Premium
Prediction Soil Fertilization Maps Using Logistic Modeling and a Geographical Information System
Author(s) -
Elprince Adel M.
Publication year - 2009
Publication title -
soil science society of america journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.836
H-Index - 168
eISSN - 1435-0661
pISSN - 0361-5995
DOI - 10.2136/sssaj2008.0369
Subject(s) - logistic regression , fertilizer , logit , mathematics , human fertilization , statistics , ecology , agronomy , biology
Multivariable fertilizer recommendation (MVFR) models allowed for variable rate fertilizer applications and precision agriculture. The objectives of this study were (i) to use binary logistic modeling to assess the importance of site variables in farms' decisions for implementing optimal N and K fertilization, and (ii) to combine multivariable logistic models in a geographic information system for the prediction of N and K fertilization class maps. These were based on experimental optimal N and K fertilization data at 67 sites and survey data for 22 site variables in an arid date palm ( Phoenix dactylifera L.) region (20,000 ha). Only nine of the 22 site variables were found to be statistically significant in influencing the probability of N and K responses: surface Fe, Mn, and Cu; profile residual NO 3 –N; surface organic matter (OM), sand, and clay; surface extract soil salinity (EC es ), and quantity of irrigation water ( Q iw ). The probability of response ( Y = 1 means success and Y = 0 indicates failure) to the levels of minor, major, and excessive N application were expressed by the logistic models: logit[Pr( Y = 1|N = 0.25,0.5,1|X)] = 0.574 − 0.015 Q iw + 0.768EC es + 0.455Fe − 1.336 Cu, logit[Pr( Y = 1|N = 0.5,1|X)] = 3.274 − 0.011 Q iw − 0.307Mn, and logit[Pr( Y = 1|N = 1| X )] = 4.394 + 1.463EC es − 0.826NO 3 –N, respectively. The corresponding models for K were: logit[Pr( Y = 1|K = 0.25,0.5,1|X)] = 4.424 − 0.104OM − 0.108NO 3 –N, logit[Pr( Y = 1|K = 0.5,1|X)] = −1.189 + 0.015 Q iw , and logit[Pr( Y = 1|K = 1|X)] = 37.582 − 0.560Mn − 0.036Sand − 0.046Clay. These logistic models were cross‐validated and combined in a geographic information system to derive N and K fertilization class maps using kriged‐interpolated data sets of the significant site variables. Logistic modeling could utilize low‐cost data for MVFR model calibration and validation and the production of soil fertilization maps with larger scales.