Share on Facebook
Share on Twitter
Share on Google Plus
Share on Pinterest
Time-series clustering of remote-sensing retrievals for defining management zones in a vineyard

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

Dr. Noa Ohana-Levi
The Remote Sensing Laboratory, Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Israel

Management zones (MZs) are efficient for dealing with spatial variability in agricultural fields and applying site-specific management (SSM). This study proposes an approach for generating MZs using time-series clustering (TSC) to enable SSM and time-specific management (TSM). TSC was applied to time series of remote sensing images in a California vineyard during four growing seasons (2015-2018) using three datasets: evapotranspiration (ET), leaf area index (LAI), and the normalized difference vegetation index (NDVI). MZs maps were generated for each dataset after computing the optimal number of clusters. These maps and the corresponding temporal dynamics characterizing each cluster (MZ) were quantitatively compared. An aggregated MZs map was generated using multivariate clustering. Additionally, time-series decomposition was used to decompose the ET time series of each pixel into three components: long-term trend, seasonality, and remainder. For each time-series component, TSC was applied and MZs maps were generated. The different ET-components MZs maps were compared for spatial similarities. Each ET-component MZs map was also analyzed against six environmental variables (elevation, slope, northness, lithology, topographic wetness index, and soil type) to check for the spatial association between the MZs maps and the local characteristics in the vineyard. Finally, the ET-components maps were used as independent variables against yield (ton ha-1) using analysis of variance (ANOVA) to assess their relations to yield variability. The findings show that the optimal number of clusters was two and that LAI TSC outperformed LAI and NDVI. The MZs maps of NDVI and LAI were nearly identical, while ET showed weaker similarities to NDVI and LAI. The aggregated MZs map was composed of ET and NDVI, with spatial patterns visually similar to a 2016 yield map. For the ET-components MZs maps, the trend and seasonality had a moderate spatial association and were statistically different from the remainder. For the trend and seasonality spatial association models, the most important predictor was soil type, followed by elevation. The ET-remainder MZs map was strongly linked with northness spatial variability. The yield levels corresponding to the two clusters in all TSC were significantly different. These findings enabled spatial quantification of ET time series at different temporal scales that may benefit within-season decision-making regarding the amounts, timing, intervals, and irrigation location. With the increasing availability of higher spatial and temporal resolution satellite imagery, TSC may be further utilized for defining within-field spatial variability and temporal dynamics of various characteristics for SSM and TSM.