivity—on the vegetated land surface. However, processing and analyzing large and noisy Earth Observation (EO) datasets present significant challenges. How can we reliably identify and interpret these anomalies? To address this, our project implemented a robust pipeline for detecting forest browning events using high-resolution EO time series data.

Proposed Approach / Solution

Given the pronounced seasonality of vegetation greenness, we focused on modeling annual greenness cycles. By constraining these cycles to a double sigmoid function, we employed a neural network to predict ecologically interpretable parameters of the sigmoid based on diverse variables, including local elevation, exposition, climate, forest type, and terrain. To model "normal" greenness cycles, the framework utilized quantile regression to produce confidence intervals, facilitating the identification of deviations indicative of anomalies. This anomaly detection model is scalable to the high-resolution EO data covering the entirety of Swiss forests and enables the creation of interactive maps that illustrate the spatio-temporal patterns of forest drought anomalies.

Impact

Strengthening the capacity for drought impact monitoring and near-term forecasting (with lead times of weeks to months) has been designated a high priority by the Swiss Federal Council. A national program for its implementation has recently been initiated under the National Centre for Climate Services (NCCS). The methodologies developed through this project contribute to an open-access, robust, transparent, and updatable workflow for detecting and quantifying drought impacts. These advancements lay the groundwork for improving drought impact assessment and forest fire forecasting methods in future research.