Factors Influencing Soil Organic Carbon Stock Variations in Italy During the Last Three Decades

Abstract

Soils contain about three times the amount of carbon globally available in vegetation, and about twice the amount in the atmosphere. However, soil organic carbon (SOC) has been reduced in many areas, while an increase in atmospheric CO₂ has been detected. Recent research works have shown that it is likely that past changes in land use history and land management were the main reasons for the loss of carbon rather than higher temperatures and changes of precipitation resulting from climate change. The primary scope of this work was to estimate soil organic carbon stock (CS) variations in Italy during the last three decades and to relate them to land use changes. The study was also aimed at finding relationships between SOC and factors of pedogenesis, namely pedoclimate, morphology, lithology, and land use, but also at verifying the possible bias on SOC estimation caused by the use of data coming from different sources and laboratories. The soil database of Italy was the main source of information in this study. In the national soil database is stored information for 20,702 georeferentiated and dated observations (soil profiles and minipits) analysed for routine soil parameters. Although the observations were collected from different sources, soil description and analysis were similar, because all the sources made reference to the Soil Taxonomy and WRB classification systems, and soil analyses followed the Italian official methods. Besides horizon description and analysis, soil observations had a set of site information including topography, lithology, and land use. The SOC and bulk density referred to the first 50 cm, thus CS was calculated on the basis of the weighted percentage of SOC, rock fragments volume, and bulk density. A set of geographic attributes were considered to spatialize point information, in particular, DEM (100 m) and derived SOTER morphological classification, soil regions (reference scale 1:5,000,000) and soil systems lithological groups (reference scale 1:500,000), soil moisture and temperature regimes (raster maps of 1 km pixel size), land cover (CORINE project, reference scale 1:100,000) at three reference dates: years 1990 and 2000, and an original update to 2008, obtained with field point observations. The interpolation methodology used a multiple linear regression (MLR). CS was the target variable, while predictive variables were the geographic attributes. Basic statistical analysis was performed first, to find the predictive variables statistically related to CS and to verify the bias caused by different laboratories and surveys. After excluding the biased datasets, the best predictors were selected using a step-wise regression method with Akaike Information Criterion (AIC) as selection and stop criterion. The obtained MLR model made use of the following categorical attributes: (i) decade, (ii) land use, (iii) SOTER morphological class, (iv) soil region, (v) soil temperature regime, (vi) soil moisture regime, (vii) soil system lithology, (viii) soil temperature, (ix) soil aridity index (dry days per year), and, (x) elevation. The interaction between decade and land use variables was also considered in the model. Results indicated that CS was highly correlated with the kind of main type of land use (forest, meadow, arable land), soil moisture and temperature regimes, lithology, as well as morphological classes, and decreased notably in the second decade but slightly increased in the third one, passing form 3.32 Pg, to 2.74 Pg and 2.93 Pg respectively. The bias caused by the variables like “laboratory” and “survey source” could be as large as the 190%.