Update:
Objective 1. Experiments to silence the Lac2 cuticle gene of stemborer larvae with RNAi injected into larvae were completed in FY18 and FY19. Preliminary results with dsRNA-covered artificial diet indicate that silencing Lac2 both delays larval pupation and kills adult beetles; however, it is not known whether such effects can be repeated in developing soybean. This summer we conducted greenhouse studies using mesh exclusion cages infested with Dectes stem borer collected from soybean and giant ragweed near Ashland bottoms. Transgenic-soybean seeds developed in Dr. Harold Trick’s lab at KSU were grown in the greenhouse in large pots and were infested with 8 mating pairs of Dectes from field collections over the course of 3 weeks. Ovipositional scars were recorded two weeks post infestation and in late-reproductive stages plants were dissected and number of developing Dectes larvae were counted. Results from the greenhouse infestations were inconclusive. There was significant Dectes ovipositon pressure across several of the cages, but heat and other environmental stressors likely contributed to the low number of tunnels observed in the artificial infestations. The trial will continue but under field conditions using natural populations, as this will give us the best indication as to whether the RNAi event will show promise in controlling this annual pest.
Objective 2. Soybean podworm (Helicoverpa zea) trapping data was further analyzed to explore potential uses for modeling insect severity given location and time of year within Kanas. The the best model for predicting the total number of moths was the model that included the interaction between all variables as it resulted in the lowest AIC values (AIC: 5549.4). Further analysis revealed that location was not a sufficient variable to model the data alone since there was such diversity in the number of different sites and knowing one site does not allow a model to predict what will happen at any new site. The best model for predicting future values of total moths was the negative binomial regression, which takes into account the hour of the day. However, due to the observation that moths are only active at night, all of the analyses with only nighttime data needed to be repeated to investigate whether the time of the night was still a good predictor of moth activity. For predicting the average number of moths, a stepwise linear function was used to predict and model the average number of moths per hour, location, or at a given temperature. The best model for predicting the average number of moths was using only time of day once again. Looking at only nighttime data, the best model for predicting future total moth counts was a negative binomial model that took into account both time of day and the average temperature since it had the lowest AIC value of 5923.6. After creating a model to predict total moth counts, a stepwise linear function was created to predict average moth counts during the nighttime hours. The best model for predicting average number of moths at night was location and time of night. Therefore, even without the extreme difference between the day and night that we saw before, the time of night still had a significant effect on the number of moths. Further analyses are being conducted and will be included in the final report.
Models to accurately estimate soybean defoliation using machine learning were updated and tested using images collected from transects in naturally infested soybean fields in central Kansas; manuscript has been submitted and awaiting peer review. Various feature extraction architectures and performances comparisons are still being conducted and a manuscript is under preparation. Some examples of these architectures included VGG19, VGG20, ResNet50, and ResNet101, etc. Future applications from this work would include implementing this concept in real-time on a smartphone to get rapid detection and defoliation classification estimates for making in-field application decisions. Obtaining this information in situ could prove valuable for researchers and soybean growers alike.
In addition, 25 fields were sampled in northeast Kansas to evaluate the presence of soybean gall midge, an emerging pests across much of the Midwest. No records have been detected to date, but agents and growers were trained on how to scout for this pest and encouraged to report to state specialists. Efforts to better understand the distribution of this new pest will continue in the coming years, as this pest has the potential to impact soybean production in the northeast corner of the state.
Objective 3. Decision guides were updated for 2022 and can be found at: http://www.ksre. ksu.edu/bookstore/pubs/Mf743.pdf. Work is ongoing to update the insect pest management information section for the Kansas Soybean Management 2022 publication. New information was added in our soybean pest management web-based decision chemical selection tool in myFields.info (https://www.myfields.info/chemical/selector/search) and discussed during field days, radio programs, newsletters, and via other educational opportunities as appropriate. For a list of all soybean pests and associated management guides on myFields.info, visit http:// myfields.info/pests.
Dectes stem borer continues to expand its range and impact on soybean production across much of central to eastern Kansas. It is imperative that effective management strategies be developed for this pest that are cost effective. Insecticides are not always a viable option because the emergence window for this pest is quite big. Incorporating host plant resistance is a more selective way to reduce Dectes stem borers in commercial fields, since the plant will be producing toxins that will directly affect the developing larvae, which are hard to target using foliar insecticide applications. Dectes stem borer are not the only pest that Kansas soybean farmers must manage. Soybean podworm is another annual pest that affects acres in several regions within the state. Predicting where this pest will be a problem can help soybean growers and consultants manage their time more efficiently. Results from this work will feed into a predictive model to alert registered users within www.myfields.info of potential flights within a given county. In addition, defoliating pests have become more of an issue in recent years; however, it is not known whether 30+ year old thresholds are still relevant for making informed decisions. There is also a need to streamline field assessments by making effective use of new technologies like machine learning. Consequently, we developed a model that is capable of detecting infested leaves that are captured through the use of a drone or small unmanned aircraft system.