Binary Logistic Modeling to Spatially Predict the Probability of Soil Fertilizer Application Classes.

Binary logistic models were developed to spatially predict the probability of NK fertilizer application classes in a date palm-arid region (20 000 ha) using soil salinity (EC_e, dS/m), profile residual nitrate (NO₃-N_p, mg/kg), the soil-test values for soil surface Fe_s (mg/kg) and Mn_s (mg/kg), and the total applied quantity of irrigation water (Q_iw, m³/tree). The NK fertilizer application classes were assigned based on a total of 67 field trials implemented at sites of wide range of soil fertility during two growing seasons. The experiments had the same design with 16 factorial combinations of N and K while P was kept constant with a total number of 200 date palms per field trial. The combination of n independent site-variables taken r at a time method was developed to estimate the target number of total regression degrees of freedom. Only six of the 24 site-variables (X) were found to be statistically significant in influencing the probability of NK responses. The probability of response (Y=1 means response and Y=0 no response) to the levels of N-application and major-N-application were expressed by the logistic models: Logit (Y=1|N= 0.5, 1, 2, or 4 |X) = 2.950 – 0.017 Q_iw + 1.066 EC_es + 0.5 Fe_s and logit (Y=1| N = 4 |X) =  -1.583 – 0.132 NO₃-N_p + 0.785 EC_ep, respectively.  The models for K-application and major-K-application were: logit (Y=1|K= 1, 2, or 4 |X) =  2.581 + 0.018 Q_iw - 0.090 NO₃-N_p and logit (Y=1|K = 4|X) =  0.453 + 0.568 EC_es – 0.456 Mn_s, respectively. These logistic models were combined in a geographic information system (GIS) to derive soil NK fertilizer application class map using the data sets of the significant site-variables.