MAPPING THE SPATIAL DISTRIBUTION OF COLOMBIA’S FOREST ABOVEGROUND BIOMASS USING SAR AND OPTICAL DATA
Keywords: biomass, carbon, SAR, multispectral, machine-learning, forest inventory
Abstract. An assessment on the amount and spatial distribution of forest aboveground biomass (AGB) for the forests in Colombia was generated using in-situ national forest inventory data (IDEAM, 2018), in combination with multispectral optical data and synthetic aperture radar (SAR) satellite imagery. ALOS-2 PALSAR-2 gamma-0 backscatter annual mosaics (2015–2017) provided by JAXA were normalised and corrected using previous ALOS PALSAR annual mosaics (2007–2010) as reference. A multi-temporal Landsat 7 & 8 composite over the whole of Colombia was used for the year 2016 ± 1. The national forest inventory in-situ plots used to train our model consisted of 5-subplots each and were collected during the period 2015–2017 in the main biomes of the country. A sample of permanent 1ha plots (PPMs) were also measured. Nationally developed allometries (Alvarez et al., 2012) were used to estimate AGB. A non-parametric random forests (RF) algorithm was used within a k-fold framework to retrieve AGB at 30 m spatial resolution for the whole of Colombia. The algorithm was trained using forest inventory plots and validated at plot (0.35 ha) and PPM level (1 ha). The accuracy assessment found coefficients of determination (R2) of 0.68 and 0.61, and relative root mean square errors (Rel. RMSE) of 49% and 34% at plot and at PPM level, respectively. The results showed that the average AGB for the country was 118.1 t ha−1 (45.6 t ha−1 for Caribe, 75.4 t ha−1 Andes, 122.5 t ha−1 Pacifico, 32.7 t ha−1 Orinoquia, and 200.5 t ha−1 for the Amazonia, regionally), and that the total carbon stocks for the country were 6.7 Pg C for the period 2015–2017.