Simulation of soil types in Teramo province (Central Italy) with terrain parameters and remote sensing data

Soil surveys are an essential source of information for land management although a limited budget often reduces the amount of data available. Even if the dataset is limited, geostatistics can provide a valid estimation tool through a weighted moving average interpolation (kriging). Often, however, the spatial variability of soil properties appears smoothed and short range variability is underestimated by this kind of interpolation technique. A more realistic distribution of a given variable on the territory can be obtained through models based on stochastic simulation.The study area was located in Abruzzo region, Central Italy, in the Soil Region 61.3 as defined by the European Soil Bureau, and includes an economically relevant viticulture district (Controguerra “DOCG area of Colline Teramane”). Relationships between soil type distribution and terrain attributes – slope, incoming solar radiation, NDVI, TWI, etc. – were established, and the most significant were used in a multinomial logistic regression to generate simulated maps. These maps, derived from a set of measured point data, auxiliary information from a Digital Terrain Model and Landsat images, were compared with the soil subsystem map 1:250,000, realized by ARSSA. The comparison indicated that the simulated distribution of the soil classes is consistent with the pedological map and fits better with the local morphology.

Research highlights
► Multinomial logistic regression (MLR) was used to produce predicted soil maps.
► Different scenarios were built, and the best compared with a traditional soil map.
► Agreement among maps, but simulations better represent the local morphology.
► MLR produced explicit and objective maps with a measurable level of accuracy.
► MLR allows to decrease costs and subjectivity of soil maps.