Modelling the effect of spatially variable soil properties on the distribution of weeds

Abstract

The patch spraying of weeds is an area of precision agriculture that has had limited uptake. This is in part due to the perceived risks associated with not controlling individual weeds. Nevertheless, the inherent patchiness of weeds makes them ideal targets for site-specific management. We propose using a mechanistic model to identify areas of a field vulnerable to invasion by weeds, allowing the creation of treatment maps that are risk averse. We developed a spatially-explicit mechanistic model of the life-cycle of Alopecurus myosuroides, a particularly problematic weed of cereal crops in the UK. In the model, soil conditions which vary across the field, affect the life-cycle of A. myosuroides. The model was validated using data on the within-field distribution of A. myosuroides on commercial farms and its co-location with soil properties. We demonstrate the important role played by soil properties in determining the within-field distribution of A. myosuroides. We also show that scale-dependent correlations between A. myosuroides and soil properties observed in the field are an emergent property of the modelled dynamics of the A. myosuroides life-cycle. Our model could therefore support effective site-specific management of A. myosuroides within fields by predicting areas that are vulnerable to A. myosuroides. The usefulness of this model in its ability to predict patch locations for A. myosuroides highlights the possibility of using similar models for other species where data are available on the response of the species to various soil properties.

Keywords: Alopecurus myosuroides; Life-cycle; Patch; Soil properties; Spatial model.

Fig. 1

Basic component structure of the A. myosuroides life-cycle model. Processes are shown in italics and components of the A. myosuroides life-cycle are boxed and capitalised. This life-cycle component is based on the model by Moss (1990) and runs in each cell of our spatially explicit model. Note: A. myosuroides seed is not spread by the combine while still on plant and so movements by cultivation only occur in the soil. This may not necessarily be the case for other weed species that retain seed heads until harvest.

Fig. 2

Numerical order of assessment of nearby squares for the natural dispersal of seeds from a plant in the centre square (labelled “1”). Cells with the same number all receive the same proportion of seed from the starting cell. If required, the pattern continues in the same manner expanding outwards.

Fig. 3

Germination counts plotted against hydrothermal time. Grey is low soil organic matter, Yellow is medium soil organic matter. Solid lines show high water input and dashed lines are low water input. The solid black line shows the resulting germination counts from Eq. (3) when parameterised for the seeds used in the experiment. See supplementary material for experimental detail.

Fig. 4

Maps of Harpenden (top row: a–d), Haversham (middle row: e–h) and Redbourn (bottom row: i–l) showing the kriged log seedling counts (first column: a, e and i) and model outputs (columns 2–4: b–d, f–h, and j–l). Each model output shows the average log seedling density in each cell across 300 realisations of the field. The simulations in the second column (b, f, and j) are the output from the model simulations with rotational ploughing as the cultivation type — ploughing every fourth year with tining at <5 cm in the intermediate years. The simulations in the third column (c, g, and k) used 10 cm tining each year, and the simulations in the fourth column (d, h, and l) used <5 cm tining. Colour scales are maintained within columns and are applicable to each cultivation type separately.

Fig. 5

Frequency distribution of scale-dependent correlation coefficients between the simulated number of A. myosuroides seedlings and simulated soil properties used as inputs into the model simulations for the field in Harpenden. The dotted line represents the observed scale-dependent correlation in the field (Metcalfe et al., 2018b). The correlations shown are between A. myosuroides seedlings and the soil properties clay (a–e), soil organic matter (f–j), pH (k–o) and water (p–t) and for each soil property a range of spatial scales are considered ranging from coarse-scale in the first column to fine-scale in the last column: 50+ m (a, f, k, p), 20 m (b, g, l, q), 7.3 m (c, h, m, r), 2.7 m (d, i, n, s), and 1 m (e, j, o, t).

Fig. 6

Frequency distribution of scale-dependent correlation coefficients between the simulated number of A. myosuroides seedlings and simulated soil properties used as inputs into the model simulations for the field in Haversham. The dotted line represents the observed scale-dependent correlation in the field (Metcalfe et al., 2018b). The correlations shown are between A. myosuroides seedlings and the soil properties clay (a–e), soil organic matter (f–j), pH (k–o) and water (p–t) and for each soil property a range of spatial scales are considered ranging from coarse-scale in the first column to fine-scale in the last column: 50+ m (a, f, k, p), 20 m (b, g, l, q), 7.3 m (c, h, m, r), 2.7 m (d, i, n, s), and 1 m (e, j, o, t).