An evaluation of alternative statistical models for predicting habitat suitability for weeds

Abstract

AbstractSinapis arvensis (wild mustard) is a competitive weed of oilseed rape in Iran. The spatial distribution of S. arvensis and its habitat suitability zonation in Fars Province, Iran, was studied using frequency ratio (FR) and weights‐of‐evidence (WofE) statistical and probabilistic models in a geographic information system. For this purpose, a dataset was prepared of weed presence/absence in oilseed rape fields across Fars Province and measures of different effective factors, including elevation, distance from roads, distance from rivers, pH, electrical conductivity, mean annual temperature, mean annual rainfall, slope degree, slope aspect, plan curvature and physical properties of the soil were collected from the sites. Boruta machine learning method was used to determine the significance of each variable that was used to map the geographical distribution of wild mustard and the appropriateness of its habitat. Using the area under the curve and receiver operating characteristic, the accuracy of the habitat suitability zonation maps generated from the FR and WofE models was assessed (AUC‐ROC). The results indicated that AUC‐ROC values for the FR and WofE models were 91% and 88% respectively. FR and WofE are good methods for predicting the likelihood of S. arvensis occurring in a oilseed rape field. Finally, the resulting weed distribution map can aid decision makers and managers in identifying areas that require enhanced management in future plans.