Day 4

Detailed paper information

Back to list

Paper title Evaluation of multi-scale multispectral sensors and machine-learning algorithms for mangrove mapping and LAI estimation
  1. Javier Blanco Sacristán King Abdullah University of Science and Technology Speaker
  2. Kasper Johansen King Abdullah University for Science and Technology (KAUST)
  3. Ibrahim Hoteit King Abdullah University of Science and Technology (KAUST)
  4. Matthew McCabe King Abdullah University of Science and Technology
Form of presentation Poster
  • C4. HAPs/UAVs
    • C4.01 Innovative UAV applications
Abstract text Mangroves provide multiple ecosystem services in the intertidal zone of tropical and subtropical coastlines and are among the most efficient ecosystems at storing carbon dioxide. For several decades, remote sensing has been applied to map mangrove distribution and their biophysical properties, such as leaf area index (LAI), which is one of the most important variables for assessing mangrove forest health. However, remote sensing of mangrove LAI has traditionally been relegated to coarse spatial resolution sensors. In the last few years, unmanned aerial vehicles (UAVs) have revolutionised mangrove remote sensing. Nevertheless, the myriad of available sensors and algorithms makes it difficult to properly select a suitable methodology to map their extent and LAI.
In this work we performed a multi-sensor (i.e. Landsat-8, Sentinel 2, PlanetScope and UAV-based MicaSense RedEdge-MX) comparison and evaluated the performance of various machine-learning algorithms (i.e. classification and regression trees (CART), support vector machine (SVM) and random forest (RF)) for mangrove extent mapping in a Red Sea mangrove forest in Saudi Arabia. The relationship between several vegetation indices and LAI measured in-field was also evaluated. The most accurate classification of mangrove extent was achieved with the UAV data using the CART and RF algorithms, with an overall accuracy of 0.93. While the relationships between field-derived LAI measurements and satellite-based vegetation indices produced coefficients of determination (r2) lower than 0.45, the relationships with UAV-based vegetation indices produced r2 up to 0.77. Selecting the most suitable sensor and methodology to assess mangrove environments is key for any program aiming to monitor changes in mangrove extent and associated carbon stock, particularly under the current scenario of climate change, and the results of this work can help on this task.