ELABORATION OF THE ITALIAN PORTION OF THE GLOBAL SOIL ORGANIC CARBON MAP (GSOCMAP)

The Global Soil Organic Carbon map (GSOCmap) published by the Food and Agriculture Organization<br> constitutes a baseline estimation of soil organic carbon stock (CS, ton ha–1) from 0 to 30 cm, on a grid at 30 arc-seconds<br> resolution (approximately 1 x 1 km). It has been produced for the Italian territory by the Italian Soil Partnership (ISP): a<br> national hub of institutions dealing with soils, either academic/research institutions, and regional soil services (RSS). The<br> RSS are the main soil data owners in Italy and play a central role in the elaboration of policies for soil management. The<br> RSS adhering to the ISP are: Calabria, Campania, Emilia Romagna, Friuli Venezia Giulia, Liguria, Lombardia, Marche,<br> Piemonte, Puglia, Sicilia, Toscana, and Veneto. A national soil database is maintained by the Consiglio per la Ricerca e<br> l'Analisi dell'Economia Agraria (CREA). The RSS contributed with soil data, with mean density of 1 point per 50 square<br> kilometres, selecting data analysed for soil organic carbon content (SOC, dag kg-1), which were representative and well<br> distributed for the following environmental covariates: land use, geomorphology, and climate. The data were selected inbetween<br> 1990 al 2013. This was necessary in order to exclude the effect of the new soil protection policies of the Rural<br> Development Programme 2014-2020. For the RSS not included in the ISP, the data were selected from the national soil<br> database. 6748 point data were finally selected. SOC values obtained with the Springer and Klee and flash combustion<br> elemental analyser methods were retained for elaborations, because the 2 methods, were found to give statistically<br> equivalent results. SOC values obtained with Walkey and Black method were, instead, corrected with an empirical factor<br> of 1.3. 2292 of the 6748 point data had also measured bulk density (BD, Mg m–3). Pedotransfer functions were calibrated<br> to estimate BD were measured BD were missing, with the following as auxiliary variables: land use, soil regions, texture,<br> and SOC. The carbon stock (CS, ton ha–1) was calculated by multiplying: 0.3 (m) * SOC (dag kg-1) * fine earth fraction (1 -<br> skeletal content expressed as daL m–3) * BD (Mg m–3). CS of the first 30 cm depth was calculated as depth-weighted<br> average. A spatial statistics method was used for the CS interpolation. The following auxiliary variables were used: soil<br> regions, soil subregions, Corine land cover 2006, lithology, soils affected by natural constrains (gleyic, histic, vertic,<br> coarse, shallow, arenic, sodic, and acid), sand content, silt content, 30-m aster-DEM, distance from coast, distance from<br> relieves, soil aridity index, annual mean precipitations, mean annual air temperature, soil inorganic carbon, and soil<br> depth. For the soil region of Po valley, the land units at 1:250,000 scale were also used. The interpolation method was a<br> general linear regression for the soil regions of Po valley, and a radial basis function for the remaining Italian territory.<br> The 6748 point data were divided, by spatial random sampling, into 10 subsets. Ten interpolations were produced, each<br> time leaving out 1/10 of the dataset. Average (fig. 1), standard deviation and confidence intervals of these 10<br> interpolations were calculated. Mean Absolute Errors (MAE) and Root Mean Squared Errors (RMSE) were respectively<br> 25.5 and 36.4 Mg/ha. A.85 Italy Map source: Country submission Point data Number of samples: 6748 Sampling period: 1990-2013 SOC analysis method: SOC values obtained with the Springer and Klee and ’flash combustion elemental analyser’ methods were retained for elaborations. Uncorrected values obtained by the Walkey and Black method were corrected with an empirical linear equation, based on previous studies and as recommended by the Italian official methods. BD analysis method: Undisturbed sampling, core method and pit method Mapping method Mapping method details: Neural Networks and GLM, according to soil region Validation statistics: Mean Error (ME) of the prediction is 1.688 Mg/ha, MAE 25.57 Mg/ha, Root Mean Squared Error (RMSE) is 36.24 Mg/ha. Contact Data Holder: Research centre for agriculture and environment Contact: CREA Consiglio per la ricerca in agricoltura e l’analisi dell’economia agraria [e-mail]