Hacking Climate and Soil Data for North Carolina’s Crops

This is a guest post from Geospatial Analytics Ph.D. students Pratikshya Regmi and Titilayo Tajudeen.

From September 12–14, 2025, we participated in the 2025 N.C. Plant Science Initiative (PSI) Hackathon at NC State University. The annual event gives undergraduates, graduate students and postdoctoral scholars the chance to demonstrate their skills in a competition to solve ag tech-related challenges. With the goal to accelerate agricultural research using data analytics, machine learning, and hardware fabrication, the Hackathon has multiple tracks and is inclusive to participants with various levels of experience.

Our project asked a straightforward question: Which crops are best suited for North Carolina when climate and soil are considered together? We set out to connect climate history, soil information, and crop records to produce practical, data-driven insights.

Our team was intentionally interdisciplinary: Titilayo and I are Ph.D. students in Geospatial Analytics, Kashish Grover is a Ph.D. student in Horticultural Science program, and Jerry Yu is a graduate student in Statistics. This mix shaped every decision we made throughout the Hackathon. While we focused on data integration, mapping, and spatial reasoning, Kashish grounded our choices in agronomic realities, and Jerry guided the extraction of soil metrics. We jointly make decisions about modelling, evaluation, and predictions while structuring our presentation to reflect the insights gained from the data, the challenges encountered, and what we would have done differently if given more time.

The work unfolded in three steps. First, we conducted careful exploratory data analysis (EDA) to understand patterns, gaps, and inconsistencies. Second, we built AI/ML baselines to relate climate and soil factors to crop yields, and we reviewed historical records to highlight major crop failure events. Finally, we used those relationships to forecast top candidate crops for North Carolina’s conditions. The goal was not just prediction, but clarity—what factors matter most, where, and why.

A key lesson for both of us was balance. In short time frames, it is tempting to favor modeling over understanding. We found the opposite is true: strong modeling depends on first knowing the data well, its scales, spatial context, quality, and missingness. Going forward, we plan to devote proportionally more time to EDA while reserving enough time for modeling.

The experience was also a bridge across fields. We each learned outside our comfort zones: we deepened our understanding of agricultural constraints, and our teammates engaged with geospatial structure and uncertainty. This exchange gave the project direction and helped us present results that could be useful to stakeholders, not just technically sound.

We participated in the Advanced Level track, “Grow Thy Crop.” Participants hacked climate data to identify the top five crops best suited for North Carolina. They developed AI/ML models to analyze crop yields using historical and current climate data, while also identifying major crop failure events in history.

We are grateful that our work was recognized with first place in the Advanced Track, but the most important outcomes were the skills we built and the connections we made. Thank you to the Plant Sciences Initiative (PSI) as well as all the funding organizations for creating an environment where collaboration across disciplines is possible and encouraged. We look forward to applying these lessons to future research and outreach.