Spatiotemporal Variation in the Weed Distribution Within an Organic Paddy Field: A Time‐Series Hierarchical Bayesian Model With Spatial Autocorrelation

The timing and direction of farming operations appear to influence the spatial structure and growth of weeds. In this study, we attempted to model the spatial distribution of weed growth based on unmanned aerial vehicle (UAV) imagery. We obtained UAV images of an organic paddy field dominated by Monochoria vaginalis at intervals of 2 weeks from June to July and calculated the coverage in each grid (1 m × 1 m). The weeds distributed along the farming operations, suggesting that the spatial autocorrelation in each direction should be considered separately. However, the semivariograms of the two directions (the direction of the farming operation and the direction perpendicular to the farming operation) did not show a statistically significant difference, implying that it is inappropriate to apply global anisotropy to weed distribution at this scale. We then fitted the time‐series spatial distribution data into a hierarchical Bayesian model, incorporating a CAR model to express the underlying spatial structure in weed growth, to analyze the temporal and spatial variation of weed growth. The weed growth rate from late June to early July seemed to be affected by the timing of intertillage weeding and temperature. The spatial autocorrelation in the direction of farming operations was greater than in the other direction. The formation of hotspots seemed to be caused by the heterogeneous distribution of the seedbank and the uneven ground surface level. These results suggested that a hierarchical model incorporating appropriate spatial autocorrelations enables quantitative evaluation of weed growth and effects of weed control during each period.