Identification of Weeds Based on Hyperspectral Imaging and Machine Learning

Weeds can be major environmental and economic burdens in New Zealand. Traditional methods of weed control including manual and chemical approaches can be time consuming and costly. Some chemical herbicides may have negative environmental and human health impacts. One of the proposed important steps for providing alternatives to these traditional approaches is the automated identification and mapping of weeds. We used hyperspectral imaging data and machine learning to explore the possibility of fast, accurate and automated discrimination of weeds in pastures where ryegrass and clovers are the sown species. Hyperspectral images from two grasses (Setaria pumila [yellow bristle grass] and Stipa arundinacea [wind grass]) and two broad leaf weed species (Ranunculus acris [giant buttercup] and Cirsium arvense [Californian thistle]) were acquired and pre-processed using the standard normal variate method. We trained three classification models, namely partial least squares-discriminant analysis, support vector machine, and Multilayer Perceptron (MLP) using whole plant averaged (Av) spectra and superpixels (Sp) averaged spectra from each weed sample. All three classification models showed repeatable identification of four weeds using both Av and Sp spectra with a range of overall accuracy of 70–100%. However, MLP based on the Sp method produced the most reliable and robust prediction result (89.1% accuracy). Four significant spectral regions were found as highly informative for characterizing the four weed species and could form the basis for a rapid and efficient methodology for identifying weeds in ryegrass/clover pastures.

The leaf-feeding beetle, Cassida rubiginosa, has no impact on Carduus pycnocephalus (slender winged thistle) regardless of physical constraints on plant growth

The leaf-feeding beetle, Cassida rubiginosa, is an oligophagous biocontrol agent capable of feeding on most species in the tribe Cardueae (thistles and knapweeds). The beetle was released in New Zealand in 2007, primarily to control Cirsium arvense (Californian thistle), with the recognition that it had potential to control multiple thistle weeds. The objective of this study was to test the impact of different densities of Cassida rubiginosa larvae (0, 50, 100, or 200 per plant) on the growth and reproductive performance of the annual thistle weed, Carduus pycnocephalus (slender winged thistle). Since the effectiveness of biocontrol agents is often enhanced when plants are stressed, different levels of growth constraint were imposed by growing the weed in different pot sizes (0.5, 1, 5, and 12 litres). We hypothesised that feeding damage by Cassida rubiginosa larvae would have a greater impact on the weed when grown in smaller pots, since root growth would be constrained, and the weed’s ability to compensate for feeding damage would be restricted. Contrary to our hypothesis, pot size had no effect on feeding damage by Cassida rubiginosa on Carduus pycnocephalus. As expected, most measures of plant performance increased with larger pot sizes, including plant height, biomass, and the number of seedheads per plant. The results of this study indicate that Cassida rubiginosa is unlikely to contribute to the control of Carduus pycnocephalus. Additional oligophagous biocontrol agents targeting the rosette stage and seed production should be considered for release in New Zealand.

Spill-over attack by the gall fly, Urophora stylata, on congeners of its target weed, Cirsium vulgare

The gall fly, Urophora stylata, was released in New Zealand in 1998 as a biocontrol agent for the thistle weed, Cirsium vulgare (Scotch thistle). In the summer of 2018, a survey was conducted to assess the field host range of the biocontrol agent in New Zealand. A random selection of 18 pasture populations under sheep and/or beef production, where C. vulgare was present, was surveyed to quantify the attack intensity (gall size relative to seedhead size) on C. vulgare, and the presence of attack on other thistle weeds within the same population. At each location, seedheads were collected from C. vulgare and all other thistle species (Cardueae) present, which included Cirsium arvense (Californian thistle), Cirsium palustre (marsh thistle), Carduus nutans (nodding thistle), and an Arctium species (burdock). In addition to attack on C. vulgare, the gall fly was recorded on C. arvense (at six locations) and C. palustre (at one location). The probability of the presence of attack on C. arvense was positively correlated with the attack intensity on C. vulgare, suggesting that attack on C. arvense is a ‘spill-over effect’ occurring where seedheads of C. vulgare are in limited supply.

Farm systems analysis of two thistles of differing seasonal pasture growth impacts in North Island hill country

Models for infestations of Californian thistle (Cirsium arvense) and variegated thistle (Silybum marianum) were used to modify fortnightly pasture growth forecasts using Agricultural Production Systems simulator software using climate and soil data from a single farm (Tangihanga Station) over four aspects and three slope classes. Modelling using Farmax software was used to estimate profitability using the current farm enterprises with or without either the Californian or variegated thistles. Modelled pasture production, based on field observations of thistle infestation, was similar to estimates using animal intake (from Farmax). Californian thistle reduced pasture production in summer and autumn, while variegated thistle reduced autumn, winter and spring pasture growth. Californian thistle had a much greater overall presence (20%) than variegated thistle (9%). Both types of thistle reduced the potential to finish lambs in summer and reduced ewe wintering numbers, while the presence of Californian thistles also reduced over-wintering cattle numbers, by reducing summer-autumn pasture accumulation. Cost of control for Californian thistle ($233/ha) over two years was higher than for variegated thistle ($184/ha) over four years. Net profitability was reduced by 24% ($87/ha) by the presence of Californian thistle, and by 37% ($135/ha) by the presence of variegated thistle.

Distribution of Puccinia punctiformis in above-ground tissue of Cirsium arvense (Californian thistle)

Cirsium arvense (Californian thistle) is a problematic weed in agricultural systems throughout New Zealand and the rust fungus Puccinia punctiformis is a potential biological control agent for this weed. Puccinia punctiformis can systemically infect thistles but the movement of the pathogen in planta is not fully understood. This research determined the level of infection in planta caused by P. punctiformis at a single time point. The concentration of P. punctiformis DNA in planta was determined to ascertain the location of the fungus within naturally field-infected C. arvense. Quantitative polymerase chain reaction was undertaken on above-ground symptomatic and asymptomatic C. arvense tissue at various locations within leaves (top, middle and bottom) and the main stem. All C. arvense shoots had detectable amounts of P. punctiformis but the concentration was 100× greater in symptomatic compared with asymptomatic shoots. In general, the concentration of fungus progressed up the leaves with a significant effect between locations (P<0.001). Puccinia punctiformis was found in planta but broadscale disease of C. arvense does not occur and the reason for this is unknown.

Impact of the biocontrol beetle, Cassida rubiginosa, on the secondary weed target, marsh thistle (Cirsium palustre)

The folivorous beetle Cassida rubiginosa was introduced to New Zealand to control the weed, Californian thistle (Cirsium arvense). Although Californian thistle is the primary host, many other thistles are accepted hosts. The objective of this study was to test if the beetle can reduce the fitness of marsh thistle (Cirsium palustre). A potted plant experiment was established with four treatments (0, 50, 100, and 200 larvae/plant). Plant growth (width, height, and number of branches) and reproductive performance (number of flowers, seeds, seed weight and percent germination) parameters were measured. No significant differences were found for any of the measured parameters, except percent germination. Higher larval densities (100 and 200) resulted in approximately 10% less germination compared with lower densities (0 and 50). Under these experimental conditions, C. rubiginosa had minimal impact on the performance of marsh thistle. For the beetle to have an impact, it would likely need to attack smaller, non-bolting rosettes, or be combined with additional stressors that might be encountered in a natural field population.

Genetic variation for tolerance to defoliation in Cirsium arvense (Californian thistle)

The pasture weed, Cirsium arvense (Californian thistle), is notorious for its ability to tolerate defoliation by herbivores, mowing, or herbicides, which is facilitated by the growth of adventitious shoots from its extensive clonal underground root system. In an outdoor potted-plant experiment, we examined the tolerance of 36 unique genotypes of C. arvense to defoliation by establishing pairs of clonal replicates that were assigned to a clipped, or unclipped treatment. Three clipping treatments were applied, and the final height, number of shoots, and biomass, were measured to compare the fitness between the clipped and unclipped clones. The majority of genotypes were negatively affected by clipping and showed a reduction in most final fitness measurements. However, some genotypes were equivalent or even greater than their unclipped counterparts indicating a large genetic range for tolerance to defoliation. The mean range in tolerance ratios (clipped/ unclipped) was 0.17 to 1.3 for shoot height, 0.26 to 1.2 for shoot density, and 0.6 to 1.2 for biomass (where a ratio of 1 indicates equivalence to the unclipped state). Since repeated defoliation is recommended for control of this weed, selection for more tolerant genotypes is possible, which may have management implications.

Californian thistle and its variable response to the biological control agent Sclerotinia sclerotiorum

Californian thistle (Cirsium arvense) is a problematic weed particularly in permanent pastures The fungus Sclerotinia sclerotiorum has potential as a bioherbicide to control this weed but its variable efficacy in historical field trials suggest that there are differences in susceptibility to S sclerotiorum within the species To test this hypothesis the responses of 32 New Zealand provenances of C arvense to a foliageapplied myceliumonbarley preparation of S sclerotiorum were compared under common conditions Significant differences between provenances were found supporting the hypothesis that there is variation within C arvense in New Zealand in its susceptibility to S sclerotiorum Further work will examine differences in the efficacy of fungal isolates against different C arvense provenances

Can differences in Cirsium arvense and Rumex obtusifolius densities within pastures be explained by soil parameters

Some organic farming literature suggests that the presence of weeds in pastures can be explained by problems with the soil Sixty 1 m2 patches of pasture with different densities of broadleaved dock (Rumex obtusifolius) were identified across six different paddocks on an organic dairy farm Various soil parameters were measured in each patch to determine if any of these correlated well with the differing weed densities Parameters measured included soil pH moisture compaction and concentrations of 12 nutrients in adjacent perennial ryegrass This process was repeated for Californian thistle (Cirsium arvense) using a different series of 60 patches Significant correlations were found between dock density and pH potassium magnesium and manganese and Californian thistle density was significantly correlated to soil pH and sulphur content However these correlations were generally weak and are not considered good indicators of whether these weed species would thrive within this particular farm

The search for white tip of Californian thistle in New Zealand

Phoma macrostoma is a cosmopolitan fungus pathogenic to a wide range of herbaceous and woody plant species but nonpathogenic to most grasses Recently a strain of P macrostoma pathogenic to Californian thistle (Cirsium arvense) and other broadleaf weeds was registered as a bioherbicide in Canada for use in turf grass and arable cropping systemsThis strain causes a condition known as white tip disease that is characterised by symptoms of yellow to white chlorosis commonly referred to as photobleaching Symptoms characteristic of white tip disease have been observed on Californian thistle in New Zealand but it is uncertain if the causal agent is P macrostoma The aim was to determine if strains of P macrostoma that cause white tip disease and are genetically similar to the Canadian bioherbicide strain are present in New Zealand The presence of such strains would facilitate the registration of this pathogen as a bioherbicide since most safety requirements will have been satisfied by overseas research People in the agricultural industry are encouraged to look for Californian thistle with white tip symptoms and send samples to AgResearch for identification