Progress 04/01/24 to 03/31/25
Outputs
Target Audience:The targeted audience of this project mainly includes not only researchers and students in agricultural engineering, agronomics, and computer science and electrical engineering but also the agricultural industry, especially the irrigation pivot industry. Changes/Problems:We requested the 2nd no cost extension to wrap up the project tasks and finish paper publications. What opportunities for training and professional development has the project provided? Aim 4. Create and implement a series of diverse STEM training and mentorship programs as well as extension and outreach programs to stimulate students' and public's interests of advancing agricultural production with STEM based technology, increase the technology adoption for a more sustainable agricultural production, and prepare the next-generation agricultural workforce with STEM expertise to embrace the Digital Agriculture era. New and updated courses and course modules at the university in STEM and technology areas PI Shi and colleagues developed and have been offering a new course for AGST undergraduate students titled Technologies and Techniques in Digital Agriculture, which utilizes the concepts and technologies from this project. Students learned the data lifecycle from data generation and collection, including public data sources and sensor and machinery data, to IoT-based data gathering, parsing, and visualization, to Python-based data processing, with irrigation scheduling as a primary use case. Co-PI Heeren and colleagues developed and have been offering an Irrigation Field Course, an experiential learning course including education and outreach activities https://ow.ly/qOIX50TMSsN. The technologies introduced in this course primarily related to the technologies involved in project Aims 1, 2, and 3. Other than the visits to center pivot manufacturers, students also get opportunities to use sensors and data loggers to get water depth, experience drone demos, and understand how UAS and AI advance irrigation management. This had 99 reactions on LinkedIn: https://www.linkedin.com/posts/waterforfood_what-does-training-the-next-generation-look-ugcPost-7254492389405286400-OT0Y?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAw_3hgBcXJkEobOuob9n36gIW0MBOQL0go A lecture on machine learning for irrigation management, developed directly from materials from the CPS project, will be developed and delivered in AGST Advanced Irrigation Management in spring 2025. Two capstone projects for undergraduate students in the Electrical and Computer Engineering (ECE) department have been completed. Four senior undergraduate students were involved in each capstone project from early September to May of the following year. The Agricultural Sensor Network (ASN) (in the first year of this CPS project) assists the agricultural community by collecting air temperature and soil moisture data. ASN is a long-range low-power wireless network consisting of sensor nodes, a router node, and a control hub. ASN collects sensor data from the five sensor nodes using ZigBee and LoRa for short- and long-range communication. The router node receives both ZigBee and LoRa communications and transmits the sensor data to a control hub to be displayed on an LCD. The longest communications range can be 8 kilometers under the designed wireless network protocol. Each sensor node is powered by disposable Alkaline batteries allowing for up to 14 days. The router node is powered by rechargeable batteries and employs a solar panel to charge the batteries, which allows low maintenance after its deployment. The control hub displays sensor data on an LCD touchscreen and is powered by a wall outlet. The Data Logger System (DLS) (in the second year of this CPS project) aimed to design a user-friendly data logger system and collect timestamped soil moisture data. The logger was designed to be installed along with moisture sensors in a field. The project provides wireless functionality through a cell network and a web application for viewing and analyzing collected moisture data. Testing was completed by placing the sensors in the soil, collecting moisture samples, and sending them to the web interface. This project was motivated by the problem that commercially available data logging devices are expensive and time-consuming. Much of the complexity comes from required additional modules and subscriptions. This complexity inhibits scalability, which is necessary for practical sensor networks. The design created was a less complex system compared to the commercially available systems. This reduction in complexity was achieved by streamlining the design to only interface with moisture sensors. However, the system still provides useful features such as timestamping and cell network connectivity. The remote environment of an agricultural field constrained the project in several ways, especially with power, operation, and data collection. The developed solution provides a cost-effective and scalable system for networked moisture data loggers. The Printed Circuit Board (PCB) was designed with surface mount components but only utilized one side of the PCB, which greatly increased the size. Utilizing both sides of the PCB for component placement would help shrink the overall system considerably. A custom manufactured waterproof and dust proof enclosure would improve the form factor and operation of the device as well. The removable SIM card should be replaced with a soldered IC. The PCB could also be split up into multiple PCBs and stacked on top of each other to take advantage of the vertical space inside the enclosure. Graduate and undergraduate student mentoring At least six graduate students and six undergraduate students have been mentored to fully or partially work on this project by the PIs. To list a couple here, Eric Wilkening, a former MS student, is now the product manager at AgriFac; Jiating Li, a former PhD student (female), is now an assistant professor at the University of Manitoba, Canada; Chong Yu, a former PhD student (female), is now an assistant professor at the University of Cincinnati. Other educational material on popular media reaching the public An AI for Food Production video Artificial Intelligence for Food Production (https://www.youtube.com/watch?v=A27sHEvZc9c&t=36s) was created with the problem that this project addresses and the technologies included in this project. It has 3.5k impressions and 79 reactions on LinkedIn and is highlighted by ASABE: https://www.linkedin.com/posts/derek-heeren-8209aa58_artificial-intelligence-for-food-production-activity-7117288316034584576-DHgy?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAw_3hgBcXJkEobOuob9n36gIW0MBOQL0go A video of irrigation pump installation was created and primarily advertised for our undergrad programs. It got 2.3k impressions and 74 reactions on LinkedIn: https://www.linkedin.com/posts/derek-heeren-8209aa58_a-huge-thank-you-to-our-industry-partners-activity-7275303968392540160-wcnh?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAw_3hgBcXJkEobOuob9n36gIW0MBOQL0go Outreach and interactions with industry The team has started interacting with the industry (primarily with the pivot industry) since the beginning of the project. The industry continues to express interest in utilizing ML and cyber-physical systems for irrigation, especially the data needed to train ML models. We expect the industry to use our data warehouse and incorporate our ML models into their products. We have approached industry partners about potentially sharing some of their irrigation data (from many crop types across several continents) to be published in an open-access database for ML, and they have been somewhat open to the idea (this would be a significant step forward if the industry was willing to share data openly). At Future of BSE Days, industry partners in the irrigation tech and fertigation tech space expressed the need for capacity development (e.g., online badges) for their employees; we are exploring as a department how we might be able to meet this need. The educational materials developed from this project can be the basis of potential online badges. How have the results been disseminated to communities of interest?We have disseminated the results to the scientific communities, academia, as well as to the public through journal publications, conference presentations, white papers, popular media videos, course materials in the curriculum, and interactions with industry. What do you plan to do during the next reporting period to accomplish the goals?Aim 1. Build an adaptive platform to support heterogeneous in-season crop, environment, and management data generation, gathering, and analysis. If we could be granted another NCE, we will finish the data warehouse/repository and analytical center with multiple years' of heterogeneous datasets, including soil properties, weather variables, remote sensing data, crop health indicators, management practices, and yield responses, as well as the developed models. We plan to finish analyzing and organizing the data, develop the data query and visualization interface, and publish them. This valuable, heterogeneous, and long-term dataset will be freely available for researchers and industry to fill an important gap in the lack of such datasets. People can use it to explore plant and environment interactions, sensing strategies for plant stresses, and model development for irrigation scheduling. Aim 2. Establish edge-cloud joint modeling, decision-making, and feedback control for variable-rate applications. In the next few months, we plan to enhance the ML model's accuracy, robustness, and generalizability by including training datasets with more environmental and field conditions (tied to Aim 1). As for the Federated learning (FL) based edge-cloud joint modeling and decision-making, we are working on a manuscript to publish the findings, which we plan to finish if granted another NCE. We expect this will be an important publication and pioneer work for federated learning in agricultural settings. We also plan to expand and adapt the developed FL algorithm on other types of models such as CNN models. Aim 3. Validate and evaluate the developed CPS-enabled VRT framework on farm testbeds with center pivot fertigation systems in terms of the yield, profit, and total chemical application inputs. We are working on a couple of manuscripts to publish the findings related to the field validation and yield results, which we plan to finish if granted another NCE. Aim 4. Create and implement a series of diverse STEM training and mentorship programs as well as extension and outreach programs to stimulate students' and public's interests of advancing agricultural production with STEM based technology, increase the technology adoption for a more sustainable agricultural production, and prepare the next-generation agricultural workforce with STEM expertise to embrace the Digital Agriculture era. This aim has been completed.
Impacts
What was accomplished under these goals?
Aim 1. Build an adaptive platform to support heterogeneous in-season crop, environment, and management data generation, gathering, and analysis. In the past four years, the project team has generated, gathered, analyzed, and organized a data warehouse with valuable heterogeneous datasets, including soil properties, weather variables, remote sensing data, crop health indicators, management practices, and yield responses. Here is a link to an example dataset (from a field located in eastern Nebraska) in the data platform/warehouse we have been creating https://go.unl.edu/xi9d. Co-PI Heeren has been leading this effort. The overall data warehouse currently has 225 GB and 16,000 files (after removing most of the raw data files) from 5 fields across the precipitation gradient (geographic locations) in Nebraska for 2 to 10 years for each field. We started working on this dataset at the beginning of this project (2021) by generating and collecting new sensor and environmental data and organizing and processing historical data from fields before the project period. Aim 2. Establish edge-cloud joint modeling, decision-making, and feedback control for variable-rate applications. Scalable machine learning (ML) models for irrigation scheduling(edge and/or cloud) - Co-PI Heeren's group has developed state-of-the-art ML models to predict soil water depletion (SWD) and Latest Date to provide dynamic, site-specific irrigation recommendations for maize and soybean. The models incorporated multi-field, multi-year datasets encompassing remote sensing data, weather variables, soil properties, crop health indicators, management practices, and yield responses. Feature selection methods, including domain knowledge-augmented techniques, correlation analysis, and Random Forest feature importance, were used to identify the most influential predictors. The domain knowledge-augmented approach consistently retained agronomically relevant features, ensuring the models' interpretability and generalizability. Several ML algorithms, including XGBoost (XGB), Gradient Boosting, Random Forest, and Decision Tree regressors, were evaluated, with XGB consistently achieving superior performance. On unseen field conditions, the XGB model accurately predicted SWD, achieving (R² = 0.72, RMSE = 22 mm, MAE = 18 mm, correlation = 0.90) for maize and (R² = 0.78, RMSE = 24 mm, MAE = 19 mm, correlation = 0.90) for soybean. Field implementation during the 2024 growing season demonstrated the ML models' practical applicability, with irrigation recommendations aligned closely with those from the well-established Soil Water Balance (SWB) method and the Spatial Evapotranspiration Modeling Interface (SETMI) tool. In the next few months, we plan to enhance the ML model's accuracy, robustness, and generalizability by including training datasets with more environmental and field conditions (tied to Aim 1). Federated learning (FL) based edge-cloud joint modeling and decision making - Co-PI Zhang has been leading the development of Federated learning (FL) based edge-cloud joint modeling and decision making. Federated learning can coordinate multiple clients (farms in agriculture) to train their local models without transferring raw data to a center. These local models are aggregated and updated iteratively to obtain a global model. Our primary goal was to transition data-driven decision-making in agriculture from centralized learning to federated learning to reduce communication costs and preserve data privacy while ensuring modeling accuracy. Leveraging the big dataset generated and the outcomes of the Scalable machine learning (ML) models for irrigation scheduling described in the section above, the FL work in our project focuses on predicting the key variable in advanced irrigation scheduling, soil water depletion (SWD). We have used the corn data in the dataset, which consists of 11,934 samples and 45 features. In Federated Learning (FL), we employ horizontal data partitioning, where each client owns the same set of features but different samples. Clients do not share their data with each other.The final global model was evaluated using an independent test set with the standard performance metrics: R², MAE, MSE, and RMSE. Results showed that FL can obtain accuracy similar to CL even when the data is not centrally stored. In addition, although FL has slightly higher errors than CL, it remains stable across increasing client numbers without worsening, proving that FL can effectively aggregate distributed data and maintain a robust global model. The Global Model of FL is more stable than the Independent Local Model across all metrics. The Independent Local Model experiences significant performance degradation at higher client numbers, likely due to data distribution imbalances or insufficient training samples. FL can effectively mitigate local model fluctuations, improving overall predictive quality. Most importantly, FL does not require any client to transfer the local data to the center, dramatically saving communication costs and preventing raw data from being under security risks. Water and nitrogen stress detection - PI Shi's group has been focusing on developing remote sensing based methods for more accurate water and nitrogen stress detections. This task was done in the previous year. Aim 3. Validate and evaluate the developed CPS-enabled VRT framework on farm testbeds with center pivot fertigation systems in terms of the yield, profit, and total application inputs. Though multiple fields are used to develop the cyber-physical framework and data warehouse, the Variable Rate Irrigation (VRI) Field (53 hectares) located at 41.1648°N, 96.4304°W in eastern Nebraska was identified in this project to validate and evaluate the developed ML models. Data from the 2023 growing season were used in model development and testing, while the field deployment of the ML models was conducted in this field site during the 2024 growing season. The VRI Field was irrigated using a Zimmatic model 8500 center pivot system (Lindsay Corporation, Omaha, NE, USA), equipped with Lindsay's Precision VRI zone-control technology. This system allowed for individual nozzle control along the center pivot lateral, enabling irrigation management tailored to each rectangular plot within the field. In-season nitrogen (Urea Ammonium Nitrate, UAN 32-0-0) application was implemented through variable rate fertigation (VRF) using Agri-Inject (MacRoy G110, Series 882) Reflex Connect system integrated with the irrigation equipment. Fertigation management was conducted during the 2023 and 2024 production years. Anhydrous ammonia was applied pre-plant using a 30' John Deere 2510H high-speed anhydrous applicator and UAN was applied during the season. PlanetScope remote sensing data (NDRE) was integrated with the Holland and Schepers nitrogen recommendation model to calculate the N rate. Prescription maps were generated using Lindsay FieldMAP and uploaded to the VRI system using Lindsay FieldNET. Multiple irrigation treatments were implemented in the study field during the 2023 and 2024 seasons. 2024 had five irrigation treatments: best management practice (BMP), 50% BMP, rainfed, SETMI, and a machine-learning-based Cyber-Physical System (CPS) treatment. While this study primarily focused on the CPS treatment for the implementation of the ML models, the other irrigation treatments, especially the BMP and SETMI, were employed to benchmark the performance of the ML systems. The preliminary 2024 yield results of the BMP, 50% BMP, rainfed, SETMI, and the CPS treatments are shown in the table above. There were no statistically significant differences across treatments based on ANOVA tests. We are working on a couple of manuscripts to publish the findings, which we plan to finish if granted another NCE.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
" Amori, P. N., Heeren, D. M., Shi, Y., Wilkening, E. J., Balboa, G. R., Goncalves, I. Z, & Rudnick, D. R. July 29-31, 2024. Potential of machine learning algorithms for timely and adaptive variable rate irrigation management. ASABE Annual International Meeting, Anaheim, Calif. Oral presentation.
- Type:
Other
Status:
Published
Year Published:
2024
Citation:
" Heeren, D. M., Mohammed, A. T., Wilkening, E. J., Neale, C. M. U., Boldt, A. L., Chandra, A., Amori, P. N., Goncalves, I. Z., Shi, Y., & Balboa, G. R. (2024). Field research report: Results from the ENREEC VRI Field for the 2021, 2022, and 2023 crop seasons. Technical report submitted to the Eastern Nebraska Research, Extension, and Education Center. Available at: https://digitalcommons.unl.edu/biosysengpres/79/
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
" Li, J., Ge, Y., Puntel, L.A., Heeren, D.M., Bai, G., Balboa, G.R., Gamon, J.A., Arkebauer, T.J. and Shi, Y. (2024). Integrating UAV hyperspectral data and radiative transfer model simulation to quantitatively estimate maize leaf and canopy nitrogen content. International Journal of Applied Earth Observation and Geoinformation, 129, 103817.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
" Duan, J., Rudnick, D. R., Proctor, C. A., Heeren, D., Nakabuye, H. N., Katimbo, A., Shi, Y., & de Sousa Ferreira, V. (2024). Estimation of corn nitrogen demand under different irrigation conditions based on UAV multispectral technology. Agricultural Water Management, 304, 109075.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
" Li, J., Ge, Y., Puntel, L.A., Heeren, D.M., Bai, G., Balboa, G.R., Gamon, J.A., Arkebauer, T.J. and Shi, Y. (2025). Devising optimized maize nitrogen stress indices in complex field conditions from UAV hyperspectral imagery. Precision Agriculture, 26(1), 3.
|
Progress 04/01/23 to 03/31/24
Outputs
Target Audience:The targeted audience of this project mainly includes not only researchers and students in agricultural engineering, agronomics, and computer science and electrical engineering but also the agricultural industry, especially the irrigation pivot industry. Changes/Problems:We have the majority of the proposed tasks completed, but for the rest of them, we are applying for an NCE to complete the project aims. What opportunities for training and professional development has the project provided?Aim 4. Create and implement a series of diverse STEM training and mentorship programs as well as extension and outreach programs to stimulate students' and public's interests of advancing agricultural production with STEM based technology, increase the technology adoption for a more sustainable agricultural production, and prepare the next-generation agricultural workforce with STEM expertise to embrace the Digital Agriculture era. New and updated courses and course modules at the university in STEM and technology areas: In Biological Systems Engineering, PI Shi continued offered the Digital Agriculture course for undergraduates included in a lot of the concepts and technologies from the project. In addition, Co-PI Heeren and colleagues developed and have been offering an Irrigation Field Course, an experiential learning course including education and outreach activities https://ow.ly/qOIX50TMSsN. The technologies introduced in this course primarily related to the technologies involved in project Aims 1, 2, and 3. Other than the visits to center pivot manufacturers, students also get opportunities to use sensors and data loggers to get water depth, experience drone demos, and understand how UAS and AI advance irrigation management. This had 99 reactions on LinkedIn: https://www.linkedin.com/posts/waterforfood_what-does-training-the-next-generation-look-ugcPost-7254492389405286400-OT0Y?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAw_3hgBcXJkEobOuob9n36gIW0MBOQL0go Graduate and undergraduate student mentoring - multiple graduate and undergraduate students have been mentored to work on this project in the departments of Biological Systems Engineering, Electrical and Computer Engineering, and Agronomy and Horticulture. This year we have a MS student, Eric Wilkening, graduated and now working at AgriFac, and two PhD students graduate students graduated or graduating soon, Jiating Li and Lin Wang, who secured positions in academia to continue research in digital agriculture, all from in Biological Systems Engineering. An AI for Food Production video Artificial Intelligence for Food Production (https://www.youtube.com/watch?v=A27sHEvZc9c&t=36s) was created with the problem that this project addresses and the technologies included in this project. It has 3.5k impressions and 79 reactions on LinkedIn and is highlighted by ASABE: https://www.linkedin.com/posts/derek-heeren-8209aa58_artificial-intelligence-for-food-production-activity-7117288316034584576-DHgy?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAw_3hgBcXJkEobOuob9n36gIW0MBOQL0go We have continued the interactions with the industry, especially the pivot industry, to seek their interests and suggestions for our project idea. The group had meetings with Lindsay and Vermont. The industry continues to express interest in utilizing ML and cyber-physical systems for irrigation, and the data needed to train ML models. How have the results been disseminated to communities of interest?We have disseminated the results to the scientific communities, academia, as well as to the public through journal publications, conference presentations, white papers, popular media videos, course materials in the curriculum, and interactions with industry. What do you plan to do during the next reporting period to accomplish the goals?Aim 1: We will finish organizing the data for the initial version of the data warehouse, which would complete this aim. Aim 2: We will (1)improve the existing ML model for irrigation scheduling, with alternative output prediction variable(s) and other ML models; (2) finish the federated learning model with the improved ML irrigation schedulingmodel. This would complete this aim. Aim 3: We are ready toperform the field test in the growing season of 2024 in eastern Nebraska to test the ML irrigation scheduling model. The treatments we will evaluate inlcude current management practice, scheduling with SETMI model, and scheudling with the model developed in this project. We will also measure the yield at the end of season. Aim 4: We will continue the student mentoring, curriculum development and improvement, and industry interactions.
Impacts
What was accomplished under these goals?
Aim 1. Build an adaptive platform to support heterogeneous in-season crop, environment, and management data generation, gathering, and analysis. In the past year, we focused on analyzingand organizing a data warehouse with heterogeneous datasets we collected, including soil properties, weather variables, remote sensing data, crop health indicators, management practices, and yield responses, for irrigation management. The datasets include: Soil properties - the soil water was monitored with TDR315 Acclima Sensors, and data were logged using a CR300 datalogger equipped with an embedded RF407 radio. Data transmission was conducted wirelessly using the proprietary PakBus communication protocol and Yagi antennas, operating at 900 MHz. The CR300 datalogger was powered by a solar panel. Data was transmitted to an embedded PC computer at the pivot point. The embedded PC was set up for remote access through Chrome's remote desktop, enabling real-time monitoring. In addition, soil texture (%sand, %clay, %silt); Ksat, AWC, FC, WP were obtained by field samplings. Crop health indicators & remote sensing data - the leaf or canopy temperature was monitored using infrared thermometers (SAP-IP IRT, Dynamax) powered by solar panels. Data from the IRT sensors were transmitted via the Zigbee wireless communication protocol to a central coordinator (model IRT-COR, Dynamax). The coordinator routed data through a USB connection to the embedded computer, accessed through Chrome remote desktop for real-time monitoring. In addition, daily SAVI was derived from regular drone flights and Planet satellite imagery. Drone-based hyperspectral data were also collected for partial field/years to investigate the possibility of better-detecting water and nitrogen stresses. Weather - weather data including daily air temperature (minimum and maximum), relative humidity, wind speed, solar radiation, precipitation, reference evapotranspiration (ETr), growing degree days (GDD), vapor pressure deficit (VPD) are obtained from local weather stations. Management & yield - the planting, irrigation, and fertigation implemented over the season, and the yield data were all recorded manually or imported from the yield monitor. Aim 2. Establish edge-cloud joint modeling, decision-making, and feedback control for variable-rate applications. A major part of our effort in Year 3 focused on developing the scalable machine learning (ML) models for irrigation scheduling led by Co-PI Heeren's group. We have developed an initial version of the state-of-the-art ML models to predict the Latest Date to provide dynamic, site-specific irrigation recommendations for maize and soybean. The models incorporated multi-field, multi-year datasets encompassing remote sensing data, weather variables, soil properties, crop health indicators, management practices, and yield responses. Feature selection methods, including domain knowledge-augmented techniques, correlation analysis, and Random Forest feature importance, were used to identify the most influential predictors. The domain knowledge-augmented approach consistently retained agronomically relevant features, ensuring the models' interpretability and generalizability. We also made important progress on the Federated Learning (FL) based edge-cloud joint modeling and decision making led by Co-PI Zhang's group, using the data we collected so far in Aim 1. As a baseline, we first implementedthe Centralized Learning (CL) framework that the FL framework is compared to, the model training process followed these steps, and the final performance was evaluated using metrics such as R², MAE, MSE, and RMSE. Then we compared these results with the results of theFederated Learning (FL). In the past year, we have employedhorizontal data partitioning, where each client owns the same set of features but different samples. Clients do not share their data with each other.The final global model was evaluated using an independent test set with the standard performance metrics: R², MAE, MSE, and RMSE. After comparing the test performances of the four global models, the best-performing model was selected as the final prediction model. The results we have so far showed that theFL can obtain accuracy similar to CL even when data is not centrally stored. In addition, although FL has slightly higher errors than CL, it remains stable across increasing client numbers without worsening, proving that FL can effectively aggregate distributed data and maintain a robust global model. The Global Model of FL is more stable than the Independent Local Model across all metrics. The Independent Local Model experiences significant performance degradation at higher client numbers, likely due to data distribution imbalances or insufficient training samples. FL can effectively mitigate local model fluctuations, improving overall predictive quality. FL does not require any client to transfer the local data to the center, dramatically saving communication costs and preventing raw data from being under security risks. PI Shi's group has been focusing on developing remote sensing based methods for more accurate water and nitrogen stress detections. Monitoring and estimating crop water and nitrogen stress levels are the basis of any management strategies. To untangle the interactive effects of water and nitrogen deficiency, we investigated the potential of drone-based hyperspectral data, mechanistic radiative transfer models (PROCOSINE and PROSAIL-PRO) together with data-driven ML models (Gaussian Process Regression), and dominant and modified vegetation indices (VIs) on resolving this issue. This research showed the potential of improved estimation accuracy and robustness for leaf and canopy nitrogen contents under varying soil water regimes, and paves the way for the optimization of nitrogen stress indicators by mitigating the effects of soil water variability. Aim 3. Validate and evaluate the developed CPS-enabled VRT framework on farm testbeds with center pivot fertigation systems in terms of the yield, profit, and total application inputs. We have been continuing collecting various data in field, but 2023 was the first year we successfully implement fertigation in field.TheVariable Rate Irrigation (VRI) Field (53 hectares) located at 41.1648°N, 96.4304°W in eastern Nebraska was identified in this project to validate and evaluate the developed ML models. In the past two years, we have made efforts to upgrade theZimmatic model 8500 center pivot system (Lindsay Corporation, Omaha, NE, USA), equipped with Lindsay's Precision VRI zone-control technology, to be able to do fertigation.In-season nitrogen (Urea Ammonium Nitrate, UAN 32-0-0) application was implemented through variable rate fertigation (VRF) using Agri-Inject (MacRoy G110, Series 882) Reflex Connect system integrated with the irrigation equipment. Fertigation management was conducted during the 2023 growing season. Ath the same time, we kept collecting data and used them in model development and testing. Now we are ready for the field test for the coming 2024season with the model develop and the tested fertigation pivot.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
" Li, J., Shi, Y., Ge, Y., Heeren, D., Bai, G., Luck, J., Puntel, L., Balboa, G.R., Gamon, J., & Arkebauer, T. (2023) Advancing the use of UAS-based hyperspectral imaging for nitrogen content estimation and stress detection. The American Society of Agricultural and Biological Engineers (ASABE), 2023 Annual meeting, 9 - 12 July 2021, Omaha, NE.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
" Wilkening, E. J., Heeren, D. M., Shi, Y., Katimbo, A., Amori, P. N., Balboa, G. R., ... & Rudnick, D. R. (2023) Development of a Scalable, Edge-Cloud Computing Based Variable Rate Irrigation Scheduling Framework. The American Society of Agricultural and Biological Engineers (ASABE), 2023 Annual meeting, 9 - 12 July 2021, Omaha, NE.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
" Yu, C., Meng, Z., Zhang, K. & Shi, Y. (2023) Efficient Multi-Layer Stochastic Gradient Descent Algorithm for Federated Learning in Cyber Physical Agriculture Systems. The American Society of Agricultural and Biological Engineers (ASABE), 2023 Annual meeting, 9 - 12 July 2021, Omaha, NE.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2023
Citation:
" Li, J., Wijewardane, N. K., Ge, Y., & Shi, Y. (2023). Improved chlorophyll and water content estimations at leaf level with a hybrid radiative transfer and machine learning model. Computers and Electronics in Agriculture, 206, 107669.
|
Progress 04/01/22 to 03/31/23
Outputs
Target Audience:The targeted audience of this project mainly includes not onlyresearchers and students in agricultural engineering, agronomics, and computer science and electrical engineering but also the agricultural industry, especially the irrigation pivot industry. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?Please see the report for Aim 4 above. How have the results been disseminated to communities of interest?We have disseminated the results to the scientific communities, academia, as well as to the public through journal publications, conference presentations, course materials in the curriculum, and interactions with industry. What do you plan to do during the next reporting period to accomplish the goals?Aim 1. Build an adaptive platform to support heterogeneous in-season crop, environment, and management data generation, gathering, and analysis. We plan to keep the generation, gathering, analysis, and organization of the heterogeneous data in the next growing season. The data and processed data will eventually contribute to the irrigation decision-making model in Aim 2, as well as a data warehouse that will be published. ?Aim 2. Establish edge-cloud joint modeling, decision-making, and feedback control for variable-rate applications. In the next few months, we plan to enhance the ML model's accuracy, robustness, and generalizability by including training datasets with more environmental and field conditions (tied to Aim 1). As for the Federated learning (FL) based edge-cloud joint modeling and decision-making, we are working on a manuscript to publish the findings, which we plan to finish if granted another NCE. We expect this will be an important publication and pioneer work for federated learning in agricultural settings. We also plan to expand and adapt the developed FL algorithm on other types of models such as CNN models. Aim 3. Validate and evaluate the developed CPS-enabled VRT framework on farm testbeds with center pivot fertigation systems in terms of the yield, profit, and total chemical application inputs. Next year, we plan to (1) complete the upgrade of the current pivot system into a fertigation system, and (2) at the same time, we will evaluate the performance of the initial developed ML based irrigation scheduling model developed in Aim 2 using the existing irrigation pivot system. Aim 4. Create and implement a series of diverse STEM training and mentorship programs as well as extension and outreach programs to stimulate students' and public's interests of advancing agricultural production with STEM based technology, increase the technology adoption for a more sustainable agricultural production, and prepare the next-generation agricultural workforce with STEM expertise to embrace the Digital Agriculture era. We plan to keep working on (1) developing STEM education materials inspired by the project ideas and findings, (2) training graduate and undergraduate students, (3) developing outreach materials to disseminate project ideas and findings to the public; and (4) the interactions with industry.
Impacts
What was accomplished under these goals?
Aim 1. Build an adaptive platform to support heterogeneous in-season crop, environment, and management data generation, gathering, and analysis. As planned, we focused on testing and improving the in-house developed IoT-based soil and crop sensing system for soil and crop stress monitoring in experiment plots in Year 2 led by Co-PI Ge, in addition to data collections using commercial sensors conducted by the rest of the team. This in-house sensing system collects RGB images, Soil Water Content, NDVI, PRI, and canopy temperature data. Image-extracted data were transmitted using WiFi as well as LoRa. The hardware development, image labeling, and deep learning model training required a lot of labor and time. From the RGB data we collected, we extracted information including crop type, canopy coverage, weed types and count, and insect types and count. This part of the project is still under continuous improvement. We have utilized deep learning models that run on edge computers to achieve this objective. With the support of Soil Water Content, NDVI, PRI, and Canopy Temperature data, we are trying to develop a machine learning model that can estimate the soil water content using only NDVI, PRI, Canopy Temperature, and Weather data. We also collected some stomata conductance, LAI, and hyperspectral data using commercial sensors. We have been continuing the collection of pre-season soil sampling data, in-season soil moisture sampling using the neutron probe sensor (considered to be the soil moisture ground truth data), in-season management data, UAV sensing with RGB, multispectral, thermal and hyperspectral data throughout the season, plant tissue sampling data throughout the season (complete nutrient levels and water content), LAI meter sampling throughout the season (considered as LAI ground truth), biomass samplings, and end-of-season yield. The data were collected in fields located in Ithaca and Clay Center in Nebraska. We have also been working on organizing the historical data from those fields as well as a couple of fields in west Nebraska. Aim 2. Establish edge-cloud joint modeling, decision-making, and feedback control for variable-rate applications. In Year 2, we focused on developing the initial machine learning based decision-making models for irrigation management led by Co-PI Heeren, and the remote sensing data based water and nitrogen stress detection led by PI Shi. We also made important theoretical progress on the federated learning (FL) algorithms for edge-cloud joint modeling and decision making led by Co-PI Zhang. Specifically, we have developed the first version of ML models to predict the Latest Date that an irrigation has to be implemented. The models are currently developed based on the data we collected in the past season, including the remote sensing data, weather variables, soil properties, crop health indicators, management practices, and yield responses. Both linear regression and Random Forrest ML algorithms were trained and evaluated on their ability to generate irrigation recommendations for maize and soybean. We expect to have preliminary results next year. Monitoring and estimating crop water and nitrogen stress levels are the basis of any management strategies. In addition to the existing stress detection models primarily based on multispectral remote sensing, we have been developing hyperspectral based methods to untangle the interactive effects of water and nitrogen deficiency, by integrating mechanistic radiative transfer models with data-driven ML models. In addition to the machine learning based decision making models that can be used on either the edge or the cloud, the project team made important progress on the federated learning (FL) based edge-cloud joint modeling. The primary goal is to transition data-driven decision-making in agriculture from centralized learning to federated learning to reduce communication costs and preserve data privacy while ensuring modeling accuracy. The theoretical framework of FL has been successfully established and evaluated. The project team had a lot of discussions on the following steps to apply the framework to the specific application of irrigation management. With the steady progress of Aim 1, we expect to have enough data to start the simulation next cycle. Aim 3. Validate and evaluate the developed CPS-enabled VRT framework on farm testbeds with center pivot fertigation systems in terms of the yield, profit, and total application inputs. We are still in the process of developing the CPS-enabled VRT framework in aims 1 and 2, and will not be able to validate the system until the next growing season. However, in Year 2, we have already started preparing the field testbed fertigation system. We identified a field in the university farm in eastern Nebraska equipped with a Lindsay Zimmatic center pivot system and VRI zone-control capability. This system allowed for individual nozzle control along the center pivot lateral, enabling irrigation management tailored to each rectangular plot within the field. We are in the process of upgrading the current irrigation-only system to a fertigation system. Aim 4. Create and implement a series of diverse STEM training and mentorship programs as well as extension and outreach programs to stimulate students' and public's interests of advancing agricultural production with STEM based technology, increase the technology adoption for a more sustainable agricultural production, and prepare the next-generation agricultural workforce with STEM expertise to embrace the Digital Agriculture era. New and updated courses and course modules at the university in STEM and technology areas: In Biological Systems Engineering, PI Shi and colleagues developed and have been offering a new course for AGST undergraduate students titled Technologies and Techniques in Digital Agriculture, which utilizes the concepts and technologies from this project. Students learned the data lifecycle from data generation and collection, including public data sources and sensor and machinery data, to IoT-based data gathering, parsing, and visualization, to Python-based data processing, with irrigation scheduling as a primary use case. In Electrical and Computer Engineering, Co-PI Zhang has been incorporating the project idea into undergraduate capstone projects in the past two years. Four senior undergraduate students were involved in each capstone project from early September to May of the following year. Two projects have been successfully completed: the Agricultural Sensor Network (ASN) assists the agricultural community by collecting air temperature and soil moisture data, and the Data Logger System (DLS) aimed to design a user-friendly data logger system and collect timestamped soil moisture data. Graduate and undergraduate student mentoring - multiple graduate and undergraduate students have been mentored to work on this project in the departments of Biological Systems Engineering, Electrical and Computer Engineering, and Agronomy and Horticulture. Last but not least, we have been keeping the interactions with the industry, especially the pivot industry, to seek their interests and suggestions for our project idea. The industry continues to express interest in utilizing ML and cyber-physical systems for irrigation, and the data needed to train ML models.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Wang, L., Li, J., Zhao, B., Baenziger., P.S., Puntel, L.A., Frels, K., Heeren, D.M., Ge, Y., & Shi, Y. (2022). Improved leaf area index estimation with multimodal UAS-derived plant traits and its application for crop model calibration. 2022 ASABE Annual International Meeting, in Houston, TX.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2022
Citation:
Chamara, N., Islam, M. D. 3, Bai, G. F., Shi, Y., & Ge, Y. (2022). Ag-IoT for crop and environment monitoring: Past, present, and future. Agricultural Systems, 203, 103497.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Chamara, N., Ge, Y., Bai, G., Shi, Y. (2022). Design and implementation of IoT sensor network for soybean soil water and vegetation index sensing and yield estimation. 2022 ASABE Annual International Meeting, in Houston, TX.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Yu, C., Shen, S., Zhang, K, Zhao, H., and Shi, Y., Energy-Aware Device Scheduling for Joint Federated Learning in Edge-assisted Internet of Agriculture Things , Proc. of IEEE Wireless Communications and Networking Conference (WCNC), Austin, TX, April 10-13, 2022.
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2022
Citation:
Wang, Lin, "Improve Yield Prediction with Uas-Based Leaf Area Index Estimation and a Hybrid Machine Learning- and Process-Based Model" (2022). ETD collection for University of Nebraska-Lincoln. AAI29322341.
https://digitalcommons.unl.edu/dissertations/AAI29322341
|
Progress 04/01/21 to 03/31/22
Outputs
Target Audience:The targeted audience of this project reached in Year 1 mainly included researchers and students in agricultural engineering, agronomics, and computer science and electrical engineering. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?
Nothing Reported
How have the results been disseminated to communities of interest?
Nothing Reported
What do you plan to do during the next reporting period to accomplish the goals?Aim 1. Build an adaptive sensing platform to support heterogeneous in-season crop, environment, and management data generation, gathering, and analysis. While continuing the data generation and collections, we plan to focus on developing and improving the corresponding data processing and analytics to prepare the inputs needed in Aim 2 in Year 2. We had already started the data processing and analytics in Year 1, mainly focused on improving LAI estimation by utilizing the UAV imagery. We plan to continue this in Year 2 as LAI is an important input for both irrigation and nitrogen management models. Meanwhile, we will have additional focuses for the data analytics in Year 2 as described below: the current issues with the data center, such as the incompatibility of interfacing with the IoT system, and the challenges of fusing the data in different modality and resolutions. the water and nitrogen stresses disentangle utilizing the hyperspectral and other multi-modal data collected. We currently plan to start with developing a method based on a radiative transfer model (PROSAIL) and machine learning models to estimate plant chlorophyll and water contents. Aim 2. Establish edge-cloud joint modeling, decision-making, and feedback control for variable-rate applications Implement the initial irrigation and nitrogen management models with the data collected in Year 1, and investigate issues to be addressed. Investigate the possibility of a method for combined irrigation and nitrogen management This relates to the Change in Project - the fused fertigation model is more challenging than we expected and needs more time to be developed.We have options on using (1) two models, (2) using one model e.g. the APSIM. With the data collected in different fields and the theoretical simulation of FL in Year 1, investigate the detailed application approaches for FL and the IoT based edge-cloud system on the variable-rate irrigation and nitrogen managements in field settings. The communications environment is envisioned to be a great challenge during this process. If successful, we also plan to optimize the model training process from the perspective of device/node selection. Aim 3. Validate and evaluate the developed CPS-enabled VRT framework on farm testbeds with center pivot fertigation systems in terms of the yield, profit, and total chemical application inputs. For Year 2 we will validate the feasibility of the workflow and operation of the proposed system. We will start with one of the fields we've been collecting the data from, and incorporate irrigation and nitrogen treatments to create different levels of water and nitrogen stresses. We currently consider the field in Ithaca, NE. Data will be collected throughout the season, and feed into the management models we currently identified. The data will also be used by the FL algorithm to improve the modeling generalization and efficiency. Aim 4. Create and implement a series of diverse STEM training and mentorship programs as well as extension and outreach programs to stimulate students' and public's interests of advancing agricultural production with STEM based technology, increase the technology adoption for a more sustainable agricultural production, and prepare the next-generation agricultural workforce with STEM expertise to embrace the Digital Agriculture era. First time implementation of the new course AGST 316 Technologies and Techniques for Digital Agriculture based on the technologies used/developed and results derived from this project. Collaboration with industry (pivot companies) and demonstrate the system to growers.
Impacts
(N/A)
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Wang, L., Li, J., Zhao, B., Baenziger., P.S., Puntel, L.A., Frels, K., Heeren, D.M., Ge, Y., & Shi, Y. (2022). Improved leaf area index estimation with multimodal UAS-derived plant traits and its application for crop model calibration. 2022 ASABE Annual International Meeting, in Houston, TX.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Chamara, N., Ge, Y., Bai, G., Shi, Y. (2022). Design and implementation of IoT sensor network for soybean soil water and vegetation index sensing and yield estimation. 2022 ASABE Annual International Meeting, in Houston, TX.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Yu, C., Shen, S., Zhang, K., Zhao, H., & Shi, Y. (2022, April). Energy-Aware Device Scheduling for Joint Federated Learning in Edge-assisted Internet of Agriculture Things. In�2022 IEEE Wireless Communications and Networking Conference (WCNC)�(pp. 1140-1145). IEEE.
|
|