IDENTIFYING CANOPY GAPS IN THE MONTEVERDE, COSTA RICA CLOUD FOREST USING VERY HIGH RESOLUTION SATELLITE IMAGERY AND GROUND-BASED MEASUREMENTS

Canopy gaps are a fundamental element of the physical structure of tropical rain forests, and they play an important role in maintaining biological diversity. Gaps caused by tree and branch falls are often associated with strong wind events, and gap size and density vary spatially throughout the forest in response to biological and physical variables. The germination, growth, and reproduction rates of pioneer plants that recruit into canopy gaps depend on both gap size and age. Our goal in this research was to develop a technique to automatically identify and measure canopy gaps in very high resolution satellite imagery (< 1 m ground sampling distance). Using a geometrically and radiometrically corrected 4-band Pleiades 1 image of the area near the Monteverde Cloud Forest Reserve (MCFR) in Costa Rica, we developed training and validation data sets by visually identifying canopy gaps and regions of uninterrupted forest canopy. Visual identification of a gap was based primarily on the presence of a distinctive shadow. We used a machine learning algorithm (random forest) to develop a classifier using the 4-band pixel reflectance values along with textural information (GLCM contrast for the near-infrared band). Using the validation data, the overall accuracy of the classifier was estimated to be 98%. The classifier was used to create a map of canopy gaps over a 3 km² region including much of the MCFR and to determine an approximate distribution of gap areas. The proportion of canopy identified as gaps increased from SE to NWover the reserve and had an average value of approximately 0.17 over the study region. The accuracy of the classifier is also tested against on-the-ground measurements collected during July-August, 2019 in the MCFR. These data will be used to estimate the frequency of both false positive and false negative identifications of canopy gaps.