Accuracy Assessment of Sentinel 2 Image-Derived Leaf Area Index (LAI) of Plant Canopies and Other Land Use/Cover Types in Mangrove Areas

• Main Authors: Apan, Armando A, Castillo, Jose Alan A, Maraseni, Tek N, Salmo, Severino G, III
• Format: text
• Published: Archīum Ateneo 2018
• Subjects: Sentinel 2; Leaf Area Index (LAI); mangroves; Philippines; Environmental Sciences
• Online Access: https://archium.ateneo.edu/es-faculty-pubs/36; https://www.researchgate.net/publication/340583195_ACCURACY_ASSESSMENT_OF_SENTINEL_2_IMAGE-DERIVED_LEAF_AREA_INDEX_LAI_OF_PLANT_CANOPIES_AND_OTHER_LAND_USECOVER_TYPES_IN_MANGROVE_AREAS
• Institution: Ateneo De Manila University

Leaf area index (LAI) quantifies the amount of foliage area per unit ground surface area. A dimensionless quantity (i.e. LAI = leaf area / ground area, m2 / m2), it is an important variable to predict photosynthetic primary production and for monitoring vegetation dynamics. The aim of this study was to assess the accuracy of Sentinel 2 image-derived LAI of plant canopies and other land use/cover types in tropical mangrove areas in the Philippines. This is part of a bigger study on mapping of above-ground biomass and soil carbon of mangrove forests. Using hand-held CI-110 Plant Canopy Imager, field-level LAI measurements were made between June-July 2015, in mangrove areas from 90 sample sites located in Honda Bay, Palawan, Philippines. A regression analysis was conducted between the a) Sentinel 2 (Level-1C, acquired in April 2016) image-derived LAI values, and b) field-collected LAI measurements. The regression analysis was accomplished using the traditional linear regression and new machine learning algorithms from the WEKA software. The accuracy of prediction was analysed using root mean square error (RMSE) and agreement (r) of predicted and observed values from leave-one-out cross-validation method. The results show that the field collected LAI data ranges from 0 - 2.71 (mean = 0.86), while the Sentinel 2 image-derived LAI values range from 0.19 - 4.32 (mean = 1.64). The regression models had 0.9419 correlation/agreement of observed and predicted value for linear regression; and had 0.9435 for using the support vector machine algorithm. The corresponding RMSE obtained were 0.3350 (linear regression) and 0.3424 (support vector machine), demonstrating a high prediction accuracy although the two datasets’ LAI values were nine months apart. This study validates the good accuracy of the Sentinel 2 image-derived LAI values, indicating their usefulness for selected applications.