Chandra Krintz

Context and Problem Statement

• The agricultural industry is currently underserved by technology and analytics despite collecting vast amounts of sensor data, historical records, and government data.
• Feeding a projected 9 billion people by 2050 with no available increase in arable land necessitates a shift toward extreme efficiency and productivity gains.
• The commercial opportunity in California's agriculture sector is estimated at $50 billion, indicating a need for immediate technological intervention.
• Analytics can be adapted from web services (e.g., Amazon recommendations) to agricultural use cases like irrigation scheduling, fertilizer application, and crop disease identification.

Technological Convergence and Applications

• The proposed solution converges Internet of Things (IoT), robotics, big data, machine learning, statistical analysis, and multispectral image processing.
• A case study involving Fresno State used interpolation algorithms on wine grape samples to distinguish high-quality from low-quality produce for differential harvesting.
• Diffusion analysis applied to the same dataset successfully identified the root causes of quality differentiation without direct manual intervention.
• Multispectral and thermal imaging enable remote monitoring to detect crop health issues, pest infestations, and irrigation leaks via temperature differentials.
• Farmers can utilize tablet interfaces and drone imagery to focus field inspections on high-probability problem areas rather than surveying entire fields manually.

System Architecture: The "Lights-Out" Appliance

• The researchers propose a hybrid cloud approach combining on-farm computation with cloud analytics to address low-bandwidth, expensive, and unreliable connectivity on farms.
• The proposed on-farm system is a self-managing, fault-tolerant appliance designed for non-expert users, comparable to a "TiVo" box or refrigerator.
• The hardware prototype utilizes six Intel NUCs, providing 24 processors, over a terabyte of disk, and 100 gigabytes of RAM for approximately $4,000.
• The system supports portability, allowing applications written in Hadoop, Spark, R, or MATLAB to run transparently on both the on-farm appliance and public cloud platforms like Amazon or Google.
• To prevent resource contention between different analytics engines (e.g., Spark, Storm, Hadoop), the system employs research-driven configurations that manage shared resources while maintaining high availability.
• The architecture adapts large-scale cloud technologies for smaller, fault-tolerant edge environments, ensuring operation continues even if individual components fail.

Economic and Social Considerations

• Data sovereignty is identified as a critical requirement; farmers must retain full control over their data, including the choice to share, sell, or keep it private.
• The solution relies on open-source software and reduced hardware costs to ensure viability for a broad population of growers.
• Validation of research is conducted through direct partnerships with industry players, farmers, and IoT sensor companies rather than isolated lab experiments.
• The researchers emphasize the need for a collaborative community to address "sticky" systems problems and achieve global food security.