Delineation of site-specific management zones using geostatistics and fuzzy clustering analysis

Abstract

The application of site-specific techniques and technologies in precision agriculture requires the identification of contiguous homogenous zones within a field, often referred as management zones (MZ). The delineation of MZ involves some sort of clustering, however there is not a widely accepted method. The application of fuzzy set theory to clustering algorithm has allowed researchers to better account for the continuous variability in natural phenomena. Moreover, the methods based on nonparametric density estimation can detect clusters of unequal size and dispersion. The objective of this paper is to compare different procedures of clustering in order to assist researchers and producers in delineating within- field management zones. One hundred georeferenced measurements of soil, radiometric and crop attributes were collected on a 12-ha durum wheat field in two seasons in south-east Italy. All variables were interpolated on a 1×1 m grid using geostatistical techniques of kriging and cokriging. We compared the following techniques: (1) iterative Self-Organising Data Analysis Technique (ISODATA); (2) fuzzy c-means method; (3) non-parametric density algorithm. All the methods produced consistent results, creating the subdivision of the field into 2 or 3 distinct classes of suitable size for site-specific management. The clusters differed mainly in soil textural properties and crop response. Fuzzy c-means algorithm provides the user with two performance indices to choose the most appropriate number of clustering for creating MZ. Nonparametric density algorithm allows to assess the local variation within each cluster.