Cover Crop Dataset for the Midwest
Role: Lead Geospatial Data Scientist
Paper in writing.
Tools: Google Earth Engine, Python, Keras
Research has shown that planting cover crops during non cropping seasons can significantly improve soil health and ultimately improve yield production. Cover crop adoption has been incentivized through monetary and subsidy benefits by the federal US government and a variety of Agricultural programs.
The cover crop dataset will feed into Land Core’s risk model which determines the risks and opportunities for farmers and Agricultural lenders such as Agricultural insurance companies when adopting regenerative practices including reduced tillage and crop rotation
Spatially explicit information on cover crops is fundamental to analyzing the impact of soil health practices on agricultural productivity, its resilience to climate risks, and carbon measuring and reporting applications over large areas. Here, we develop a remote sensed cover crop dataset using the Harmonized Landsat Sentinel (HLS) data for major states in the U.S. Midwest. Normalized Difference Vegetation Index (NDVI) time series from HLS are used to extract the cover and cash crops green up dates and growing season features using the within-season emergence (WISE) algorithm.
Still updating ….
Predicting Maize Yields at Large Scale using Remote Sensing in Mexico
The advancement of remote sensing applications has been enabled with increased computing power and higher resolution images. It is no doubt that the biggest challenge in remote sensing applications is that data is often expensive, difficult to obtain, and hard to work with. In this project, we develop a method to predict yields using satellite imagery with only aggregated training data (i.e. we are predicting the yield of an individual farm from data aggregated in a municipality level).
We introduce a dimensionality reduction technique that uses a 10D histogram to track the pixel count of NDVI over 10 months (May-February). This method intuitively shows how many pixels moves from brown to green or green to brown w. We believe this method will be useful for policy makers who are developing subsidy programs to target rural farmers. Additionally researchers can find this model useful to fill data gaps they may have with their training data.
I have presented this project in the TWEEDS, AGU conferences and Google Earth Booth.
Exercise is Medicine - on Campus
UC Berkeley Graduate Student Researcher
University Health Services (UHS)
Tools: Excel, Word, R.
The Exercise in Medicine program, initiated by the American College of Sports Sciences, promotes physical activity as a vital aspect of overall health. This program was recently introduced at the University Health Services (UHS) with the goal of improving the health and well-being of students by incorporating movement into daily campus life.
My research focus was on reducing barriers to access physical activity facilities for marginalized populations on campus. As part of this effort, I co-led the organization of a swimming access series event for black students, which was attended by 67% of participants who had never used the swimming pools before.
Feedback from attendees indicated that the event was valuable and should be continued at the school. I conducted research on methods to improve and assess the program’s impact on student health outcomes. I also prepared impact evaluation reports for all the events done developing KPIs for the program.
This dynamic role involved both on-the-ground interactions with students to gather feedback and the development of creative event ideas and planning. I am grateful to have worked with an excellent team and have been able to prioritize physical activity in my daily life, including walking and stretching as personal favorites.
Predicting Plant Functional Types in US
Data Science Intern at Regrow
Mentors: Matt Jones & Sam Barret
Tools: Google Earth Engine, Python, Keras, Numpy, Pandas
Regrow is a climate tech company that is fighting climate change through sustainable farming. I joined Regrow in summer 2022 as a Data science intern where my project was in developing a model that predicts Plant Functional Types (PFT) in US. PFTs, also known as land cover types, are fractional products which are reliably indicative of the heterogenous landscape of rangelands since it is common to find several PFTS on a field plot such a bare soil and shrubs.
Building upon on Allred et al. (2021) methodology, I used Landsat 5,7,8 to extract the predictors of the model which were Bands 1-6, Normalized difference vegetation index (NDVI) and Normalized burn ratio two (NBR2). These predictors were collected on 64 day composites over 2 years for each field plot.
This was the project that sparked my passion for geospatial data science for resource management. I worked with an amazing team that is still pivotal to my career till date. I have presented this project at the AGU 2022.
Farming Communities in Philippines
UC Berkeley -Masters of Development Engineering Project
Course: Designing, Scaling and Evaluating Development Engineering Applications
Team: Emman Uy, Patricia Quaye, Calvin Chen, Alan Huynh
In this project, we collaborated with the government of Zamboanga Del Norte, a province in the Philippines with a high rural population facing poverty. The government implemented a program to empower and uplift these rural farming communities by providing them with resources such as tractors, farming tools, and seeds to improve agricultural production. The program also included transportation services to facilitate the movement of goods to markets and reduce food loss. While this was a successful poverty alleviation initiative, it was not financially sustainable as it relied heavily on government funding.
Therefore, we worked with the government to devise a strategy that would allow the program to be more self-sustaining while also preserving the ecosystem. Through 100 interviews with farmers, government officials, and other stakeholders, we found that the farmers greatly valued the program and would like it to continue. Based on our findings, we recommended the establishment of a squash flour factory as a potential means to increase farmer income and production.
IoT based Livestock Monitoring System
2018 – Present
Tungana – a company I co-founded
In Kenya, livestock farming is popularly done in open spaces where cows typically grazing in open spaces under the supervision of herders. With climate change there has been more scarcity of pasture which has led cattle to wander further leading to lost cattle. Herders have also perished, for instance, two herders were killed by elephants when they were herding. To address these issues, we are working with a ranch in Taita, Kenya to develop GPS-enabled ear tags for cattle that can transmit the location of the animals to facilitate the retrieval of lost cows.
This system will also provide insight into the grazing patterns of the cattle, allowing for more informed grazing management decisions, such as the placement of water troughs, given that climate change has impacted the water consumption of cattle. This a project still in progress where the main challenges have been internet and power access in remote regions. As a measure to address this, we have a Wireless Planet Initiative that addresses rural internet connectivity gaps that can easily hinder innovation and implementing solutions at last mile.