Progress 07/01/23 to 06/30/24

Outputs
Target Audience:Our phase I project, focusing on tracking and quantifying the pollination behavior of honey bees, targeted farmers that grow crops which require managed bee pollination. To develop our deep-tech solution, we partnered with the USDA under a CRADA. This work is particularly relevant to the USDA Agricultural Research Service (ARS) given its mission to ensure agricultural productivity and sustainability. By examining the specific interactions between honey bees and these crops, the study provides crucial insights into optimizing pollination strategies to enhance crop yields. Understanding bee visitation patterns can lead to improved cultivation practices that maximize pollination efficiency. In blueberries, for example, pollination is critical for fruit set and quality. Our findings can inform ARS initiatives aimed at supporting and enhancing bee populations and health, thereby ensuring robust pollination services. This research aligns with the USDA ARS's goals of advancing agricultural science and promoting the health of pollinators, which are vital for the success of diverse agricultural systems. This work will then be commercialized for small and medium sized farms that are looking to reduce the use of managed honeybees and increase the reliance on native bee populations. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We hired a technician under this phase I that is being trained to use our lidars and service them. USDA ARS has trained an intern to work on this project and collect our scouting data. How have the results been disseminated to communities of interest?We have not yet published, however, we are starting to see results which will be used for future papers by the USDA ARS group. We have started sharing our results with potential commercial partners/customers and are moving towards a commercial trial. We have also gained interest from honey makers and apiaries as a result of the phase I outcomes. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? TO 1 We started our field trials with blueberry potted plants, which were placed in a greenhouse with three tables. Each table fit 50 4-gallon potted plants per table and was kept under drip irrigation. Although we planned on blueberry being our main crop for the trials, we encountered challenges due to limited flowering from these young blueberry plants, resulting in low bee activity. Consequently, we adjusted our approach for the Phase I experiment, opting to use African basil plant, as suggested by the USDA. This decision was made to ensure adequate floral resources for bee attraction and activity. To ensure a controlled trial, we took several measures to reduce the number of variables that may affect our results. To prevent any pollinators other than the European honeybee from affecting the trial, we set up our experiment in a temperature-controlled greenhouse at the USDA site. We used a fully enclosed greenhouse to ensure that the honeybees we placed there were the only ones pollinating and sealed off any openings to prevent any from leaving. The greenhouse uses drip irrigation and has temperature controls that we could monitor. We also implemented our own sensors to monitor environmental variables such as temperature, relative humidity and dew point. While the temperature in the greenhouse was set to maintain itself at a constant 80°F, we instead found it fluctuating quite a lot. Data from December 8th, 2023 showed the temperature in the greenhouse fluctuating by almost 20°, and as expected, our lidar's bee activity data correlates with those fluctuations. This was seen throughout our dataset, with the highest activity usually taking place in the afternoon. We believe that the ramp up of activity in the afternoon also has to do with the location of the greenhouse. The greenhouse only has two transparent walls, thereby its orientation causes a significant increase in sunlight entering into the space in the afternoons. The size of the honey bee colony necessary (the colony size to greenhouse size ratio - HGR) was estimated for the 50 by 25 ft greenhouse that we used. The initial colony size was projected based on previous studies as well as expert recommendations. Since we did not find any studies that used honey bees in a greenhouse (greenhouses typically use bumble bees), we based our initial estimate on what is typically used in outdoor blueberry crops, downsized slightly based on the recommendation of the USDA entomologist. That initial estimate turned out to still be too large and eventually we brought it down to only 2 frames. TO 2 To implement a method for PVI, our team optimized our feature engineering algorithms and made necessary adjustments to our ML code to facilitate automated bee detection within a greenhouse environment. A comprehensive data collection protocol for LiDAR-based bee detections was established, configuring our optics and data acquisition systems to capture relevant information. Field trials were then conducted to validate the accuracy and reliability of these automated detection algorithms. This validation process involved comparing automated detections with manual observations. During the experiments, USDA personnel would record their scouter data multiple times each day. Their method for scouting pollination activity included walking a certain predetermined path through the greenhouse rows and manually recording sightings of bees near flowers. They would mark down the number of bees seen and the position where they were spotted. The collected USDA data was filed in a logbook and then shared with our team. We uploaded the data into our database and then normalized it to values between 0 and 1, which was then compared to our pollination visitation index (PVI). For our PVI calculations, we consider other variables that help us exclude false positives while keeping our thresholds sensitive enough to capture a high number of true events. Those include the time of day, temperature, and intensity of light. For example, a positive bee event seen at 3pm will be weighted differently than a hit at 1am. We know from empirical data that bee activity in the middle of the night is not common, albeit not impossible, and so we established a methodology for assigning weights to the different variables at play. As a result, the PVI is a quantitative measure designed to assess bee pollination activity within a certain environment. We use a normalized value between 0-1 so that we accurately compare the pollination activity across the different scenarios, and as a result the PVI offers a comprehensive perspective on pollination dynamics. A higher PVI value indicates a greater frequency of bee visits, reflecting heightened pollination activity. We define the normalized PVI as: PVInorm=(Xi-min(X1:Xn))/(max(X1:Xn)-min(X1:Xn)) Where i refers to the index of an observation interval out of n intervals, beginning at i=0, and Xi is defined as the weighted number of detected bee events during interval i: Xi= NB(Ot)× wT(T)× wh(h),×wt(t)) And where NB is the number of bee detections, Ot is the duration of an observation interval, wT is the weight assigned to temperature, T is the temperature, wh is the weight assigned to humidity, h is the humidity, wt is the weight assigned to time of day, and t is the time of day. Defining PVI=0 involves comprehensive monitoring and data collection during no bloom periods to establish a baseline of non-pollination activity. This process includes systematically observing and recording honeybee behaviors, utilizing automated detection systems to continuously monitor flight activity and flower visitations, and conducting thorough observations when no flowers are in bloom. During a 20-minute scan of our system, if zero bees are detected/sighted we would define this as PVI = 0 (i.e there was no pollination activity, even though there are still bees onsite). We define PVI=1 as the maximum number of honeybee detections recorded by both the USDA partner and our system during a full season of data collection. This threshold may be refined after collecting and analyzing more data across multiple seasons, ensuring a robust and accurate representation of PVI=1. The maximum value will vary between different environments, crops, and seasons. Therefore, as we increase the number of data trials, our max ('1') will become more accurate and we will likely have different max values set for different crops or terrains. By analyzing peak detection data and establishing this upper limit, growers can quantify the maximum pollination potential. When we execute the correlation function in Excel to get the correlation value between our data and the USDA, we found that we had a correlation value of 0.70 (1 being perfect correlation, and 0 being absolutely no correlation). We also checked whether there was a correlation between PVI and temperature. The results from achieving our Phase 1 technical objectives demonstrate the technology's capability for measuring pollination efficiency in real-time. Under this phase I project, we successfully established a robust data collection protocol and developed accurate algorithms for automated bee detection. Moreover, our analysis revealed correlations between bee activity and environmental factors, providing insights into bee movement patterns and foraging behavior within the greenhouse. In conclusion, the Phase I project has demonstrated the technical feasibility of using LiDAR technology to quantify bee pollination activity in a greenhouse environment. Moving forward, we will refine our methodology and extend our research to larger-scale field settings. We anticipate that our approach will yield valuable insights into pollination dynamics and contribute to the broader understanding of ecosystem health and biodiversity conservation. ** Figures that were referenced in this text can be provided.

Publications