INTEGRATING ON-FARM INFORMATION TO OPTIMIZE WATER MANAGEMENT

• Sponsoring Institution: National Institute of Food and Agriculture
• Project Status: COMPLETE
• Funding Source: AFRI COMPETITIVE GRANT
• Reporting Frequency: Annual
• Accession No.: 1011174
• Grant No.: 2017-67012-26125
• Cumulative Award Amt.: $116,931.00
• Proposal No.: 2016-04882
• Project Start Date: Feb 15, 2017
• Project End Date: Feb 14, 2019
• Grant Year: 2017
• Program Code: [A7201]- AFRI Post Doctoral Fellowships

Recipient Organization
OREGON STATE UNIVERSITY
(N/A)
CORVALLIS,OR 97331

Performing Department
AG Biol & Ecol Engineering

Non Technical Summary
This purpose of thisproject isto find new solutions in determining site-specific crop water demand, and to do so utilizing practical and low cost on-farm technologies.Irrigated agriculture constitutes the greatest consumptive water use globally, so that irrigation efficiency measures are an important part of global efforts to best utilize this limited resource. However, greater irrigation efficiency must be achieved while simultaneously maintaining or increasing crop yields and farming profitability. Incremental water use decisions are made at the local level by farmers under many real world constraints; consequently they face significant risks in operating large and complex irrigation systems. These decisions should be supported by reliable information upon which to base operational plans and irrigation scheduling. Implementing precision irrigation effectively depends upon highly resolved estimates of crop water demand so that application rates match demand precisely both in location and timing. Efficient irrigation planning depends on timely, reliable, and site-specific information in order to anticipate crop water demand, irrigate adequately to prevent drought stress, and maximize yield from the available resource.Growers and irrigation specialists currently have many resources at their disposal, including regional and satellite based ET estimates, state and local soil mapping, and scientific irrigation planning software. However, these methods do not provide site-specific and real time measurements of actual crop water demand, and farmers do not have reliable means by which to validate the accuracy and precision of these estimates. For this information to be directly useful in irrigation planning, it should be validated by on site measurements. Reliable, local, and real time information is required to realize the full potential of precision agriculture.This project will develop a method to determine the crop water requirement in real time, utilizing existing and affordable instrumentation, such as on-farm weather stations. Artificial neural networks (ANNs), which are used in many complex systems such as airplane flight control and nonlinear hydrologic analysis, will be developed to coordinate and prioritize the required field data. Trained in conjunction with a research grade eddy covariance (EC) system, the ANN will be able to provide a robust, site specific method to determine evapotranspiraton. This solution is proposed to be directly useful by estimating crop water demand without an ongoing need for research grade instruments, thereby providing a robust and low cost information useful to on-farm water management and efficiency measures.

Animal Health Component

40%

Research Effort Categories

Basic

10%

Applied

40%

Developmental

50%

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
111	0210	2050	50%
111	5360	2020	50%

Knowledge Area
111 - Conservation and Efficient Use of Water;

Subject Of Investigation
0210 - Water resources; 5360 - Drainage and irrigation facilities and systems;

Field Of Science
2020 - Engineering; 2050 - Hydrology;

Keywords

evapotranspiration

irrigation optimization

local data

neural networks

water use efficiency

Goals / Objectives
This purpose of thisproject isto find new solutions in determining site-specific crop water demand, and to do so utilizing practical and low cost on-farm technologies. There are three overarching goals of this post-doctoral fellowship. The first of these concern educational outcomes and professional development of the post-doctoral project director, undergraduate research assistant, and additional students pursuing agricultural engineering and related disciplines. The second goal is to conduct field experimentation and develop computational models that directly address the needs of irrigators and water managers to improve water use efficiency, and to augment the information used in irrigation decision support. The third project goal is to disseminate the results and new methods as directly useful products throgh outreachto producers, agriculture professionals,and researchers.For the purpose of professional development and educational goals, this project includes mentoring and training of the project director by the primary mentor and a supporting mentor. Both mentors provide resources and skills in field experimentation and scientific methods, as well as supporting the PD's professional skills and teaching. The project budget also includes support for an undergraduate research assistant, who will both benefit from the learning experience, exposure to porfessional practice, and by directly contributing to the project outcomes. The PD will also assist in teaching and offer independent course work to undergraduates in the agricultural engineering and science disciplines. Specific objectives supporting this goal are:PD will develop and document the current status of discipline specific knowledgePD will independently coordinate, design, and conduct a field experiment campaign and develop analyses of resulting dataPS and Mentor will collaborate to offer educational opportunities to undergraduates including coursework, seminars, and mentoring.The concrete goals of this project are to conduct field based experiments and to develop computational models that provide crop water use information that is directly useful to farmers, water managers, and irrigation specialists. These efforts will be conducted in coordination with growers and industry professionals, as well as university extension and agriculutral research station staff. The specific models proposed are neural networks, which will be implemented in resolving improved estimates of evapotranspiration and crop water use. These models will utilize experiemental field data and publicly available information. The intending goals of these efforts are to identify the most cost effective, robust, and appropriate methods to determine site-specific crop water demand, to improve the reliability and transparency of ET estimation methods; to enable farmers to obtain actual ET estimates that are site specific at a spatial scale relevant to irrigation decisions; and to allow farmers to obtain affordable ET information rapidly and on demand.PD will coordinate with cooperating growers and industry professionals to conduct field experiemnts. Input from these professionals will guide the project outputs and products during planning phases and execution of safe and well designed field experiments.PD will implement and maintain field experiments on collaborating farms, with support as needed from mentor and undergraduate research assistant.PD will develop neural network models that identify the most critical measurements, evaluate the robustness of on-farm sensor systems, and determine cost effective approaches to determine crop water demand.Progress will be tracked internally by PD and mentor, and regularly documented in a project website and quartertly reports made available to participating partners.When possible, PD will coordinate with other researchers to expand the scope of impact. This may include coordination of experiments at agricultural research stations, university researchers, or industry professionals. The intent of these efforts will be to address the most relevant and up to date methods in the products and solutions posed by the project.Finally, this project will also prepare products and reports that are directly useful to farmers, agricultural professionals, and researchers.The products and research outcomes will be disseminated through outreach at industry and extension events, authorship of publications in peer-reviewed journals, and by presenting results at professional conferences.The study results will also consider the potential impact of neural network ET estimates on publicly available irrigation information networks (such as Agrimet and CIMIS), and provide some initial proposals for next steps to implement this method.Methods, results, and neural network models will be documented and available for download through data-sharing and archiving at Oregon State University.

Project Methods
The project will be completed in four phases, which will allow for the flexible implmentation of the educational, experimental, and outreach goals.The first preparatory phase will consist of literature review, method refinement, field coordination, and the majority of the educational activities. This phase will be completed by spring 2017, and will include offering an undergraduate course which uses methods developed for the project implementation. Students will be asked to review these course materials using standard university evaluation protocol, and this will incorporated into future development of course curriculum. During this first phase, PD will develop a web based platform for sharing progress, regular reporting, and ultimately for distribution of project products. PD will take the mentorship training course offered at Oregon State University for Junior Faculty members. Following this, the PD will hire, coordinate and mentor an undergraduate research assistant. Progress and challenges from this activity will be covered as part of regular weekly meetings. Evaluation will occur through a mentor evaluation form, filled by the undergraduate research assistant.The second phase of the project will implement a campaign to collect the data required for the neural network method. Eddy covariance will provide the control estimate of evapotranspiration (ET) used in the neural network model development. Eddy covariance (EC) systems with sonic anemometers and open path infrared gas analyzers (IRGAs) will be deployed and co-located with existing and supplemental meteorological stations. To ensure representative ET measurements, data will be quality controlled following standard correction procedures and data treatment protocols. The control estimate will be compared against low cost instruments commonly utilized by farmers to measure air temperature, relative humidity, wind speed and wind direction, shortwave radiation, ground heat flux, and other parameters. Experimental field deployments will be documented and photographed and this information posted on the project web page.The project schedule includes an initial one month training period (co-located EC data and meteorological data collected) during the early to late spring of 2017 (warm, wet conditions). This time period is based on a conservative estimate of two weeks from a previous neural network trained with EC data. This timing allows for multiple training strategies and ensemble averaging. Recognizing that this training period is limited to one set of conditions, a second training period is planned for fall 2017 (hot dry season). In addition to providing a validation of the initial solutions found during the first data collection period, this re-evaluation will test modelling strategies under different field conditions. This iterative process provides an additional layer of robustness to the study with multiple opportunities to re-evaluate and ensure project success. Flexible scheduling also simplifies coordination with the Collaborating Producer's schedule, which ultimately depends on weather and other unforeseen circumstances. During the interim period between the two field deployments, additional short term deployments in coordination with existing agricultural weather monitoring networks is anticipated.The third phase of developing the neural network model will occur concurrently with phase two, with model development supported by ongoing field data collection. Artificial Neural Networks (ANNs) are biologically inspired computational models that allow computers to learn and adapt from data and approximate, complex (non-linear) multi-input functions. Due to their power, flexibility, and robustness, ANNs are used in a wide variety of applications ranging from hydrologic analysis to airplane flight control. The structure of an artificial neural network is comprised of nodes ("neurons"), where each node routes one input variable. The network calculates the contribution of each input variable, and uses feedback from a control estimate to train functional relationships.In addition to improved ET estimation, the ANN is more robust than other methods typically used to determine ET, such as the Penman-Monteith (P-M) equation. If one parameter (i.e. input) is missing from the ANN, performance will be diminished, but the method will continue to predict evapotranspiration. This approach allows the development of a flexible, low cost, and networked method appropriate to on-farm monitoring. Supporting objectives for the ANN phase are: a) Determining the appropriate length of a training period; b) determining the optimal ANN structure and method; c) documenting best training strategies; and d) determining the most important input variables. Established ANN methods will be used, which are widely available in various open source computational toolboxes/libraries in programming languages such as Python, as well as Matlab's Neural Network toolbox. Ensemble ANN predictions of fluxes will also be implemented to provide estimates of mean ET and determine uncertainty in the predictions.The last phase of the project will be to disseminate the results and products including the ANN model and method. PD will attend the ASABE conference in July 2017 to present the progress and initial findings of the study. PD will deliver a seminar online and to the faculty of the Water Resources Engineering program at Oregon State University in the 3rd quarter of the project. Extension efforts may include field days or direct outreach to farmers, and will disseminate methods and solicit feedback on the direct application of the ANN method to determine crop water use and measure ET. The project results will be prepared in no less than three manuscripts submitted to refereed journals. The model, data and methods will be documented and archived in an open source format at the project website.

Progress 02/15/17 to 02/14/18

Outputs
Target Audience:This project worked directly with cooperating farmers, with scientists at university, ARS, and Extension, and communicated results at professional conferences (ag. engineering and water resources) and through formal coursework. Cooperating farms included a medium size farming corporation (via company irrigation managers and operation managers) producing annual crops and livestock, local family farms operating hazelnut orchards, and the regional hazelnut growers association. Scientists included researchers and technicians at a plant pathology lab in the Corvallis ARS office, Extension researchers and technicians at the North Willamette Research Station (Aurora). Scientists at corporate farm supply (Crop Production Services) were also reached in this project. Presentations were made at two professional conferences- International Meeting of the Amer. Society of Agricultural and Biological Engineers (ASABE), and at the Amer. Water Resources Association (AWRA). Additional seminars ont his research were delivered at Oregon State University, Stockholm University (Sweden), and Navarinno Enivironmental Observatory (Greece). In addition to mentored undergraduate reserachers, formal coursework and short courses were offered at Oregon State University at both undergraduate and graduate levels. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Training and professional development opportunities for the PD, undergraduate, and graduate students were enabled during the first year of this project. These included undergraduate courses offered by the PD, and on the job training of research assistants. Graduate students were advised and the PD also lectured in graduate courses and offered series of training workshops. The PD also availed himself of opportunities to develop professionally at conferences and industry meetings, contribute to community and regional service work, apply for professional advancement, and develop his network of industry and research professionals. Teaching and training opportunities directly benefitted students and trainees, as well as the PD who practiced teaching skills. The PD delivered an invited seminar on evapotranspiration and agricultural water use for the Department of Physical Geography at Stockholm University. This was followed by a short course for hydrology students on modelling methods in GIS (February 16, 2017). A similar course on modelling in GIS was given at OSU for the course 549 Regional Hydrologic Modeling in 2017, and this course will be repeated in January 2018. The PD co-taught a week long field course (GE 7049 Ecohydrology: a Mediterranean perspective) at the NEO Institute in Navarrino, Greece. Two seminars were presented; the first emphasized the relationships between agriculture, hydrology, and water resource planning. The second covered the physics of evaporation and the field measurement of ET. The PD offered a series of four workshops on programming for engineering and natural resource science. The PD provided initial technical consultation for a pilot project using machine learning techniques to measure crop water use; this was a separate project conducted jointly by OSU Computer Science faculty, Biological and Ecological Engineering faculty, and a private venture, and funded by Oregon BEST. Currently, the PD is assisting Oregon Department of Water Resources as a technical advisor on estimating current crop water use and projecting future needs for state-wide place-based planning guidance. The PD participated in a small workgroup session in December 2018, and continues to review planning documents developed by ODWR staff. The PD also planned and proposed a frost protection study, to be conducted in 2018 after final discussions with the cooperating grower (post grape harvest and processing). The PD is planning ongoing research collaborations in 2018 with USDA researchers at ARS Corvallis, and with OSU scientists at the North Willamette Research Station. Undergraduate research assistants (RAs) were trained and assisted with field work, technical writing, data processing, and software coding. With support from the project, RAs were also able to attend their first professional conference (AWRA) in November 2017. The PD also trained two additional undergraduate students in field methods and experimental design during collaborative research projects. The PD continues to advise one student for research training (credited engineering coursework) as a continuation of her summer field research. The PD advises four undergraduates on the process of applying to graduate programs. A graduate student from Stockholm University is working on the secondary technical objective of using data from existing weather networks to interpolate site specific ET and crop water demand. The PD assisted this student in applying for a Fulbright award for travel and study in the US. Initial work was undertaken on this project during the graduate course in Greece, and the student travelled to the US for three weeks in August 2017. During this trip, the grad student assisted with field experiments and began writing her thesis. The PD meets weekly (remotely) with this graduate student to advise on her thesis and research. The graduate student will complete her degree in April and has been accepted as an intern at Stockholm Environmental Institute, where as part of this position, she will continue work on the network neural network method. How have the results been disseminated to communities of interest? Invited Seminar: The PD gave an invited talk on crop water demand estimation and outlined the neural network project at the OSU WRGP Water Resource Engineering seminar on January 18, 2017. The seminar was titled "Innovations in Water Resource Engineering: Measuring ET with low cost sensors and neural networks". Invited Seminar: On February 16, 2017, the PD gave an invited talk on evapotranspiration and agricultural water use for the Department of Physical Geography at Stockholm University. Conference Presentation and Proceedings: The PD presented initial project results at the ASABE conference in July. The talk was titled "Training Low Cost Sensors to Estimate Site-Specific Evapotranspiration with Neural Networks", and a summary paper was published in the Conference Proceedings (10.13031/aim.201700694). Conference Presentation: The PD presented results at the AWRA conference for a special session on evapotranspiration at the AWRA conference in November 2017. Publication: PD published "Mapping Soil Texture by Electromagnetic Induction: A Case for Regional Data Coordination" in SSSAJ (doi: 10.2136/sssaj2016.12.0432). The paper demonstrates the use of ANNs to improve spatial resolution in mapping soil using electrical conductivity measurements. Publication: PD submitted a manuscript "Computational Efficiency for the Surface Renewal Method" to Atmospheric Measurement Techniques (doi:10.5194/amt-2017-123). The manuscript describes efficient calculation methods for analyzing high speed data in atmospheric flux studies (using the surface renewal method). Positive peer reviews were received, and publication is anticipated in early 2018. Publication: The "Mapping Soil Texture..." publication (SSSAJ) was promoted in CSA News (doi:10.2134/csa2017.62.1107). Media coverage: Collaborative experiment during total solar eclipse received significant national and local media exposure (Eugene Register Guard, Oregon Statesmen Journal, Associated Press, and reprinted in Wall Street Journal). Invited Seminar: As part of interviewing for a faculty position at University of Idaho, the PD gave a research seminar on August 28, 2017. The seminar was titled "Ongoing Research in Precision Agriculture" and covered aspects of the ANN project as well as related research. Publication: The PD co-authored a manuscript on atmospheric boundary layer development based on research conducted in 2015. The paper was published in Environmental Fluid Mechanics and ius titled "A High Resolution Measurement Of The Morning ABL Transition Using Distributed Temperature Sensing And An Unmanned Aircraft System" (doi:10.1007/s10652-017-9569-1). Industry Meetings: In December 2017, the PD presented project results to Oregon hazelnut growers, and solicited feedback on future studies. The meeting promoted adoption of technology, and identified potential cooperating growers for future research and development. Consultation: the PD is assisting Oregon Department of Water Resources as a technical advisor on estimating current crop water use and projecting future needs for state-wide place-based planning guidance. The PD participated in a small workgroup session in December 2018, and continues to review planning documents developed by ODWR staff. What do you plan to do during the next reporting period to accomplish the goals?Remaining project goals concern dissemination of information to communities of interest and publication of findings. The PD is currently preparing three manuscripts based on results of field experiments conducted this field season. The first documents the neural network method used in the project, and evaluates the most cost-effective strategies for applying this method more broadly in on-farm applied monitoring. The second evaluates results of the water flux measured in the irrigated fields, and concerns energy balance closure and measurement error associated with standard research methods. The third will document a strategy for large scale determination of unknown crop coefficients for hazelnuts, and other crops under non-optimal growing conditions. The PD is also schedule to present at two conferences in 2018; the international ASABE meeting (Detroit,MI in July); and Agricultural and Forest Meteorology (Boise, ID in May). The PD is collaborating on preparation of additional manuscripts. The graduate student at Stockholm University who is co-advised by the PD will prepare a report on the application of neural networks to interpret regional weather data to estimate site-specific and crop-specific ET. The undergraduate advised by the PD is preparing a report on errors in ground heat flux measurement and the use of neural networks to reduce error in this aspect of ET measurement. Following the eclipse experiments, two manuscripts are in preparation on the averaging time for flux (relevant to the measurement of crop water demand) and on the surface renewal method of flux measurement. Continuation of field research from this project also will be ongoing in 2018. The PD is coordinating with scientists at USDA-ARS Corvallis and OSU Extension North Willamette Research Station to conduct three experiments that build on results from this project. The first will study the deployment of many low cost weather stations to improve decision support for irrigation and other management practices. This experiment will be conducted in multiple vineyards in the Willamette Valley. Second, the PD is coordinating with Extension scientists to continue measuring evapotranspiration in orchards, and improve understanding of crop water demand variability in new disease resistant varieties of hazelnuts, Finally, the PD is collaborating with Extension scientists to develop a surface renewal based system to measure water usage in nurseries, especially in potted nursery stock. All of these projects are funded by the collaborating partners. An additional study to predict frost using neural networks to guide the use of irrigation based protection is proposed in collaboration with a local vineyard in Wren, Oregon. The PD is scheduled to meet with the vineyard owner/manager to finalize plans for field experiments to be conducted in spring 2018. Ongoing research will continue efforts to generate open-source and low cost methods to use the neural network outputs. Beyond publication in refereed journal, these efforts will include publication of data obtained in this study (via Oregon State's open data repository); prototyping of Arduino based loggers that incorporate trained data algorithms; investigation of satellite based remote sensing data into ANN training methods.

Impacts
What was accomplished under these goals? All goals and objectives laid out for the first year of this project have been completed. The Project Director (PD) conducted a series of field experiments in coordination with two participating growers and Oregon State University (OSU) Extension faculty and staff. Experimental design and field studies were conducted at three irrigated sites (one annual crop, one orchard, and over research trials at Extension facility), with a total of more than seven months of data collected. Two additional field experiments were conducted in collaboration with OSU and USDA-ARS staff at vineyard sites, with a total of seven months of data collected. A large scale field study on turbulent transport processes was conducted during the 2017 eclipse, in conjunction with scientists from OSU, Lawrence Livermore National Laboratory, and UC Davis. Two undergraduate students were employed for nine months; they assisted with laboratory procedures, field experiments, data analysis, and received additional professional and technical training. The PD assisted USDA-ARS staff with calibrating, maintaining, and deploying experimental equipment for collaborative studies. A graduate student from Stockholm University visited for one month, assisted with field research, and received training in field methods and data analysis. Over the course of the first year, the PD disseminating initial findings through publications, professional presentations, training and teaching activities. In addition to continued annual guest lectures at Oregon State and Stockholm University (Sweden), the PD co-taught a course at the NEO Institute (Greece). The PD also offered a series of workshops on applied programming for engineering undergraduates and graduate students at Oregon State (attended by ten students). The PD presented initial results of field experiments at two conferences (ASABE and AWRA), published a conference proceedings paper at ASABE, and gave two additional invited seminars. During the first year of the project, the PD published two peer-reviewed manuscripts (Soil Sci. Soc. Amer. Journal - first author; and Environmental Fluid Mechanics, co-author), with a third (Atmos. Meas. Tech. - first author) in the final stages of editorial review. Three additional manuscripts are in preparation, and the PD has submitted abstracts to present at two conferences in 2018 (ASABE and AgForMet). The PD also presented findings at two meetings with participating growers, and gave a talk for several Willamette Valley hazelnut growers and professionals from farm supplier Crop Production Services to demonstrate potential benefits and applications of the research findings. In collaboration with two Extension research scientists (OSU), the PD plans to extend the field research into the 2018 field season. At the request of Oregon Department of Water Resources staff, the PD is participating in a technical review of resource projection and technical monitoring procedures for a state place based planning program. At each public presentation, USDA-NIFA funding was highlighted as the primary source of material support for this project. Finally, during the first year of the project, the PD pursued professional development and advancement. The PD applied for five positions related to agricultural water management, focusing on tenure track academic positions. In August the PD interviewed at the University of Idaho, and has subsequently accepted a position at UI as Assistant Professor of Precision Agriculture in the Department of Soil and Water Systems, to start in April of 2018.

Publications

• Type: Journal Articles Status: Accepted Year Published: 2018 Citation: Kelley, Jason, and C. W. Higgins. "Computational Efficiency for the Surface Renewal Method." Atmospheric Measurement Technology, 2017 in press, amt-2017 123.
• Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Kelley, Jason, Chad Higgins, Taylor Vagher, and Willow Walker. "Neural Networks and Low Cost Sensors to Estimate Site-Specific Evapotranspiration." In 2017 ASABE Annual International Meeting, p. 1. American Society of Agricultural and Biological Engineers, 2017.
• Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Kelley, Jason, Chad Higgins, Taylor Vagher, and Willow Walker. "Measuring Site Specific ET using Neural Networks." In Evapotranspiration and Crop Water Use: Current Research and Applications, 2017 AWRA Conference Proceedings, American Water Resources Association, November 7, 2017.
• Type: Journal Articles Status: Published Year Published: 2017 Citation: Kelley, Jason, Chad W. Higgins, Markus Pahlow, and Jay Noller. "Mapping Soil Texture by Electromagnetic Induction: A Case for Regional Data Coordination." Soil Science Society of America Journal 81, no. 4 (2017): 923-931.