Comparing single tree and area based approaches for biomass estimation of Cape Verdean xerophytic forests using visible aerial imagery

Models to estimate the AGB over dry Cape Verdean woodlands were developed using visible high-resolution aerial orthophotography. The proposed method is based on the integration of clustering techniques combined with the Visible Atmospherically Resistant Index (VARI) and segmentation algorithms for tree crowns extraction. This allowed for the minimization of constraints due to poor spectral contrast between the background and tree crowns, especially for brighter parts of the crowns and shadowed parts of the scene. Both single tree and area based approaches were tested and their performances compared on the basis of field data from the National Forest Inventory of Cape Verde (CV-NFI). In the first approach, AGB was calculated as a function of crown width and tree height by the allometric equations developed specifically within the CV-NFI. In the second approach, regression analysis was used in deriving models for biomass as a function of the crown projected area. The accuracy of the values predicted was measured by the Root Mean Square Error (RMSE) against the allometric-based (field-measured) biomass. The models produced similar accuracy in the AGB predictions with NRMSE% of 42% for the first approach and 45% for the second. The mean AGB as estimated from the CV-IFN data for the study area of 14399 ha was 12.701 Mg ha-1. This compared with 11.380 Mg ha-1 predicted by the area based model and 10.278 Mg ha-1 by the single tree model. The findings demonstrate that promising results can be achieved and as expected, the reliability increases with the area for which mean values are presented. Improvements of the forest maps and the stratification in homogeneous layers can lead to enhanced AGB estimations, something which opens opportunities for the practical application of the models for monitoring and management purposes.