A zone-based approach for processing and interpreting variability in multi-temporal yield data sets

The availability of combine yield monitors since the early 1990′s means that long time-series (10+ years) of yield data are now available in many arable production systems. Despite this, yield data and maps are still under-exploited and under-valued by professionals in the agricultural sector. These historical data need to be better considered and analyzed because they are the only audited means by which growers and practitioners can assess the spatio-temporal yield response within a field. When done, time-series of yield maps are mostly processed by classification-based algorithms to generate spatial and temporal yield stability maps or to provide yield or management classes. This work details an alternate segmentation-based methodology to first generate and then characterize contiguous within-field yield zones from historical yield data. It operates on the yield data rather than interpolated yield maps. A seeded region growing algorithm is proposed that enables both the specification of seeds and zone segmentation in a multivariate (multi-temporal yield) attribute space. Novel metrics to assess the yield zoning are proposed that are derived from textural image analysis. The zoning algorithm and metrics were applied to two fields with long time-series (6+ years) of yield data in combinable crops. The two case studies showed that the proposed zone-based approach was effective in delimitating relevant within-field yield zones. The generated zones had differing temporal yield responses between neighbouring zones that were of agronomic significant and interest to the production systems. As this is a first attempt to apply a segmentation algorithm to yield data, areas for future development applications are also proposed.