Progress 01/15/24 to 01/14/25
Outputs
Target Audience:The target audience during the reporting period included a mixture of students, faculty, and staff at Appalachian State University and conference attendees at the Annual Meeting of American Association of Geographers. While there is no official data on Appalachian State visitors available, it is most likely visited by > 100 students, faculties, university staff, and probably community members. AAG 2024 in Hawaii had more than 5000 participants, including geographers and interdisciplinary scholars from various parts of the world. The attendees were from academia, government, and industry. Further, since the conference was in hybrid format, there were many virtual attendees. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?We have been training these students: Alan Huff (Aug 2024 - current): He is a Bachelor student of Computer Science at ASU. He is working on analyzing the machine learning portion. S M Rakib Karim (Jan 2024 - current): He is a PhD Student of Computer Science at the University of Missouri Columbia. He is working under the guidance of Dr. Hossain (Co-PI). The students are actively involved in exploring large-scale geospatial and agricultural datasets and developing machine learning models to analyze the interplay between windbreaks, wind speed, and crop loss. Through this work, they are gaining hands-on experience in spatial data analytics, model development, and interdisciplinary climate resilience research. Jack Dickson (Jan 2025 - current): He is a Bachelor student of Geography at ASU. He is helping us with the spatial analysis of the data. Colin Tiller (Jan 2024 - May 2024): We hired Mr. Tiller, who was studying BS in Computer Science, was hired to do data curation and analysis. He processed all the input datasets. ?Tyler Collin (Aug 2024 - Dec 2024): He was a graduate student of Computer Science at ASU. Through ASU funding, he worked with Dr. Tinghao Feng, Assistant Professor of Computer Science at ASU on curating various input datasets necessary for the geovisualization and advanced analysis of the data. How have the results been disseminated to communities of interest?We disseminated the results in the following ways: Teaching in the classroom: Thapa taught GHY5150: GIS Seminar course where the students learned about Python programming using windbreak data. This is an example of taking research into the classroom. The students did multiple assignments using this dataset and even presented two posters at the ASU Research Day in April 2024. The ASU Research Day is mainly attended by students, faculties, university staff, and probably some local community members. Since the university serves more than a quarter of the rural population, we assume that the study has been disseminated to the rural population as well, to whom the information will be relevant. ?Geovisualization: We have been trying to visualize the data in an effective manner using a GitHub page. Dr. Tinghao Feng, a Computer Science faculty at ASU, is working on data visualization. We have created a GitHub page that displays spatial and temporal input and output data. It will point out the heat map of the output variable and the trend of the input variables. For a given county that we select, we can visualize the trend of input variables and how they are correlated with the output variables. We plan to keep working on it. What do you plan to do during the next reporting period to accomplish the goals?This is the final year of the project, so we plan a few activities: Data analysis and model development: Spatiotemporal analysis: As mentioned in objective 3, we are conducting spatiotemporal hotspot analyses using both conventional and machine learning methods to detect high-risk counties. As preliminary analysis suggests notable correlations between windbreak density, wind speed, and indemnity across counties, we will investigate spatial and temporal patterns to identify regions most vulnerable to wind-related crop loss. We also plan to model and predict potential losses under various scenarios of windbreak coverage. Econometric and machine learning models: We are also developing econometric and machine learning models to assess the relationships. We will emphasize temporal models to capture how windbreak effectiveness evolves over time. End-to-end methodological framework - We are in the process of finalizing an end-to-end methodological framework integrating econometric and machine-learning approaches. Knowledge Transfer and Dissemination: Information dissemination via a conference: We will share our findings with a wider audience in Fall 2025. Our goal is to present the result to our stakeholders during a fall conference, such as the Perennial Farm Gathering in 2025 (date to be confirmed) or other gatherings of agroforestry-related stakeholders. Website update: Publish a GitHub page that visualizes spatial and temporal datasets. The information will be available on the Center of Agroforestry website. Peer-reviewed manuscripts and professional conferences: We are working on multiple peer review publications that we plan to submit in the fall of 2025. Our group also plans to attend Fall conference (American Geophysical Union fall gathering or similar conferences) to disseminate findings to wider audiences. Information dissemination through other mediums: In the fall of 2025, we will develop podcasts (https://agroforestry.libsyn.com/) (average monthly listeners ~2,000), and YouTube videos (current subscribers 482) featured through the University of Missouri Center for Agroforestry (UMCA) accounts. The policy briefs will be shared in UMCA newsletters "Action in Agroforestry" and "Green Horizons" (current membership of ~1500).
Impacts
What was accomplished under these goals?
Rich geospatial datasets on windbreak: We have curated input data products for all of our study areas (i.e., the four states). The input data are - average and maximum wind speed (m/s), indemnity ($), windbreak density around specific crop fields, total crop area (m2) for all the major crops, and storm events for each county over the study time period. All the data is aggregated at the county level. Statistical analysis of data: For Objectives 2 and 3, we investigated three key research questions: RQ1: Is there a relationship between windbreak coverage and crop insurance indemnity? RQ2: Does wind speed correlate with indemnity payouts? RQ3: Do windbreaks and wind speed interact in influencing indemnity? To address these questions, we identified two independent variables--tree density (as a proxy for windbreak presence) and wind speed--and one dependent variable, indemnity (crop insurance payouts). We conducted correlation analyses across four major crop types: corn, wheat, soybeans, and sorghum. Our analysis revealed a negative correlation between tree cover and crop loss, suggesting that windbreaks help mitigate wind-related crop damage, thereby lowering indemnity payments. This supports the hypothesis that windbreaks serve as effective natural barriers against environmental stressors. In addition, we designed and tested preliminary regression models to predict crop loss based on windbreak density and wind speed. Machine learning model valuation: We have used the above data to run machine learning models. A table was created containing input and output features: crop coverage, maximum and average wind speed, and maximum and average precipitation. Crop_id, GEO_ID, and date were the primary keys. The output variables were indemnity loss ratio, net planted quantity, and type of cause of loss. For all the data, 12-month time series ending on the corresponding indemnity for a given GEOID and crop ID was generated. Incomplete timeseries were filtered, and paired time series with the output features (indemnity, loss ratio, and net planted quantity). The data was split into train and test sets then normalized and transformed. The input and output features were normalized separately for the time-series data and for the non-time-series data. We used latitude and longitude as features instead of the GEOID for training purposes, and for the crop id, we used one-hot encoding, meaning that there is one column for each possible crop id representing whether that id applies to a given row. We created a train-test split based on the last month GEOID, crop id, and date of each time series. We used the same train-test split for the non-time-series table. We trained random forest models and gradient boosting models on the non-time-series data. Predicting the indemnity divided by the net planted quantity gave better results for all models than did predicting the indemnity or loss ratios themselves. This is probably because the indemnity scales with the amount of land planted. The amount of land planted is not considered as an input feature but it most likely impacts the indemnity by increasing the amount that is at risk of damage, so it should be accounted for. The best models had an R squared value of around 0.1 to 0.2 on the test set. The gradient boosting model had very similar results on the test set to the LSTM model, and our current attention model did not perform quite as well overall. The model structure, learning rate, and number of heads could all still be optimized to improve results.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Hites, J.*, Bennett, W. *, Wilson, D. *, Thapa, B. 2024. Spatiotemporal analysis of tree windbreaks and crop loss from wind events in Kansas and Nebraska, USA. 27th Annual Celebration of Student Research and Creative Endeavors, Appalachian State University, NC, USA. April 18.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Hope, G. *, Ehlenberger, J. *, Thapa, B. 2024. Spatial analysis of wind events and wind-related crop loss for the Midwest, USA. 27th Annual Celebration of Student Research and Creative Endeavors, Appalachian State University, NC, USA. April 18.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Thapa, B., Hossain, T K S M, Kwon, M Y. 2024. Effectiveness of Windbreaks in Reducing Crop Loss in the Midwest, USA. Annual Meeting of American Association of Geographers, Honolulu, HI, USA. April 16-20.
|
Progress 01/15/23 to 01/14/24
Outputs
Target Audience:The target audience during the reporting period mainly consisted of around 100+ students visiting the university research day event at Appalachian State University. While there is no official data on the visitors available, it is most likely visited by students, faculties, university staff, and probably community members. Changes/Problems:As reported in the year 1 progress report, the PI and co-PIs have moved to other institutions or changed their roles. The Initial PI, Dr. Lovell, has left the University of Missouri along with other co-PIs, Dr. Thapa and Dr. Hossain. As a result, the University of Missouri worked on creating a sub-award for Appalachian State University (Thapa) and Texas Tech University (TTU) (Hossain). The award was dispersed in early 2023 for ASU and in the middle of 2023 for TTU. Because of the change of institutions, it took some time for the co-PI to find the resources needed for the project. To be efficient, Dr. Thapa re-hired Mi Young who initially worked on this project until she got a full-time position. There were also some delays in getting access to the High-Performance Computing (HPC) at ASU. However, we have access to HPC and also hired a computer science student in Spring 2024 to continue the work. Overall, the project remains on schedule and the objectives are being completed as planned. What opportunities for training and professional development has the project provided?We trained two students on this project. We re-hired Mi Young, a graduate student of Master of Data Science & Analytics at the University of Missouri, for the fall of 2023. She has been trained in geospatial data curation and analysis. An undergraduate student at Appalachian State University (ASU) presented the poster on Research Day in April 2023. The co-PI, Dr. Thapa, has moved to ASU in fall 2022. She was trained on the issue of how windbreaks affect wind-related crop damage. For the poster, she did the literature review and drafted a component of the poster. A Ph.D. student in Computer Science at the University of Missouri, also working with Dr. Hossain (co-PI) on exploring the datasets and developing machine learning models for capturing the relationships between windbreak, windspeed, and crop loss. How have the results been disseminated to communities of interest?An undergraduate student presented the poster at ASU Research Day event. It was mainly attended by students, faculties, university staff, and probably some local community members. Since the university serves more than a quarter of the rural population, we assume that the study has been disseminated to the rural population as well, to whom the information will be relevant. What do you plan to do during the next reporting period to accomplish the goals?Our plans for the next reporting period are: Objective 2: Task 1 and 2: We plan to run different regression and machine learning models and identify the most appropriate model in year 2. We will explore three types of machine learning models to characterize the data and trends: a) temporal, b) spatial, and spatiotemporal. Based on our initial exploratory analysis, the results suggest exploring the relationship between various variables at the spatially segmented level. This will help us create a draft of the end-to-end methodological framework using machine learning and econometric models. Objective 3: Task 1 & 2: We will also conduct spatiotemporal hotspot analysis using regular and machine-learning approaches as outlined in objective 3. As the first step, we are exploring each county's tree density, wind speed, and indemnity data to see if there are any temporal and spatial patterns in the data. We will then develop a spatio-temporal model to identify counties vulnerable to crop loss due to lack of windbreaks. We also plan to predict the loss that could happen for a county. Under our knowledge transfer and dissemination plan, Dr. Thapa and Dr. Hossain will develop new lecture materials on Data Science and Geospatial applications in agriculture and climate resilience research. We plan to submit manuscripts as well.
Impacts
What was accomplished under these goals?
Data curation and analysis: Since the PI moved to a new institution in fall 2022, we re-hired Mi Young to work on this project for 2023. She initially worked on this project during the spring and summer of 2022 until she graduated from the University of Missouri with a degree in Data Science & Analytics. One of the co-PIs, Dr. Hossain, moved to the University of North Texas in Fall 2022. He and his students have developed a pipeline that integrates three data sources: tree density, wind speed, and indemnity. The exploratory data analysis is performed on this integrated dataset. Currently, we are writing an article on the findings. Objective 1: Quantify windbreaks and tree cover on and around agricultural land using high-resolution landcover maps In 2022, we completed the data collection and preparation for two states, Nebraska and Kansas. For 2023, we did that for another two states, North Dakota and South Dakota. We quantified tree cover on and around agricultural land for the remaining two states for all crops. Objective 2: Develop predictive models to assess the effectiveness of windbreaks in reducing wind-related crop loss using econometric and machine learning algorithms: We have mainly focused on curating data for all the states for all four crops: corn, soybeans, wheat, and sorghum. We are doing the predictive modeling part for year 3 (2024).
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Pannell, T., Thapa, B. Understanding the relationship between windbreaks and wind-related crop loss. 26th Annual Celebration of Student Research and Creative Endeavors, Appalachian State University, NC, USA. April 19, 2023.
|
Progress 01/15/22 to 01/14/23
Outputs
Target Audience:The target audience of the conference presentation: Dr. Thapa presented the preliminary findings at the 5th World Congress on Agroforestry. The conference was attended by 800 people from all continents that included researchers, farmers, and First Nations representatives. Changes/Problems:While these changes will not likely slow the proposed work of 2023, we would still like to inform some changes in the roles and institutional affiliations of PI and co-PIs. Dr. Benjamin O. Knapp is the new PI after Dr. Sarah Lovell left the University of Missouri in December 2022. Dr. Thapa, Dr. Hossain, and Dr. Cai, the co-PIs, have also moved from the University of Missouri and joined Appalachian State University, the University of North Texas, and the United States Forest Service, respectively. What opportunities for training and professional development has the project provided?We hired Mi Young, a graduate student of Master of Data Science & Analytics at the University of Missouri, for the spring and summer of 2022. She has been trained in geospatial data curation and analysis. Dr. Thapa presented the preliminary findings at the 5th World Congress on Agroforestry. It was a valuable networking opportunity as the event was attended by close to 800 people from all continents Researchers, farmers, and First Nations representatives attended the conference that was held in Canada. How have the results been disseminated to communities of interest?Dr. Thapa presented the preliminary findings of the project at the 5th World Congress on Agroforestry in July 2022. The conference was attended by close to 800 scholars and farmers from all parts of the world. What do you plan to do during the next reporting period to accomplish the goals?Our proposed year 2 activities are as follows: Objective 1: Quantify windbreaks and tree cover on and around agricultural land using high-resolution land cover maps: We plan to complete all the tasks under this objective. We will be estimating tree cover on all six crops in our study area using the algorithm that have been developed and tested in year 1. We will also collect and curate additional variables that have been identified during stakeholder consultation meetings in year 1 and prepare it for the analysis. Objective 2: Compile all the above-identified variables for all crops and all the study area. Once all the input variables are curated, we plan to run different regression and machine learning models and identify the most appropriate model in year 2. We will explore three types of machine learning models to characterize the data and trends: a) temporal, b) spatial, and spatiotemporal. Based on our initial exploratory analysis, the results suggest exploring the relationship between various variables at the spatially segmented level. This will help us create a draft of the end-to-end methodological framework using machine learning and econometric models. Objective 3: We will also conduct spatiotemporal hotspot analysis using regular and machine-learning approaches as outlined in objective 3. Under our knowledge transfer and dissemination plan, Dr. Thapa and Dr. Hossain will develop new lecture materials on Data Science and Geospatial applications in agriculture and climate resilience research. We will also work towards creating a draft manuscript for submission. The preliminary result from the model will be presented during a special session at Agroforestry Symposium in Jan 2024 (date to be fixed). This is an annual event organized by the National Agroforestry Center at the University of Missouri.
Impacts
What was accomplished under these goals?
Objective 1: Quantify windbreaks and tree cover on and around agricultural land using high-resolution landcover maps We have made notable progress towards this goal. We quantified tree cover on and around agricultural land in Nebraska and Kansas for corn as a model crop. We quantified windbreak and non-windbreak tree density on and around corn fields for Nebraska and Kansas (two of our states) using 1-meter land cover map and 30-meter crop cover map from USDA. We are also collecting additional data based on feedback from our consultation meetings. Develop predictive models to assess the effectiveness of windbreaks in reducing wind-related crop loss using econometric and machine learning algorithms: In year 1, we curated a subset of data for corn for two of the four study states, i.e., NE and KS, and developed a test econometric model, i.e., mixed effect model. In addition, we are working on collecting additional data on soil erosivity and precipitation gradient. For the machine learning model, we worked towards developing the first three modules in an end-to-end pipeline that supports executing machine learning models: a) data integration, b) data curation, and c) data exploration. We integrated windbreak data (e.g., tree-related information) and weather data (e.g., wind speed and precipitation) at the farm level. As an initial step, we curated the data by removing the farms with missing and aberrant values, which we plan to explore in our next version of the curation module. We explored the data by performing correlation and partial correlation between various features of the curated data. In particular, we estimated the relationship between tree ratio, wind speed, precipitation, and indemnity (i.e., crop loss). We did not find a global trend at the state level. We are now focusing on exploring local trends in the data.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Thapa, B., S. Lovell, Z. Cai, KSM T. Hossain, MY Kwon. 2022. Agroforestry for climate risk management: Effectiveness of windbreaks in reducing crop loss in the Midwest, USA. 5th World Congress on Agroforestry, Quebec City, Canada. July 17-20.
|