UTILIZATION OF SOIL-PLANT INTERRELATIONS THROUGH THE USE OF MULTIPLE REGRESSION AND ARTIFICIAL NEURAL NETWORK IN ORDER TO PREDICT SOIL PROPERTIES IN HUNGARIAN SOLONETZIC GRASSLANDS

Soil and plant interrelations are strong enough in semi-natural solonetzic grasslands to permit the use of plant cover as predictor variable for soil salinity, sodicity and alkalinity. Four data sets were analysed which covered 4-7 plant association types, with sample sizes ranging from 20 to 120 and quadrat sizes 0.16 to 20 m(2); and correlation coefficients (R) of the multiple regression equations established between plant cover (independent or predictor variables) and soil (dependent or predicted variables) usually ranged from 0.65 to 0.80. Utilization of neural networks improved the prediction further and provided typically R values of 0.8. Plant cover observations consequently can be used to improve the precision of numerical maps of soil properties on solonetz soils and to delineate risk areas more precisely faster at a lower cost.