Spatial analysis of weed patterns
Keywords: Spatial analysis, weed patterns, Mead’s test, space-time correlograms, 2-D correlograms, dispersal, Generalized Linear Models, heterogeneity, soil, Taylor’s power law. Weeds in agriculture occur in patches. This thesis is a contribution to the characterization of this patchiness, to its analysis, and to its prediction, and some of its results may be useful for weed management. Spatial patterns of six weed species monitored in contiguous quadrats are characterized, using Mead’s test. Five of the six analysed weed species showed aggregation at several levels of scale. The only wind dispersing species, Taraxacum officinale was random at all scales. Next, 2-D correlograms were used to analyse spatio-temporal behaviour of weed patterns for 15 weed species groups throughout three years. Chenopodium album, C. polyspermum, E. crus-galli and S. nigrum were strongly aggregated and also exhibited the largest incidence and highest maximum weed density of the species studied. 2-D correlograms showed that patterns of C. polyspermum and S. nigrum were stable in location. Patches of one species, E. crus-galli appeared to shift from year to year. The four patchy weed species, C. album, C. polyspermum, E. crus-galli and S. nigrum, showed consistent relations of moderate strength with soil variables (pH, texture fraction or organic matter) over the three years of study using Generalized Linear Models with a Poisson log link. Models with spatially uncorrelated and spatially correlated error terms were compared, using Taylor’s power law (TPL) as a link function, resulting in modest decreases in model significance when the spatial correlation in errors was accounted for, and in a few cases, there were big differences in model significance. Spatial correlation remained in the residuals of the regression, demonstrating that factors other than the selected soil variables also contributed to the spatial correlation in the weeds. Dispersal of weed seeds in fields by harvest and rigid-tine cultivator was studied in continuous maize using a range of plant species as model weeds. The rigid-tine cultivator significantly contributed to the dispersal in the driving direction, most likely by dragging plant material with seeds through the field. Irregularities were found in the tail of the dispersal kernels, probably as a result of deposition of plant debris in the headlands by machinery. Taylor’s power law was used to predict the weed free fraction in the field using spatially implicit weed count data. The general model gave accurate predictions for most weed species, but for some, e.g. E. crus-galli, a species specific model was required to achieve adequate accuracy.
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Doctoral thesis biblioteca |
| Language: | English |
| Subjects: | geostatistics, spatial distribution, spatial statistics, spatial variation, statistical analysis, weed biology, weed control, weeds, geostatistiek, onkruidbestrijding, onkruidbiologie, onkruiden, ruimtelijke statistiek, ruimtelijke variatie, ruimtelijke verdeling, statistische analyse, |
| Online Access: | https://research.wur.nl/en/publications/spatial-analysis-of-weed-patterns |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Keywords: Spatial analysis, weed patterns, Mead’s test, space-time correlograms, 2-D correlograms, dispersal, Generalized Linear Models, heterogeneity, soil, Taylor’s power law. Weeds in agriculture occur in patches. This thesis is a contribution to the characterization of this patchiness, to its analysis, and to its prediction, and some of its results may be useful for weed management. Spatial patterns of six weed species monitored in contiguous quadrats are characterized, using Mead’s test. Five of the six analysed weed species showed aggregation at several levels of scale. The only wind dispersing species, Taraxacum officinale was random at all scales. Next, 2-D correlograms were used to analyse spatio-temporal behaviour of weed patterns for 15 weed species groups throughout three years. Chenopodium album, C. polyspermum, E. crus-galli and S. nigrum were strongly aggregated and also exhibited the largest incidence and highest maximum weed density of the species studied. 2-D correlograms showed that patterns of C. polyspermum and S. nigrum were stable in location. Patches of one species, E. crus-galli appeared to shift from year to year. The four patchy weed species, C. album, C. polyspermum, E. crus-galli and S. nigrum, showed consistent relations of moderate strength with soil variables (pH, texture fraction or organic matter) over the three years of study using Generalized Linear Models with a Poisson log link. Models with spatially uncorrelated and spatially correlated error terms were compared, using Taylor’s power law (TPL) as a link function, resulting in modest decreases in model significance when the spatial correlation in errors was accounted for, and in a few cases, there were big differences in model significance. Spatial correlation remained in the residuals of the regression, demonstrating that factors other than the selected soil variables also contributed to the spatial correlation in the weeds. Dispersal of weed seeds in fields by harvest and rigid-tine cultivator was studied in continuous maize using a range of plant species as model weeds. The rigid-tine cultivator significantly contributed to the dispersal in the driving direction, most likely by dragging plant material with seeds through the field. Irregularities were found in the tail of the dispersal kernels, probably as a result of deposition of plant debris in the headlands by machinery. Taylor’s power law was used to predict the weed free fraction in the field using spatially implicit weed count data. The general model gave accurate predictions for most weed species, but for some, e.g. E. crus-galli, a species specific model was required to achieve adequate accuracy. |
|---|