Time-series clustering of remote sensing retrievals for defining management zones in a vineyard

Management zones (MZs) are efficient for applying site-specific management in agricultural fields. This study proposes an approach for generating MZs using time-series clustering (TSC) to also enable time-specific management. TSC was applied to daily remote sensing retrievals in a California vineyard during four growing seasons (2015–2018) using three datasets: evapotranspiration (ET), leaf area index (LAI), and normalized difference vegetation index (NDVI). Distinct MZs were delineated based on similarities in pixel-level temporal dynamics for each dataset, using dissimilarity index to determine the optimal number of clusters and compare TSC results. The differences between the cluster centers were calculated, along with the ratio between the centers’ differences and the range of each dataset, denoting the degree of difference between MZs centers. Similarity between MZs from each factor was quantified using Cramer’s V and Frechet distances. Finally, an aggregated (multi-factor) MZ map was generated using multivariate clustering. The resulting MZs were compared to a 2016 yield map to determine the significance of differences between means and distribution among MZs. The findings show that LAI TSC achieved the best cluster separation. The NDVI and LAI MZs maps were nearly identical (Cramer’s V of 0.97), while ET showed weaker similarities to NDVI and LAI (0.61 and 0.62, respectively). Similar findings were observed for the Frechet distances. The yield values were found to be significantly different among MZs for all TSC maps. TSC may be further utilized for defining within-field spatial variability and temporal dynamics for precision irrigation practices that account for spatial and temporal variability.