Geostatistical analysis of the spatial distribution of soil salinity. Barbara Cafarelli 1 , Alessio Pollice 2 1 Dipartimento di Scienze Economiche, Matematiche e Statistiche - Università di Foggia, Largo Papa Giovanni Paolo II, 1, 71100 Foggia, Italy, b.cafarelli@unifg.it . 2 Dipartimento di Scienze Statistiche - Università di Bari, Via C. Rosalba 53, 1, 70124 Bari, Italy, apollice@dss.uniba.it Abstract: A correct evaluation of the causes and amount of salinity in a soil has an agronomical as well as environmental relevance and is dealt with in the context of precision agriculture. In this paper a geoadditive model is used to analyse the spatial distribution of an indicator of soil salinity and its nonlinear relations with soil physical, chemical and hydraulic features. Keywords: Geoadditive Models, Low Rank Formulation, Sodium Adsorption Ratio 1. Introduction A correct evaluation of the causes and amount of salinity in a soil has an agronomical as well as environmental relevance and is dealt with in the context of precision agriculture, a management strategy based on the use of several sources of information and information technology to support decisions concerning the agricultural practice. As a matter of fact precision agriculture makes often use of methodologies taking into account spatial and temporal variability associated to every aspect of agricultural production processes, to improve cultivation output and environmental quality. The analysis of soil physical, chemical and hydraulic traits has an outstanding role to determine potential causes of its spatial and temporal variation. The adoption of soil management practices and natural resources conservation policies can thus take advantage of relevant spatial statistical methods, which can be helpful in a differential farm management, enabling farmers to calibrate different actions according to cultural practices and soil conditions. In the agronomical context the use of universal kriging to build maps of physical, chemical and textural variables in the presence of other covariates is quite common (Pozdnyakova L.e Zhang R., 1999). Nevertheless such predictor requires the covariates to have a linear effect and such assumption is seldom verified and often violated. In this paper a geoadditive model is used to analyse the spatial distribution of a soil salinity measure, the Sodium Adsorption Ratio (SAR) and its nonlinear relations with some explanatory covariates, soil physical, chemical and hydraulic features, in order to investigate which causes, if natural causes or secondary causes related to human improper soil use, have played a role on high salinity levels of soils and ground waters within the Muravera-Villaputzu (Sardinia, Italy) coastal plain. This area is well known for its citrus groves, experiencing a remarkable decrease in its production in the last few years. Knowledge of the soil salinity spatial variation is crucial for an efficient cultivation management and especially to properly define irrigation features, given that salinity excesses can be weakened by good quality water lixiviation (Pozdnyakova L. e Zhang, 1999). 1