Non Technical Summary
USDA-ARS Tucson, AZ has acquired a vast reservoir of multi-sensor data, including thousands of frame photographs and millions of sensor measurements, from field experiments with managed honey bee colonies. The reservoir is a loose collection of image directories, CSV files, Excel spreadsheets, and digital hive inspection text logs. The project will explore and exploit this reservoir and make it public as a curated dataset for the U.S. precision apiculture community. The project will use data-driven AI to design, train, and evaluate on the curated dataset automated digital frame analysis tools as well as predictive models to anticipate the future status of colonies from non-invasive senor data series aligned with beekeeper observations.

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
402	3010	2080	100%

Knowledge Area
402 - Engineering Systems and Equipment;

Subject Of Investigation
3010 - Honey bees;

Field Of Science
2080 - Mathematics and computer sciences;

Goals / Objectives
Short-Term (3 years) Goal: To explore and exploit the USDA-ARS Tucson, AZ2014-2022 digital data reservoir of field experiments with managed colonies and tomake the reservoir public as a F.A.I.R dataset for the U.S. precision apiculture com-munity at USDA Ag Data Commons.Long-Term (5 years) Goal: To strengthen the precision apiculture research programat USU through its research collaboration with USDA-ARS Tucson, AZ

Project Methods
1) Image Annotation: We will remove redundant or unusable photographs and annotate instances of the 7 categories in Objective 2, Other categories wiill be added/removed as needed.2) Data Alignment: We will align each labeled frame photograph, whenever feasible, with frame mass (kg), mass of adult bees and sealed brood (kg), queen status (present/absent), experimental treatment group (integer), ultimate colony survivorship (yes/no), in-hive temperature (degrees Celsius), hive weight (kg), and in-hive CO2 concentration (ppm).3) Image Classification: We will train and test theree categories of models to classify different areas of frame photographs: deep learning models (e.g., YOLO) and standard machine learning models (e.g., Random Forests).4) Time Series Forecasting: We will use the long short-term memory (LSTM) networks to build predictive models as time series forecasters and compare their relative performance.5) Accuracy Evaluation: We will evaluate the accuracy of each model with four standard metrics: precision, recall, F1, and intersection over union (IOU).