An important aspect in precision agriculture is the representation of in-field spatial variability as a basis for dividing a field into homogenous management zones (MZ) in order to promote optimization of resources application. MZs can serve as the basis for applying spatially- and temporarily-variable rate inputs such as irrigation. This has important economic implications on the sustainable management of resources in water-limited regions. Management decisions often utilize sampling and/or sensors placed within MZs with the expectation that they will capture and represent a zone’s properties. Yet, to optimize sampling distribution several challenges should be addressed, including how to accurately represent the spatial variability in the field while considering cost-effectiveness and how to integrate multivariate big-data from a variety of sources?
Spatial sampling aims to estimate the attributes of spatially distributed phenomena in un-sampled locations by maximizing the probability of capturing spatial variability. Traditionally, decisions concerning the number of samples/sensors and their locations use non-spatial methods that do not address spatial autocorrelation and/or spatial heterogeneity.
A model and tool based on machine learning and spatial statistics have been developed to provide a decision support tool for distributing samples/sensors in a field while accounting for the spatial structure of multiple variables. The tool is based on stratified-based sampling where MZs act as the confining strata, and on assigning weights to MZs based on their degree of spatial autocorrelation, to reduce sampling redundancy. The tool outputs the optimal number of samples and their location and has been tested on a variety of case studies in Israel, Czech Republic, Spain and the U.S. The tool is available in a user-friendly interface for producing dynamic sampling maps to support sustainable precision management in drylands.