Forest carbon in Italian forests: Stocks, inherent variability and predictability using NFI data

National forest inventories (NFIs) are a primary source of data for carbon stock reporting. In Italy the most comprehensive assessment of forest carbon has been carried out under the second NFI. The NFI was conducted between 2003 and 2007 with an additional field exercise in 2008 and 2009, which collectively aimed to exhaustively survey four out of five forest carbon pools recognized by IPCC. The final estimates were published in 2013 and confirmed the major role of soil as a carbon store. Italian forests contain 1.24 × 109 Mg organic carbon, 57.6% of which is in soil (715.7 × 106 Mg), 38.1% in the above-ground biomass (472.7 × 106 Mg), 2.3% in litter (28.3 × 106 Mg) and 2.0% in deadwood (24.9 × 106 Mg). This article analyzed the variability of total above-ground living biomass and soil organic carbon stocks along with environmental gradients related to elevation and biogeographic regions; examined the carbon storage capacity of the different forest categories; investigated the relationships, at plot level, between both litter carbon and soil organic carbon, and other site/stand features, as well as the relationships between the four carbon pools investigated at the level of forest categories. Forest category proved to be the main driver of variability in carbon stocks, especially for the total above-ground biomass (total-AGB). The observed variation in total-AGB with altitude was primarily due to the distribution of forest categories across altitude classes. There was less variation in SOC between altitude classes than in total-AGB, and soil generally remains the main carbon pool at any altitude. The environmental parameters related to the biogeographic regions did not have a clear effect on total-AGB or soil carbon stocks. Significant correlations between both litter carbon and soil organic carbon and different site/stand features were found in the plot level data, but the amount of variation explained by the multiple regression analysis was too small for use in predictions. More variation was explained when fitting data at the forest category level.