The AI-based method is much more cost-effective than the car survey and human-based GSV surveys for largescale species surveys. Because of the strong effects of climate changes, species’ habitats and favorable environmental conditions can change rapidly, and the changing climate may accelerate invasive species spread . Based on the AI-based species mapping method and the integration with other ecological models, we can predict the spread of invasive species and monitor the population closely. The prediction can then allow us to allocate our resources better to control the spread of invasive species.Our study developed a novel weed survey method integrating GSV and a deep learning model. The overall accuracy of johnsongrass detection in GSV panoramas could achieve 77.5% with an 85% recall and 73.9% precision. The recall could be improved by adding more training samples of johnsongrass in the early vegetative stage. However, young johnsongrass is difficult to identify by human as well, growing trays and the labor of labeling will increase. The precision could be improved by utilizing different CNN image detection models. Our trained model detected 2,031 images with johnsongrass presence out of 269,489 GSV images in California, Oregon, Washington, and Nevada. The locations of those 2,031 images were used to create a distribution map of johnsongrass along the major roads in these four states.
We explained and gave examples of possible potential applications and further data analysis based on the johnsongrass map we created. We compared the cost of the car survey, human-based GSV survey, and AI-based GSV survey, and the result demonstrated that the AI-based GSV survey could spend much less time and money on larger-scale roadside species mapping. However, our method cannot retrieve images on a specific date for a single location as the current Google API always returns the latest images to us from their database. Our methods can be improved as the GSV database increase the size and provide more options for images from a single location but at different time. Overall, our work presented that an AI-based GSV survey can be cost-effective on roadside invasive species surveys. Future work will focus on expanding the scale of johnsongrass mapping, and our model will be applied to other invasive species to create more large scale distribution maps.Conventional weed management practices that solely depend on intensive use of herbicides are known to cause ecological and health hazards , and have triggered societal demand for alternative weed management strategies . Effective and sustainable weed control is also a top priority for organic agriculture . The National Organic Regulations and Guidelines prescribe the use of preventive measures as a first line of defense against weeds and other crop pests with no chemical weed control. Because of the lack of effective non-chemical weed management strategies, certified organic croplands in the US faces insignificant increases .
One of the fast growing alternative weed management strategies that may fulfill an ecologically desirable pest management alternative is the use of cover crops . Cover cropping systems involve the use of live plants or their residues as surface mulches . Cover crops not only suppress weeds, but may also improve growth and productivity of the subsequent crops . Many authors showed the usefulness of cover crops as a weed management strategy, but most were from cover crop inter-planting with the main crop . Growers are hesitant to use cover crop inter-planting, asbecause of competition for resources and yield reduction of the main crop . This makes the off-season cover cropping rotation a preferred alternative. However, relatively little evidence exists for the weed management potential of offseason cover crops. Limited resources show that off- season cover crops may provide added economic benefits including soil preconditioning, and supply of additional nutrient to the subsequent crop . This research assessed the effectiveness of summer cover cropping systems for weed management in a winter broccoli crop. More specifically, it evaluated the responses of major weed population densities and their respective biomass to two cropping strategies; through a) planting two different cover crops as a summer rotation after which the vegetable crop is planted for growth during the subsequent season and b) planting the primary vegetable crop on a summer fallow treatment. In order to assess the additional effects of summer cover cropping, we examined soil weed seed bank and the time it may take for hand weeding in each cropping treatments. It was proposed that cover crops reduce soil weed seed pressure and the need for supplemental weed control.A three-year field study was conducted from 2007-2009 at the University of California South Coast Research and Extension Center in Irvine, CA on a loamy-sandy soil.
Three summer cropping treatments were employed: 1) French marigold , 2) cowpea , seeded at 56 kg/ha, and 3) a summer dry fallow as the untreated control. Cowpea was chosen because it is a drought hardy legume, resistant to weeds and enhances some beneficial organisms . Marigold was chosen because it is known to control a broad range of nematodes . Each treatment plot was 12 m long x 10.7 m wide . The cover crops were direct-seeded in the last week of June in the center of 14 planting rows of each treatment plot, watered through drip-tubing and grown for three months. The fallow control plots did not receive water during the summer. Each cover crop treatment plot was planted with the same cover crop in each of the three years of study. Plots were separated from each other with a 3 m wide buffer bare ground. The three treatments were replicated four times in a completely randomized design. At the end of the summer cropping period , the cover crops were mowed at the soil line, chopped, and the residues left on the ground. Concurrently, alternate rows of each of the cover crop treatments were incorporated into the soil at about 0.4 m intervals using a hand-pushed rotary tiller in preparation for broccoli transplanting. The fallow plots were not tilled. Plots for cover crop and broccoli planting are shown in Figure 1a. At the beginning of the subsequent cropping season , broccoli seedlings were transplanted in double rows into the tilled strips of the summer cover crop and fallow plots at an inter and intra-row spacing of 13 and 35cm, respectively . Broccoli transplants were drip irrigated and fertilized with emulsified fish meal at 5 gallons/acre rate. Broccoli was chosen because it is a high-value vegetable crop that is sensitive to weeds, insect pests, nematodes , and requires high soil nutrients . All plot treatments were maintained in the same location for all three years of study in order to assess a cumulative effect of cover crops over time.Weed population density was obtained by sampling at 4 , 8 , and 12 weeks after broccoli transplanting. Weed population count was accomplished using a 50 cm x 50 cm quadrat randomly thrown twice within each treatment plot, grow tray then counting each weed species that had emerged within the quadrat. The population density of each weed species within a plot was recorded as the average of the two quadrat counts. Following the early and mid sampling periods, all plots were hand weeded, recording the duration of time required for weeding. Weed dry biomass was determined by clipping the aerial portion of each weed species observed within each quadrat, drying the samples for 7 days at 700C, and then weighing. The total weed dry biomass of each weed species was recorded and averaged for the two quadrat samples taken per plot. All weed species population density and dry biomass data were analyzed using ANOVA and the means separated using the student T-test.Soil samples were collected three times during each of the three trial years; at the time of cover crop planting , at the time of cover crop incorporation and at broccoli harvest . For each treatment plot, a W-shaped pattern was followed to collect twenty soil cores of 10 cm deep each following soil sampling procedures of Forcella et al. . Weed seed populations from each sampling were assessed using a simple greenhouse weed seed germination test. A set of 500 g soil from each of the sampling periods were spread in flats, placed in the greenhouse and kept moist and well drained. The soil was stirred after the 1st two weeks to expose buried seeds to light and trigger germination. Emerged seedlings were counted and removed every two weeks for one month. After one month, the soil was placed in a cold room for 30 days to simulate conditions needed by some weeds for breaking their dormancy and again placed in the greenhouse for one month and germination counted again.
Weed seedlings were identified to species and the number of individuals that had emerged from each sample was recorded and pulled from the flats at regular intervals. Flats were checked regularly at 3-4 day interval for newly emerged seedlings to assure that no plants emerged and died between counting.The most dominant weed species during all years was Portulaca oleracea , accounting for 70-85% of all weed populations. Other weed species wereChenopodium album , Solanum nigrum , Amaranthus species , Malva nicaeensis , Sonchus oleraceus , Convolvulus arvensis , Capsella bursa-pastoris and Erodium cicutarium . Urtica urticaurens and Oxalis corniculata were observed in some plots, but rarely. Data on weed population densities were presented in Table 1.1 , Table 1.2 , and Table 1.3 . Population densities of common purslane at the early sampling of 2007 were significantly lower for the cover crop treatments compared to the summer fallow . At this sample date, which was just before the initial hand weeding the population density of common purslane within the fallow summer plot peaked at 370 plants per m2 Therefore, the cover crops reduced purslane populations to one-fifth and less than one tenth in broccoli that followed either a summer cowpea or marigold cover crops, respectively. All weed population densities following initial hand weeding were generally low for all treatments and did not vary among the cropping treatments. However, the total population density of all weeds combined, mostly accounting for variations in purslane population densities was lower by 5 and 4 times if broccoli followed summer marigold and cowpea, respectively, compared to these on a fallow plot . Cover crop weed population suppression was more prominent against broad leaves than grass weeds. Grass weeds were generally of low densities and were unaffected by the summer cropping treatments of the first year . Weed population densities at mid and harvest time sampling were lower than the early sampling period . The individual weed population densities at the mid and harvest time sampling were not significantly different among cropping treatments , except for higher total broad leaf and the combined all weed species in the fallow compared to both cover crop treatments . Weed population densities for the early sampling of the second year resembled that of the first year , with Portulaca oleracea remaining as the most dominant weed. The effect of cover crop weed suppression was also similar. Accordingly, the population density of Portulaca oleracea at early sampling of the second year was reduced by 3 or 4 times if broccoli was planted after summer marigold or cowpea respectively, compared to the summer fallow . The supplemental hand weeding further reduced weed population densities for 2008 as can be seen from the lower weed population densities during the mid and harvest time sampling . At mid-season and harvest time samplings of 2008, population densities of common purslane were still was significantly lower in the cover crop treatments compared to the fallow . Similar to the 2007 observation, the broadleaved weeds were still more suppressed with cover cropping and hand weeding interactions that the grass weeds. Weed population densities during the third year were generally lower than the previous two years. Common purslane was the most abundant weed for early sampling of 2009, but was reduced by 6 and 12 times in marigold or cowpea treatments, respectively compared to the fallow treatment . Cover crop suppression of Solanum nigrum , Amaranthus species and Erodium cicutarium became significant only for this year. The cover cropping treatment continued to suppress common purslane and Amaranthus species at mid sampling in 2009. The population densities of these weeds at the mid sampling were lower for the summer cover crop than the fallow treatment .