Progress 09/01/23 to 08/31/24
Outputs
Target Audience:Our research initiatives target a diverse audience, including scholars, partners, and farmers. This study's technical advancements and performance evaluations will be primarily communicated through peer-reviewed academic journals, posters, and various report formats. These outcomes can be accessed by the scholars and the attendees of conferences or workshops. Our partners will gain valuable insights from comprehensive reports that bolster collaborative efforts and support our mutual goals. Farmers will find practical tools and solutions in our work, such as the AlfAdvisor web platform (https://alfalfa.d2s.org), which efficiently provides crop information and serves as a reference for harvest decisions. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?1. A PhD graduate student is being trained under the supervision of PI Zhang. The student has been working on alfalfa field data collection and machine learning model development for alfalfa yield and quality prediction. The student has published a paper on peer-reviewed journal, submitted a new manuscript and completed a conference poster. 2. A postdoc is currently working with PI Zhang. The postdoc has been devoted to developing alfalfa yield and quality traits prediction models using multisource satellite remote sensing data. Three papers are published, and one manuscript is under review with the postdoc as the first author. 3 A Master's graduate student was trained under the supervision of Co-PI Digman. She has completed two conference presentations and successfully defended her thesis. 4. A PhD student is being trained under the supervision of Co-PI Mitchell. The graduate student has been working on the economic model development to support the alfalfa harvest decision making. 5.A Ph.D. student is receiving training under the guidance of Co-PI Jung. The student has been trained in both basic and advanced Python programming, as well as in JavaScript, FastAPI, and SQL for full-stack web development. Additionally, the graduate assistant has been engaged in designing cyberinfrastructure and developing the project's full-stack components. How have the results been disseminated to communities of interest?The project results have been disseminated through invited research seminars and talks. Some of the research results have also been incorporated into PI Zhang's class for teaching and student training activities. Some of the research progress and results were presented by PI Zhang's group at the ASABE 2024 annual international meeting in Anaheim, CA. Also, one of the outcomes of our project is a web-based platform accessible to the target audience and the public through https://alfalfa.gdsl.org. The source codes of the developed web platform are available at a GitHub repository (https://github.com/gdslab/AlfAdvisor). This is a public GitHub repository so that anyone can deploy this system on any computing infrastructure. What do you plan to do during the next reporting period to accomplish the goals?Our future plans are listed below: (1) Validate yield and quality prediction model (2) Quantificationally evaluate the relationship between water consumption, carbon uptake, and yield. (3) Develop an all-sky daily water consumption and carbon uptake estimation model for crops using multisource satellite data. (4) Building hybrid model with deep learning and crop growth model to improve the accuracy and transferability /generalization of the proposed method. (5)All models, including those for yield and quality, drying rate, and economics, have been successfully integrated into the web platform, which is now publicly accessible. The objective for the upcoming year is to solicit feedback from collaborators and the target audience to enhance the website's efficiency and user-friendliness. Objecitve 2 and 3 have been completed.
Impacts
What was accomplished under these goals?
1. The following tasks have been accomplished for Objective 1. (1) Alfalfa sample collection and processing Alfalfa samples were collected in both Wisconsin and New York over the growing season in 2024. There were two fields in Wisconsin and four growth periods in each field from May to September. Before each cut, three to four samplings were performed in each field with around three days intervals between two samplings. Seven samples were randomly selected for each sampling in a field, leading to a total of about 188 samples in Wisconsin. A total of 120 field sections from 24 different alfalfa fields in New York were also sampled prior to mowing, over four growth periods. The GPS coordinate of each sample site was recorded. The samples were weighed immediately after cutting and dried for the dry matter yield. Quality traits were measured using lab NIR analysis. (2) Image processing and feature extraction Satellite: Harmonized Landsat Sentinel (HLS) product was downloaded from NASA official website (https://hls.gsfc.nasa.gov/). Although satellite viewing angle correction was conducted in HLS product, the correction accuracy was further evaluated in this project. It is found that the correction performance is limited in some cases (e.g., under greater solar zenith angle). To address this issue, a practical method was proposed to further improve the consistency between Landsat and Sentinel-2 optical satellite data by coupling a physical radiative transfer model and machine learning approaches. UAV: Multispectral UAV images of each field in Wisconsin were collected and processed for geographic correction, radiometric correction, merging and clipping. The spatial resolution of the image is approximately 3 cm. Several vegetation indices such as NDVI, EVI2, CIgreen, and CIrededge, and statistical features were extracted for each field to model alfalfa yield and quality traits. (3) Yield and quality model development and validation Satellite: We have developed empirical models to jointly predict alfalfa yield and quality traits. We investigated the phenology characteristics of alfalfa in Wisconsin using optical and SAR data. We also helped the implementation of the yield and quality prediction model on the website and provided necessary materials to the website-based tool. Model transferability may be a considerable issue if only using empirical modeling technologies to predict alfalfa yield and quality traits. Crop yield usually presents a significant relationship with water consumption and carbon uptake during the growing season, helping us better understand water-carbon-crop nexus. Adding priori knowledge into the yield estimation model may increase the model's transferability. We attempted to accurately derive water consumption and carbon uptake from multisource optical and SAR satellite data and then included the explored relationship into the yield estimation model. UAV: Four machine learning models, namely linear regression (LR), support vector regression (SVR), random forest (RF), and extreme gradient boosting (XGB), were developed using the UAV image features, environmental factors from national weather service and manually recorded to accurately estimate sixteen quality traits of alfalfa. Most traits achieved competitive accuracy, with an R² for DMY reaching 0.921. The importance of each type of feature and the contribution of specific features to all traits were accurately assessed. Overall, increasing the variety of input features enhanced the accuracy of quality trait estimation, with environmental information being particularly significant. Cross-field transferability was also investigated in the three alfalfa fields. (4) The drying rate model development and validation task was completed last year. Nothing new to report this year. 2. The following tasks have been accomplished forObjective 2. The economic model of the farmer's short-horizon alfalfa harvest decision was developed. This model takes outputs from the yield quality model for the 7-day horizon and the forecast from the weather data API. It then runs the drying rate model for each half-day increment over the 7-day horizon. This can be done with and without tedding. The drying window is calculated along with the probability of precipitation in the drying window. Subsequent losses from rain during the drying window and mechanical damage are estimated and the optimal harvest timing in the 7-day horizon is determined. Additional functionality such as calculating the expected value of tedding is also available for implementation in the economic model code. The economic model code was written in python and delivered to Co-PI Jung's lab for implementation. 3. The following tasks have been accomplished for Objective 3: For Objective 3, Co-PI Jung's group has successfully implemented models developed in Objectives 1 and 2. The implemented models are configured web services (https://alfadvisor.gdsl.org) and the following key features were implemented: Drying Rate Model: The drying rate model has been fully integrated into the website. This model calculates the drying rate of farmers' fields for the next seven days by utilizing weather data, including the probability of precipitation, along with user-specified parameters such as initial and target moisture content. Users are also required to specify if tedding is planned, as this affects drying conditions. The model then computes the drying rate and the resulting moisture content of the alfalfa, and the results are visualized on the website through figures that shows moisture content variation over time. Satellite Imagery Download and Preprocessing: The platform allows users to download Sentinel-1 and Harmonized Landsat Sentinel-2 (HLS) satellite imagery for any specified fields from Google Earth Engine (GEE) and the Land Processes Distributed Active Archive Center (LP DAAC), respectively. After downloading, the data undergoes preprocessing to ensure it is properly formatted for use in the yield and quality model developed in Objective 1. Preprocessing involves several key steps: clipping the satellite imagery to the field boundary to reduce data size and expedite processing time, verifying that the data is neither empty nor corrupted (which can occur if the field is located in a part of the satellite imagery tile where no data is available), calculating the vegetation indices required by the yield and quality model, and generating the appropriate data format as specified in the documentation of the yield and quality model. Yield and Quality Model Implementation: Once the data has been preprocessed and prepared for the yield and quality model, the model is executed to generate results. The outcomes of the models include the alfalfa yield and four quality indicators for every pixel (30 m spatial resolution) within the specified field. The results are then processed in two ways: (a) as a pixel-based visualization of yield and quality, offering an overview of field variation, and (b) integrating yield and quality model outputs with the economic model for further analysis. Economic Model Implementation: The economic model integrates outputs from the yield and quality model, weather forecast information (e.g., temperature, precipitation), and additional user inputs such as whether the products will be sold or fed to cattle (producing hay) and whether tedding is planned. Based on this input information, the model applies the appropriate economic functions to forecast conditions over the next seven days. The output includes the net return to the producer and recommendations for optimal harvest timing.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2024
Citation:
Chen, J., Yu, T., Cherney, J.H., Zhang, Z. (2024) Optimal Integration of Optical and SAR Data for Improving Alfalfa Yield and Quality Traits Prediction: New Insights into Satellite-Based Forage Crop Monitoring. Remote Sensing, 16, 734.
- Type:
Journal Articles
Status:
Published
Year Published:
2024
Citation:
Chen, J., and Zhang, Z. (2024) Estimation of all-sky high-resolution gross primary production across different biome types using active microwave satellite images and environmental data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 17, 12969-12982.
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
An improved fusion of Landsat-7/8, Sentinel-2, and Sentinel-1 data for monitoring alfalfa: Implications for crop remote sensing. International Journal of Applied Earth Observation and Geoinformation, 124, 103533.
- Type:
Journal Articles
Status:
Published
Year Published:
2024
Citation:
Yu, T., Zhou, J., Ranjbar, S., Chen, J., Digman, M. F., & Zhang, Z. (2024). Evaluation of the Effect of Sentinel-1 SAR and Environmental Factors in Alfalfa Yield and Quality Estimation. Agronomy, 14(4), 859.
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Chen, J., Zhou, J., Cherney, J. H., and Zhang, Z. Prediction of within-season alfalfa yield and quality traits using fused near real-time PlanetScope, Landsat-8/9, Sentinel-2, and Sentinel-1 satellite data. Precision Agriculture. (Submitted)
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Yu, T., Xu, Y., Zhou, J., & Zhang, Z. Pre-Harvest Estimation and Contribution Analysis of Alfalfa Quality Traits Using Multi-Type Features and Machine Learning. Biosystems Engineering (Submitted)
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Chen, J. and Zhang, Z., A new multisource optical and SAR satellite remote sensing data fusion framework toward capturing fine-scale alfalfa growth. Accepted by and presented at American Society of Agricultural and Biological Engineers (ASABE) International Meeting, July 28- July 31, 2024, Anaheim, CA, USA.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Yu, T., Xu, Y., Zhou, J., & Zhang, Z. Estimating Pre-Harvest Alfalfa Quality Traits Using Multi-Type Features and Machine Learning. Accepted by and presented at American Society of Agricultural and Biological Engineers (ASABE) International Meeting, July 28- July 31, 2024, Anaheim, CA, USA.
- Type:
Other
Status:
Published
Year Published:
2024
Citation:
Extension Publication: Jerry Cherney, Zhuo Zhang, Jinha Jung, Paul Mitchell, and Matthew Digman. 2024. AlfAdvisor: A program to assist with alfalfa harvest management decisions. Pro Dairy Manager, p. 2-3. March, 2024. Progressive Dairy magazine.
|
Progress 09/01/22 to 08/31/23
Outputs
Target Audience:The target audience for our research activities encompasses a wide spectrum, including alfalfa academic communities, conference attendees, collaborators and partners and practitioners. The technical roadmap, model performance, and results analysis of this study will primarily be disseminated and shared in the form of peer-reviewed academic papers and various types of reports. The target audience could get access to our research findings through these publications. Furthermore, conference attendees (e.g., American Geophysical Union and American Society of Agricultural and Biological Engineers) will have the opportunity to engage directly with our work through presentations, discussions, and networking events. Collaborators and partners will benefit from our research by accessing detailed reports and insights that support cooperative efforts and advance our shared objectives. Practitioners in the field will find practical applications and solutions in our work, aiding in their daily practices and enhancing their effectiveness. Our commitment to transparent and accessible communication ensures that our research reaches and benefits a broad and diverse audience, fostering collaboration and innovation within our domain. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?1. A PhD graduate student is being trained under the supervision of PI Zhang. The student has been working on alfalfa field data collection and machine learning model development for alfalfa yield and quality prediction. A manuscript is under review with the student as the second author. A conference abstract with the student as the first author was submitted to AGU 2023. 2. A postdoc is currently working with PI Zhang. The postdoc has been devoted to developing alfalfa yield and quality traits prediction models using multisource satellite remote sensing data. Three manuscripts are under review with the postdoc as the first author. 3.A Master's graduate student wastrained under the supervision of Co-PI Digman. The student has been working in an alfalfa field, collecting alfalfa samples and using different windrow densities to determine different drying rates for alfalfa. She has completed two conference presentations and successfully defended her thesis. 4.A Ph.D. student is being trained under the supervision of Co-PI Jung. The graduate student was trained in basic and advanced Python programming. The graduate student was also trained in FastAPI programming for full-stack web development. The graduate assistant has been working on designing cyberinfrastructure and the full-stack development of the project. 5. A PhD student is being trained under the supervision of Co-PI Mitchell. The graduate student has been working on the economic model development to support the alfalfa harvest decision making. How have the results been disseminated to communities of interest?The project results have been disseminated through research publications, presentations and extensiontalks. Some of the research results have also incorporated into PI Zhang's class for teaching and student training activities. Some of the research progress and results were presented by PI Zhang's and co-PI Digman research groups at the ASABE 2023 annual international meeting in Omaha, NE. Also, Co-PI Mitchell gave an extension talk at the Manitowoc County Forage Council Annual Meeting. The AlfAdvisor web platform is accessible to the public at https://alfadvisor.d2s.org. However, it's worth noting that since the project is still in progress, our source codes are not currently available to the public. It will become publicly accessible once the project is completed. What do you plan to do during the next reporting period to accomplish the goals?For Objective 1, our plans are (1) Propose a methodological framework to jointly predict alfalfa yield and quality traits and develop a spatial downscaling approach to map fine-scale yield/quality distributions. (2) Investigate and obtain the phenology characteristics of alfalfa from multisource optical and SAR data. (3) Develop novel machine learning models to integrate features obtained from drone multispectral images and environmental factors to further enhance prediction performance. (4) Combine drone multispectral images with satellite data to further enhance the model performance (5) support the implementation of the drying rate model on the website as well as provide input to the cost model from a production agriculture perspective. For Objective 2, our plans are: Further extend the current decision model include(1) determining optimal storage method from a farmer specified set based on relative profits, drying requirements, rain forecast, (2) using the Milk2016 metric for analyzing the yield-quality tradeoff when a farmer does not intend to sell the hay, (3) including multiple fields of alfalfa in the analysis to provide an optimal harvest order based on respective expected profits and equipment transportation costs, (4) incorporating expected profits from future cuttings and opportunity costs of end-of-season stands, and (5) estimation of losses from rainfall to interpret risk of harvesting prior to forecasted precipitation. For Objective 3 (web tool), our plans are: Implement the updated drying rate model on the website. Preprocess Sentinel-1 and HLS data for input into the yield and quality model. Implement the Yield and Quality model on the cyberinfrastructure. Add features to the web application for visualizing and manipulating the results of the yield and quality model. Integrate the economic model provided by another group into the cyberinfrastructure.
Impacts
What was accomplished under these goals?
1. The following tasks have been accomplished for Objective 1. (1) Alfalfa sample collection and processing Alfalfa samples were collected in both Wisconsin and New York over the growing season in 2023. There were three fields in Wisconsin and three to four cuts in each field from May to September. Five samples were randomly selected for each sampling in a field, leading to a total of 261 samples in WI and the other 300 samples in NY. The samples were weighed immediately after cutting and dried for the dry matter yield. Quality traits were measured using lab NIR analysis. (2) Yield and quality model development and evaluation using satellite data Multisource satellite images were used in this project. First, an accurate data fusion method was needed when multisource optical and SAR data were used together to monitor alfalfa growth. Consequently, an improved fusion framework was proposed to combine Landsat-7/8, Sentinel-2, and Sentinel-1 data. After conducting the new data fusion framework, improved long-term optical and SAR data were generated for alfalfa monitoring. The proposed framework was optimized by the radiative transfer theory and vegetation growth mechanism. Second, the impacts of some key variables on data fusion wereinvestigated and examined. Meanwhile, we proposed and compared two modeling strategies, i.e., fusion-then-prediction (FTP) and prediction-then-fusion (PTF) in predicting alfalfa yield and quality traits. Alfalfa parameters were accurately predicted from multisource satellite data. Last, we optimally integrated optical and SAR data for improving alfalfa yield and quality traits prediction. Finally, predictions of alfalfa parameters were further improved through optimal integration of optical and SAR data. (3) UAV image processing and feature extraction UAV images of each field were processed for geographic correction, radiometric correction, merging and clipping. The spatial resolution of the image is approximately 2 cm. Then several vegetation indices such as NDVI, CIgreen, and CIrededge, and statistical features were extracted for each field. (4) Drying rate model This research investigated the potential of using automated stands to develop a drying rate model for alfalfa, and compared their drying rate with in-field wilting rates. The study aimed to establish whether the difference in drying rates between the two methods is significant enough to create a corrective coefficient. The study also examined the reliability of the Rotz & Chen (1985) drying rate model, highlighting the need to update it for nighttime drying and re-wetting factors. The benefits of automated stands include reduced human intervention, minimized weather-related delays, and continuous data collection. This supports creating a comprehensive alfalfa drying model, which might extend to other herbaceous biomass. Trials involved windrow sampling, manual screens, and automated stands. The findings showed no detectable difference between automated stands and manual screens at high density. However, at low windrow density, trials resulted in statistically detectable differences. In these cases, we propose a correction factor, but more data is necessary to fit a model. From current data, the Rotz & Chen (1985) equation can still be used to predict the drying rate of alfalfa during active drying times at this model will be implemented into the web tool. Based on our data, soil moisture does not display a strong enough correlation to drying rate to merit inclusion in the equation and could result in a simplification of the model. 2. The following tasks have been accomplished for Objective 2. (1) Formulation of objectives and constraints based on available data. Using available data from satellite imaging and drying rate models, as well as available weather forecasting data, the objective and constraints of a farmer planning to harvest a single field of alfalfa was characterized. Assumptions include a profit maximizing, risk neutral farmer who is valuating their alfalfa harvest based on current market price structures. Simple constraints on rain were determined that set thresholds for the maximum allowable quantity of precipitation during the drying period and the maximum allowable probability of rainfall exceeding the quantity threshold. An additional assumption of independence from future cutting decisions was made. (2) Development of decision support model theory and algorithm. An algorithm that takes as input available data and farmer preferences of storage type, current market prices for alfalfa based on quality metrics, and the farmer's level of aversion towards precipitation was developed. Calculations of partial budget profits from harvest were derived for the decision horizon. Possible informative, user-interface representations of the tradeoffs between yield, quality, and risk of losses for incorporation into web platform were considered. 3. The following tasks have been accomplished for Objective 3. For Objective 3, Co-PI Jung's group configured a web server (https://alfadvisor.d2s.org) as a first step, and the below features have been implemented so far. (1) Developing and Implementing User Authentication and Authorization: Like many other websites, we require users to complete a sign-up process by providing essential information such as "First name," "Last name," "email address," and password. Once a user successfully finishes the sign-up process, the provided information is securely transmitted to the database and stored. (3) Farms and Fields Management: We implemented features to manage farms and fields. Once a user logs into their account, a query search is initiated on the farms table within the database. If, for instance, a user's account has just been created, there are no farms associated with it. Consequently, the map's bounding box encompasses the entire United States. On the other hand, if a user has only one farm, the bounding box spans 1 degree by 1 degree, with the farm positioned at the center of the map. Finally, if a user has multiple farms, the map's bounding box for visualization is adjusted to include all of the farms within its scope. Like farm management, we offer field visualization, creation, and editing features. However, fields come with boundaries that necessitate specific visualization and editing capabilities,When a user opens a farm, a query search request is dispatched to the backend. (4) Retrieving weather forecast data for the selected field: Upon receiving the field boundary polygon layer, we have developed a code that enables the automated retrieval of weather forecast data for each field. This is achieved by providing the geographical coordinates of each field stored in the database, which allows us to programmatically access weather forecast data via the NOAA API Web Service. The provided forecast data is based on two well-established models:Global Forecast System (GFS) andHigh-Resolution Rapid Refresh (HRRR). (5) Implementing the Drying Rate Model:The Drying Rate model has been implemented using the Dry-Bulb equation. To perform this calculation, we require two user-provided values: a) The initial moisture content (in percentage), which is the starting point. b) The desired moisture content (in percentage) that you wish to achieve. Users input these values, and we incorporate them into our calculations to determine the necessary drying time to reach the desired moisture content from the initial state.The process to compute the moisture content involves calculating the area under the drying rate graph, which represents the amount of water that has evaporated in an hour. We then subtract the amount of moisture lost from the initial moisture content to determine the current moisture content. The same process will be repeated for each hour until the moisture content reaches the target moisture content.
Publications
- Type:
Journal Articles
Status:
Submitted
Year Published:
2023
Citation:
Jing, Z.,Tong Y., Ranjbar, S., Jiang C., Digman, M., & Zhang, Z. (2023). Alfalfa yield and quality estimation using Sentinel-1 synthetic aperture radar backscattering coefficients. Manuscript Submitted for Publication.
- Type:
Journal Articles
Status:
Submitted
Year Published:
2023
Citation:
Chen, J. and Zhang, Z. An improved fusion of Landsat-7/8, Sentinel-2, and Sentinel-1 data for monitoring alfalfa: implications for crop remote sensing. Manuscript Submitted for Publication.
- Type:
Journal Articles
Status:
Submitted
Year Published:
2023
Citation:
3. Chen, J., Zhou, J., Cherney, J. H., and Zhang, Z. Prediction of within-season alfalfa yield and quality traits using fused near real-time PlanetScope, Landsat-8/9, Sentinel-2, and Sentinel-1 satellite data. Manuscript Submitted for Publication.
- Type:
Journal Articles
Status:
Submitted
Year Published:
2023
Citation:
4. Chen, J., Yu, T., Cherney, J. H., and Zhang, Z. Optimal integration of optical and SAR data for improving alfalfa yield and quality traits prediction: new insights into satellite-based forage crop monitoring. Manuscript Submitted for Publication.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
1. Zhou, J., Yu, T., Ranjbar, S., Chen, J., Digman F. M., & Zhang, Z. Alfalfa Yield and Quality Estimation Using Sentinel-1 Synthetic Aperture Radar Backscattering coefficients. Presented at American Society of Agricultural and Biological Engineers (ASABE) International Meeting, July 10 - July 12, 2023, Omaha, NE, USA.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Fares, D., Shinners, K., Zhang, Z. and P. Mitchell. Determining the Reliability of Automated Systems for Collection of Field Wilting Rates of Herbaceous Biomass. Presented at the American Society of Agricultural and Biological Engineers (ASABE) International Meeting, July 10 - July 12, 2023, Omaha, NE, USA.
- Type:
Conference Papers and Presentations
Status:
Submitted
Year Published:
2023
Citation:
2. Yu T., Zhou, J., & Zhang, Z. (2023). Alfalfa Yield and Quality Traits Prediction Based on UAV Remote Sensing, Vegetation Model and Machine Learning. Submitted to AGU 2023.
|
Progress 09/01/21 to 08/31/22
Outputs
Target Audience:We have reached Team Forage in the Division of Extension at the University of Wisconsin-Madison, for accessing the Wisconsin Alfalfa Yield and Persistence Program (WAYP) database. The WAYP database updates yearly since 2007 the yield and quality of alfalfa harvested from production fields over the life of the alfalfa stand. To date the WAYP program has collected 15 years of field-level alfalfa yield and quality data and UW Extension has been the keeper of the database. One main objective of the WAYP is to verify alfalfa yield and quality from production fields over the life of the stand. UW-Extension has been collecting long-term data to quantify changes in stand productivity as fields age. This dataset has been used by our team to associate with satellite imagery for developing models to estimate alfalfa yield and quality. The model performance has been evaluated and summarized in a manuscript that will be published in a peer-reviewed journal and disseminated in research seminars, extension outreach activities, and conferences. The basics, expected outcomes, and impacts of this project were introduced in two extension articles by well-recognized media among forge stakeholders. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?1. A PhD graduate student is being trained under the supervision of PI Zhang. The student has been working on alfalfa field data collection and machine learning model development for alfalfa yield and quality prediction. A manuscript is under preparation with the student as the first leading author. 2. A Masters graduate student is being trained under the supervision of Co-PI Digman. The student has been working in an alfalfa field collecting alfalfa samples and using different windrow densities to determine different drying rates for alfalfa. A manuscript is under preparation with the student as the first leading author. 3. A Ph.D. student is being trained under the supervision of Co-PI Jung. The graduate student was trained in basic and advanced Python programming. The graduate student was also trained in FastAPI programming for full-stack web development. The graduate assistant has been working on designing cyberinfrastructure and the full-stack development of the project. How have the results been disseminated to communities of interest?The project results have been disseminated through invited research seminars and talks from Microsoft Research and Pheasants Forever. Some of the research results have also been incorporated into PI Zhang's class for teaching and student training activities. Some of the research progress and results were presented by Co-PI Digman at 2022 ASABE annual meeting in Houston, TX under the title "Development of an Automated System to Collect High-Resolution Field Wilting Rates of Herbaceous Biomass". What do you plan to do during the next reporting period to accomplish the goals?For Objective 1, our plans are: (1) Continue alfalfa field sampling to provide enough ground truth data for yield and quality model development and validation. (2) Extend the current model input dimensions by incorporating features from optical satellite imagery and those with higher temporal and spatial resolutions. (3) Develop novel machine learning models to integrate environmental factors to further enhance prediction performance. (4) Collect drone multispectral images and combine drone images with satellite data to further enhance the model performance. (5) Develop a machine learning model to be able to predict the drying rate of alfalfa based on weather forecasting For Objective 2, our plans are: (1) Predict the milk per acre index for each of the stored forage options, based on the Milk2016 model with the predicted alfalfa yield and quality as well as the drying rate as inputs. (2) Estimate the value of each stored forage option based on the predicted index as per user-defined management costs and milk price. For Objective 3, our plans are: (1) Adding a user authentication feature to the web application so that users can manage multiple field locations easily. (2) Developing additional features to the cyberinfrastructure to support Objectives 1 and 2. (3) Working with other PIs to integrate predictive models (drying rate, yield, quality, and economic) into the cyberinfrastructure.
Impacts
What was accomplished under these goals?
1. The following tasks have been accomplished for Objective 1. (1) Alfalfa sample collection and processing Alfalfa samples were collected in both Wisconsin and New York over the growing season in 2022. There were three fields in Wisconsin and four cuts in each field from May to September. Before each cut, four samplings were performed in each field beginning from alfalfa being 15" tall with around 3 days intervals between two samplings. Five samples were randomly selected for each sampling in a field, leading to a total of 240 samples (3 fields × 4 cuts × 4 sampling/cut × 5 sites) in WI and the other 240 samples in NY. The samples were weighed immediately after cutting and dried for the dry matter yield. Quality traits were measured using lab NIR analysis. (2) Satellite image processing and feature extraction (PI Zhang) Satellite images were downloaded from the Sentinel-1 high-resolution Level-1 Ground Range Detected (GRDH) products, for the study areas from 2016 to 2021, as listed in the WAYP database. The images were processed for border noise removal, thermal noise removal, radiometric calibration, and terrain correction. Then backscattering coefficients VV and VH along with the VH/VV backscattering ratio were extracted for each field of interest. (3) Model development and evaluation Four machine learning models, namely Support Vector Regression (SVR), Multilayer Perceptron (MLP), Random Forest (RF), and Extreme Gradient Boosting (XGB), were developed using the satellite image features and ground data from the WAYP database to estimate the alfalfa dry matter yield and quality traits. We also evaluated the effects of image collection timing on the estimation performance in each of the alfalfa traits. Field yield and quality maps have been developed from the model estimates to visualize the spatial variations with a resolution of 10 × 10 m, which will be incorporated into the web platform for public access. (4) Collect drying-rate data Hay drying data has been collected over the course of two summers (2021 and 2022). These drying rate data was collected using automated and manual drying stands. The automated drying stands use Arduino MKR GSM 1400 to raise linear actuators every 15 minutes, the linear actuators are connected to screens that hold either 7.25kg or 9.07kg of alfalfa. When the screens are raised load cells read the weight of the alfalfa and send that information to a Campbell Micrologger 21X. Over 40 trials have been conducted during the summers of 2021 and 2022. This data is analysed and filtered to remove any illogical data points, and noise. (5) Develop drying rate models The data collected during the summers of 2021 and 2022, were analyzed in Excel to develop a drying rate curve. The slope of the drying rate curve is taken to determine the drying rate of alfalfa under weather conditions during that same time period. 2. For Objective 3, Co-PI Jung's group configured a web server (https://alfadvisor.d2s.org) as a first step, and the below features have been implemented so far. (1) Field boundary tool: Users can either draw their own field boundary on an interactive map or upload an already existing field boundary polygon file. The field boundary polygon layer will be used to compile required data layers programmatically from other resources. Users can also download the field boundary polygon layer in either shape or GeoJSON file formats. (2) Weather forecast tool: Once the field boundary polygon layer is provided, we implemented a code to download weather forecast data over the field programmatically. Weather forecast data are pulled using NOAA API Web Service (https://www.weather.gov/documentation/services-web-api). The weather forecast data will be a critical data source for the drying rate prediction model development. (3) Sentinel data download: We are currently working on downloading Sentinel 1 and 2 data over the field programmatically. This tool will download the Sentinel data programmatically and visualize multi-temporal layers of the growing season. These Sentinel layers will be a critical data source for the alfalfa yield prediction model.
Publications
- Type:
Journal Articles
Status:
Under Review
Year Published:
2022
Citation:
Ranjbar, S., Zhou, J., Digman, M., & Zhang, Z. (2022). Alfalfa yield and quality estimation using Sentinel-1 synthetic aperture radar backscattering coefficients. Submitted to Journal of ASABE in 2022.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
B. Pike, D.F. Fares, and M.F. Digman. Development of an automated system to collect high-resolution field wilting rates of herbaceous biomass, Poster presentation at the 2022 ASABE Annual International Meeting.
|