Mapping Woody Volume of Mediterranean Forests by Using SAR and Machine Learning: A Case Study in Central Italy

In this paper, multi-frequency synthetic aperture radar (SAR) data at L- and C-bands (ALOS PALSAR Envisat/ASAR) were used to estimate forest biomass in Tuscany, Central Italy. The ground measurements of woody volume (WV, m3/ha), which can be considered as a proxy biomass, retrieved from the Italian National Forest Inventory (NFI). After preliminary investigation assess sensitivity backscatter C- L-bands an approach based on artificial neural network (ANN) was implemented. ANN trained using backscattering coefficient L-band PALSAR, HH HV polarization) C-band (Envisat ASAR inputs. Spatially distributed WV values for entire test area derived by integration (fusion) canopy height map Ice, Cloud, Land Elevation Geoscience Laser Altimeter System (ICESat GLAS) NFI data, order build significant truth dataset training stage. analysis showed moderate correlation almost negligible C-band. Despite this, algorithm able exploit synergy SAR frequencies polarizations, estimating with average Pearson’s (R) = 0.96 root mean square error (RMSE) ≃ 39 m3/ha when applied R 0.86 RMSE 75 validated direct NFI. Considering heterogeneity scenario (Mediterranean mixed forests hilly landscape) small amount available respect spatial variability different plots, obtained results satisfactory. Moreover, successful use global maps implementing suggests possibility apply wider areas or even scales.