Continuous monitoring

The vast amount of Copernicus satellite imagery, combined with the latest big-data processing infrastructure and advanced AI/ML and time series analysis algorithms, makes continuous monitoring of LULC a logical next step in the CLMS portfolio. The main objective of this method is to roll out tools for continuous monitoring and to apply these to generate innovative forestry, urban, and LULCC products. The focus is on developing and testing structural time series modelling, advanced change detection approaches of explainable AI/ML and continuous mapping of Land Surface Characteristics (LSC).

In the frame of EvoLand, we are testing a model updated with every new observation able to compensate for inter-annual phenological differences in deciduous forests. Changes are detected as statistically significant differences between model prediction and real observation. The model is also able to predict future states and generate up-to-date cloud-free (model) images as input for AI/ML analysis.

EvoLand will investigate ML algorithms for classifying the main forest disturbance cause based on innovation intensities, a temporal summation of innovations (Moving or Cumulative Sum), spectral properties, and an analysis of model components, patch size and relevant auxiliary information (e.g., EFFIS fire danger forecasts).

Through the Continuous monitoring method, we will also work on methods to establish time series of Land Surface Characteristics (LSC). To do so we will integrate the strengths of advanced DL techniques (for temporal, contextual information) and pixel-based classifiers (to retain the level of spatial detail). LSC maps will be probability based which are consolidated throughout time and stored for further retrieval and use, e.g., as input to land cover mapping or change alert products.