Source: UNIV OF WISCONSIN submitted to NRP
USING HYPERSPECTRAL REMOTE SENSING TO DEVELOP DECISION SUPPORT MODELS FOR POTATO NITROGEN MANAGEMENT
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
AFRI COMPETITIVE GRANT
Reporting Frequency
Annual
Accession No.
1022016
Grant No.
2020-68013-30866
Cumulative Award Amt.
$475,000.00
Proposal No.
2019-06671
Multistate No.
(N/A)
Project Start Date
May 1, 2020
Project End Date
Apr 30, 2024
Grant Year
2020
Program Code
[A1102]- Foundational Knowledge of Agricultural Production Systems
Recipient Organization
UNIV OF WISCONSIN
21 N PARK ST STE 6401
MADISON,WI 53715-1218
Performing Department
Horticulture
Non Technical Summary
Our long-term project outcome is to improve environmental stewardship among crop farmers by adopting new nitrogen management tools that maintain or increase profitability. We propose to conduct multi-year field research to integrate cutting-edge hyperspectral remote sensing technology with precision agriculture methods to improve nitrogen management of potato production systems, thus mitigating impacts of agriculture on the environment, because farmers need new tools to apply the right amount of nitrogen at the right time to achieve maximum profitability and sustainability. We have developed the following short-term project objectives: 1. Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-season and end-of-season potato yield and size, also to collect hyperspectral imagery; 2. Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages, and identify the best time of applying remote sensing during the field season; 3. Develop decision support models to predict end-of-season potato yield, size, and profitability; 4. Validate models by conducting field trials on commercial potato farms; 5. Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professionals. We have assembled an interdisciplinary team with expertise in potato production, remote sensing, and economics. We have recruited 3 cooperating potato farmers to test our models on their farms and leaders of 2 growers' associations to serve on our Adivisory Board. This project will establish a research foundation for our outreach programs and help train the next generation in sustainable agricultural production.
Animal Health Component
60%
Research Effort Categories
Basic
30%
Applied
60%
Developmental
10%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
20513101060100%
Knowledge Area
205 - Plant Management Systems;

Subject Of Investigation
1310 - Potato;

Field Of Science
1060 - Biology (whole systems);
Goals / Objectives
To maintain yield and profitability, potato farmers must satisfy the need of the crops for nitrogen. To minimize environmental degradation and decrease the financial risks associated with regulatory and legal uncertainty surrounding nitrate in groundwater, potato farmers need new management tools to help them apply the right amount of nitrogen at the right time throughout the growing season. Commonly used methods of monitoring potato plant nitrogen status are lab-intensive, time-consuming, and sometimes misleading, and there are no publically available tools to predict end-of-season tuber yield and profitability using in-season information.Our long-term project outcome is to improve environmental stewardship among crop farmers by adopting new nitrogen management tools that maintain or increase profitability. The intermediate project outcome is to meld cutting-edge hyperspectral remote sensing technology with precision agriculture methods for nitrogen management to address the ongoing groundwater quality issue faced by the potato industry. To achieve these outcomes, we developed short-term project objectives:Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-season and end-of-season potato yield and size, as well as to collect hyperspectral imagery.Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages and identify the best time of applying remote sensing during the field season.Develop decision support models to predict end-of-season potato yield, size, and profitability.Validate models by conducting field trials on commercial potato farms.Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professionals.
Project Methods
Objective 1) Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-season and end-of-season potato yield and size, as well as to collect hyperspectral imagery-At final harvest, all tubers in rows 6 and 7 of each subplot will be machine harvested to evaluate both tuber yield and the tuber size profile (i.e., tuber yields by size categories 0-113g, 113-170g, 170-284g, 284-367g, 367-567g, >567g).Hyperspectral imagery will be collected following aschedule using the HySpex (Norsk Elektro Optikk, Norway) full-range (400-2500 nm) imaging system (Table. 4) in operation at UW-Madison. The VNIR-1800 camera has 186 spectral bands between 400 and 1000 nm with a spectral resolution of 3.26 nm. The SWIR-384 camera has 288 spectral bands between 953 and 2518 nm with a resolution of 5.45 nm. The HySpex is flown on one of two State of Wisconsin aircraft: Department of Natural Resources Cessna-180 and Department of Transportation Cessna-210. The flight altitude will be ~365 m above the ground, giving a spatial resolution of 0.25 m in the VNIR and 0.5 m in SWIR.Objective 2) Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages, and identify the best time of applying remote sensing during the field season-Objective 2a): Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stagesTwo methods are proposed to achieve objective 2a. The first method directly links hyperspectral reflectance signal with tuber variables via a partial least squaredregression (PLSR) model. Our preliminary results from the 2018 season will be incorporated into the modelling work here. A permutational analysis will be performed by the following steps: (1) the original dataset will be randomly split 70/30 into calibration and validation datasets; (2) the calibration dataset is used to build the PLSR model, whereas the validation dataset is used to validate the PLSR model; and (3) this calibration/validation process will be run for 500 times. Finally, the mean root-mean-square error (RMSE) and coefficient of determination (R2) will be used to evaluate model performance. This method may fail to work, because the remote sensing reflectance signal is mainly from the aboveground canopy, whereas the belowground tuber variables are invisible to remote sensors.Our second method seeks to: (1) use hyperspectral reflectance signal to model aboveground variables with a PLSR model:Objective 2b): Identify the best time of applying remote sensing during the field season-The models of different dates built under objective 2a will be compared based on their R2 and RMSE. The date with the highest model R2 and lowest RMSE will be identified as the best time for applying remote sensing to predicting real-time potato yield and size during the growing season. Analyses of imagery and field data from different dates will guide the selection of sampling and image dates for Year 2 at HARS. Combined analyses from Years 1 and 2 will guide validation of the remote sensing modelling work in a commercial farm setting in Year 3. The resulting maps from all three years will provide the basis for subsequent decision support modelling work. The best date of using remote sensing for in-season real-time potato yield and size prediction will then be generated into management recommendations to potato farmers.Objective 3) Develop decision support models to predict end-of-season potato yield, size, and profitability-Economic Model: The economic model will combine the yield function and tuber size profile function to estimate projected grower returns as a function of applied nitrogen (N), thermal time (T), and the hyperspectral reflectance signals (R). At any given sampling date, the yield function predicts the expected harvested yield using the observed hyperspectral signals, the applied nitrogen (both planned and already applied), and the expected thermal time from the sample date to harvest (Te). At a given sample date, the yield curve is known based on the observed thermal time T and reflectance signal T (solid blue line) without destructive or labor-intensive field sampling. The model then predicts the expected yield curve for a high N application rate (dashed orange line) and a low N application rate (dashed green line) using the long-range forecast for daily temperatures until harvest time (Te). Expected harvested yield is higher with a higher application, but the size profile also shifts as well. Thus, the economic tradeoff is between how much higher the yield is and its value based on size-specific prices relative to the cost of obtaining this higher yield.The decision support tool will focus on making this N recommendation function an easy-to-use package in which the farmer enters pertinent information (size-specific tuber prices, fertilizer prices, variety, location, date, etc.), as well as the reflectance data, and then the tool generates the recommendation based on the user-entered variables and estimated parameters. The initial work will be a spreadsheet tool with the location-specific weather data and forecasts downloaded from Dark Sky API (https://darksky.net/dev). Once the tool is field tested and refined as part of this project, additonal support will be sought from the Wisconsin Potato and Vegetable Growers Association to develop a mobile app.Objective 4) Validate models by conducting field trials on commercial potato farms-In the season of 2022, field trials will be conducted on three commercial farmers' fields. Farm 1 grows Russet Burbank, Farm 2 grows Soraya, and Farm grows Lamoka. On each farm, there will be about 0.4 hectare of the tested variety grown under 100 kg N ha-1, and there will be at least 0.4 hectare of the tested variety grown under standard N rate, which is farm-specific. Three times of in-season sample collection and remote sensing imaging will be coordinated by the farmers. At harvest, tuber yield and tuber size profile will be collected from four replications of 3 meters per N treatment/farm. Field trials on commercial potato farms will be used as validation of the analyses at HARS. We will implement the models developed in Years 1 and 2 on Year 3 for direct validation. Models from Years 1 and 2 will be reformulated into alternative forms as needed, depending on the validation results.Objective 5) Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professional-We will create project website "Using Hyperspectral Remote Sensing to Manage Potato Nitrogen" that will introduce the project, participants' research and outreach activities for the project, and include a link to the spreadsheet with the decision support models. Project research findings will be timely posted on the website and participants' personal research and outreach sites; publish extension bulletins summarizing key research results and making recommendations about best time of the growing season to apply remote sensing to potato N management; and each year present research findings at regional field tours and workshops and highly attended winter commodity schools.

Progress 05/01/20 to 04/30/24

Outputs
Target Audience:Potato and vegetable growers and processors, and allied industry in the state of Wisconsin and in states with similar problems of nitrate leaching and groundwater contamination. Changes/Problems:As indicated before, there's still a general resistance from the growers to AI-driven techniques like remote sensing and computed-assisted modeling. More extension and outreach activities are needed in the future to further our effort. What opportunities for training and professional development has the project provided?2 PhD students have been trained, one has graduated. One postdoc also gained to research opportunities to work on this project. At least four undergraduate students have been trained to conduct or assist with field work, UAV operation, and image processing. How have the results been disseminated to communities of interest?Through extension talks at grower education meetings, field days, and webinars, and extension articles published on grower magazines, newspapers, websites. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? Our long-term project outcome is to improve environmental stewardship among crop farmers by adopting new nitrogen management tools that maintain or increase profitability. The intermediate project outcome is to meld cutting-edge hyperspectral remote sensing technology with precision agriculture methods for nitrogen management to address the ongoing groundwater quality issue faced by the potato industry. To achieve these outcomes, we developed short-term project objectives: Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-seasuson and end-of-season potato yield and size, as well as to collect hyperspectral imagery. - three years of field trials have been completed. This trial has generated three peer-reviewed papers and one review paper. As well as at least 20 extension presentations, two extension YouTube videos, and ten newspaper interviews (some examples of the products have been listed in the previs section). Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages and identify the best time of applying remote sensing during the field season. - three years of hyperspectral images have been collected from each growing season. all images have been processed, all reflectance data have been extracted, and data has been stored in a safe cloud-based environment. This imaging data contributed to a large dataset that included multi-year imagery and enabled us to conduct an ongoing meta-analysis for predicting potato traits in-season and end-of-season. We plan to write at least another two papers about the meta-analysis. Develop decision support models to predict end-of-season potato yield, size, and profitability. - several different models such as the partial least squares regression, random forecast, support vector machine, k-nearest neighbors, have been developed. we found that random forecast is the best-performing model across different varieties, and site-year environments. Validate models by conducting field trials on commercial potato farms. - validation models are being programmed using on-farm data.We have collaboratedwith three farms to validate the models based on their specific environments to see how robust our models were. The growers who we were collaborating with were highly interested in these models and would like to use those models to guide their nitrogen fertilization in the future growing seasons. But, these are the only proactive growers who are willing to adopt new cutting-edge technologies. Some of them have indicated that they have purchased a drone carrying hyperspectral sensors to conduct mapping over their fields this summer (2024). Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professionals. - extension activities have been conducted to communicate with the growers about the findings from this project. I have conducted some one-on-one interviews with the growercommunity and realized that there is still resistance to AI-driven new technologies. Most of the growers are pretty conservative, and they want to see if these new techniques are actually usable and profitable. The conclusion is that we need to do more extension and outreach activities to continue with our efforts for wider adoption of our research work.

Publications

  • Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Alkhaled, A., P.A. Townsend, B.C. Heberlein, W.B. Hills, Y. Wang. 2024. Using hyperspectral remote sensing to develop new insights into precision nitrogen management for potato production. Precision Agriculture.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Alkahled, A., Y. Wang. 2023. Using hyperspectral spectroscopy in sustainable management of potato on irrigated sandy soil. 2023 American Society of Agronomy - Crop Science Society of America Annual Conference.
  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Alkhaled, A., P.A. Townsend, and Y. Wang. A review - Remote sensing for monitoring potato nitrogen status. 2023. American Journal of Potato Research. https://doi.org/10.1007/s12230-022-09898-9
  • Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Use of Aerial Hyperspectral Imaging to Monitor and Predict Potato (Solanum tuberosum L.) Growth, Nitrogen Status, and Yield
  • Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Alkhaled, A. and Y. Wang. 2024. Developing a robust yield prediction model for potatoes using multi-faceted and multi-year data. Agronomy Journal.


Progress 05/01/22 to 04/30/23

Outputs
Target Audience:The Wisconsin and national potato and vegetable growers and processors. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We have provided three extension talks at different grower education meetings to train our growers about how to use hyperspectral imagery in potato nitrogen management. This technology is cutting-edge and innovative, and more education will help to facilitate its wider adoption. How have the results been disseminated to communities of interest?Three different extension activities, such as talks and one-on-one meetings with the stakeholders. What do you plan to do during the next reporting period to accomplish the goals?We will repeat this same effort including publishing research papers and conducting extension activities.

Impacts
What was accomplished under these goals? To achieve these outcomes, we developed short-term project objectives: 1. Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-season and end-of-season potato yield and size, as well as to collect hyperspectral imagery. We have finished the second year of research trial at the UW Hancock Ag Research Station to collect field measurements and hyperspectral imagery.A plot trial was conducted with two potato varieties (Soraya and Russet Burbank) under four nitrogen treatments (with different N application rates and timings). During the growing season, we collected petiole tissues, whole leaves, vine samples, tubers for nitrate-N and total N% testing on a weekly basis from early tuber bulking to late tuber bulking. Hyperspectral imagery was collected on the same days as the field measurements were made. 2. Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages and identify the best time of applying remote sensing during the field season. We have developed different partial least squares regression models as well as machine learning models using spectral data collected at different growth stages of our potato crops. 3. Develop decision support models to predict end-of-season potato yield, size, and profitability. We have validatedthose models using cross-year data from different varieties/different locations. 4. Validate models by conducting field trials on commercial potato farms. We haven't received all data from commercial fields to validate our models. This work is underway. 5. Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professionals. We have presented our research findings from this project at numerous extension meetings/workshops. Research results from this project were shared with the stakeholders, to indicate that hyperspectral imagery could be used in guiding growers about when to apply how much nitrogen fertilizer to different potato varieties at different growth stages.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Wang Y., and A. Alkhaled. 2022. Hyperspectral remote sensing and machine learning for potato nitrogen management. American Society of Agronomy  Crop Science Society of America  Soil Science Society of America Annual Conference. Baltimore, MD. November 6-9.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Alkhaled, A., and Y. Wang. 2022. Applying hyperspectral imaging and machine learning to understanding crop growth and productivity. American Geophysical Union Fall Meeting. Chicago, IL. December 12-16.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Crosby, T.W., and Y. Wang. 2022. Prediction of nitrogen status and yield in potato (Solanum tuberosum s.) using hyperspectral remote sensing and PLSR-based modeling. Potato Association of America Annual Meeting. Missoula, MT. July 17-21.
  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Alkhaled, A., P.A. Townsend, and Y. Wang. 2022. Remote sensing for potato nitrogen management - A Review. American Journal of Potato Research. Review Published: 01 January 2023 Pages: 1 - 14.


Progress 05/01/21 to 04/30/22

Outputs
Target Audience:Wisconsin potato growers, processors, and the allied industry (a total of around 300 people). Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?One Ph.d. student is being trained to conduct the field work and imagy processing work under this grant (co-advised by me and Dr. Phil Townsend), and we are going to hire one more Ph.d. student do conduct the economic analysis research supervised by Dr. Mitchell How have the results been disseminated to communities of interest?Multiple national professional society meetings such as the Potato Association of America Annual Meeting, the American Society of Agronomy, Crop Science Society of America, Soil Science Society of America Annual Conference; Multiple stakeholder conferences such as the Wisconsin Potato and Vegetable Growers Association Annual Conference, the Potato Expo (of North America). What do you plan to do during the next reporting period to accomplish the goals?We will heavily focus on the modelling research during the next reporting period, we hope to establish models that predict the in-season potato nitrogen status and the end-of-season tuber yield with high R2.

Impacts
What was accomplished under these goals? Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-season and end-of-season potato yield and size, as well as to collect hyperspectral imagery. - we have finished the secondyear of field data collection at the UW-Hancock Ag Research Station, and the field data included in-season plant growth parameters including tuber bulking, canopy development, petiole nitrate values, leaf total N, leaf area index (LAI); as well as end-of-season traits including tuber yield, specific gravity, internal defects, external appearance, total N removal by tubers and by vines. Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages and identify the best time of applying remote sensing during the field season. - we have finished the second year of imagery collection over the field trial at the Hancock Research Station. We totally flew eight times over the course of the growing season from plant emergence to tuber maturity, with about a 10-day interval. We ensured that we took field measurements on the same dates as the we collected those hyperspectral imagery. The hyperspectral camera we used had over 400 spectral bands ranging from 400 nm to 2400 nm. Develop decision support models to predict end-of-season potato yield, size, and profitability. - we are using different modeling approaches such as machine learning, biophysical models (partial least square regression (PLSR)), or a combination of both to build the deecision support models; Validate models by conducting field trials on commercial potato farms. - this will be done in the season of 2022. All models abovementioned will be validated by collecting both field measurements and hyperspectral images on three collaborative potato farms near the UW Hancock Ag Research Station. Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professionals. - I, as a state extension specialist, have attended multiple stakeholder conferences to talk about research results from this project, and the talks were well-received; additionally, many growers approached me after the conferences to ask about collaboration opportunities on this subject.

Publications

  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2021 Citation: Yi Wang and Phil Townsend. How can we use remote sensing and machine learning technologies to manage nitrogen for potatoes? 2021 Potato Expo. Virtual.
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2021 Citation: Yi Wang, Trevor Crosby, Phil Townsend. Using hyperspectral remote sensing to predict potato nitrogen status and final yield. 2021 Potato Association of America Annual Meeting. Virtual.
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2021 Citation: Yi Wang. Use of cutting-edge technologies in sustainable nitrogen management of potato production in Wisconsin. 2021 Wisconsin Potato and Vegetable Growers Association Annual Conference. Virtual.


Progress 05/01/20 to 04/30/21

Outputs
Target Audience:-Wisconsin Potato and Vegetable Growers (about 110) -Midwest Food Processors (about 40) -Wisconsin Fresh Market Vegetable Growers (about 30) Changes/Problems:The summer of 2021 will be repetition of the first year. What opportunities for training and professional development has the project provided? Nothing Reported How have the results been disseminated to communities of interest?Multiple extension/outreach portals: Weekly newsletters (Wisconsin Vegetable CropUpdates); Videos on a YouTube channel; Trade journals such as the Badger Common'Tater, the Potato Grower magazine. What do you plan to do during the next reporting period to accomplish the goals?Stay active for communication with the audience via multiple extension approaches.

Impacts
What was accomplished under these goals? 1-Conduct plot-scale trials with multiple varieties, nitrogen application timings and rates to measure in-season and end-of-season potato yield and size, as well as to collect hyperspectral imagery. - we finished the first year of the field study. 2-Use hyperspectral imagery to predict real-time belowground tuber yield and size at different growth stages and identify the best time of applying remote sensing during the field season. - we collected hyperspectral images over the course of the entire growing season. Images of the research plots at the UW Hancock Ag Research Station were taken by two hyperspectral cameras mounted on UAVs that cover the full spectrum between 400 nm and 2400 nm. Images were collected on a weekly basis between June 20th and August 15th, 2020, which covered the most critical growing stages of potato crops grown in Wisconsin. We are close to be completed with the image processing. The graduate student is working on developing the models for the prediction. 3-Develop decision support models to predict end-of-season potato yield, size, and profitability. - this will start in the second half of the second year. 4-Validate models by conducting field trials on commercial potato farms. - this will start in the third year of the study. 5-Create and deliver training materials about using remote sensing in potato N management for farmers and other agricultural professionals. - published extension products have been recordedin this report.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Yi Wang, Nanfeng Liu, Phil A. Townsend, and Zhou Zhang. 2021. Using remote sensing and machine learning in potato nitrogen management. Potato Expo. Virtually.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Yi Wang, Trevor W. Crosby, Nanfeng Liu, and Phil A. Townsend. 2021. Using remote sensing and machine learning for nitrogen management of potato production. Wisconsin Potato and Vegetable Growers Association Winter Conference. Virtually.
  • Type: Other Status: Published Year Published: 2020 Citation: Nicole Miller. 2020. Research project employs high-tech tuber tools to assess true nitrogen needs of potato crops. Badger Common'Tater. 72(12): 16-19.