Secondary forests (SFs) are one of the major carbons sinks in the Neotropics due to the rapid carbon assimilation in their above-ground biomass (AGB). However, the accurate contribution of SFs to the carbon cycle is a great challenge because of the uncertainty in AGB estimates. In this context, the main objective of this study is to explore full polarimetric Advanced Land Observing Satellite/Phased Array L-band Synthetic Aperture Radar-2 (ALOS/PALSAR-2) data to model SFs AGB in the Central Amazon. We carried out the forest inventory in 2014, measuring 23 field plots. Supplementary land-use classification history was used to create 120 additional independent sample plots by adjusting growth curves using SFs age and previous land-use intensity from field plots and literature database. Multiple Linear Regression (MLR) analysis was performed to select the best model by corrected weighted Akaike Information Criterion (AICw) and validated by the leave-one-out bootstrapping method. The best-fitted model has six parameters and explained 65% of the above-ground biomass variability. The prediction error was of Root Mean Square Error of the Prediction (RMSEP) = 8.8 ± 3.0 tonnes ha−1 (8.8%). The most explanatory variables for modelling secondary forest AGB were those that result from multiple scattering (Shannon Entropy), volumetric scattering (Bhattacharya decomposition), and double-bounce scattering (ratio VV/HH, vertically transmitted and received polarization/horizontally transmitted and received polarization). Including past-use of SF areas in the model with the Landsat time series classification, as the frequency of clear cuts and the number of years of active land-use before abandonment, the MLR has increased by 10%, achieving 71% of the variability explained by the model. The uncertainty report showed that ground truth AGB estimation (inventory, allometry, and plot expansion factors) might represent 50% of the errors in the modelling estimation. In contrast, Synthetic Aperture Radar (SAR) inversion models (SAR error and regression) have achieved 20%. The results showed that additional information on secondary forest land-use history could improve the performance of AGB recovery models, as well as can be used to expand the sampling units on tropical forests.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)