Performing Department
(N/A)
Non Technical Summary
Specialty Crop production margins are eroded by input costs, impacts of weather, pests and diseases, and market price fluctuations. Producers routinely overproduce to hedge against losses from environmental impacts and ensure sufficient supply to meet retail account demand, further reducing their average margins. If they had greater certainty in advance of how much they will produce and when it will be harvest-ready, they could consistently improve the match between supply and demand, reducing overall losses and increasing margins. This project will develop precision crop maturity forecasting models for a number of key fresh vegetable crops based on computer vision analysis of imagery collected over fields during the crop growing cycle, weather and solar irradiance data, and knowledge of phenological development stages of each crop. During Phase I, we will analyze historical planting, harvest and weather data from one of the top-10 producers in the industry for three representative crops combined with photosynthetically active radiation (PAR) data from satellites to determine the extents to which temperature and PAR history profiles can provide sufficiently accurate predictive models of harvest maturity dates. We will also collect high resolution aerial imagery, weather, and radiation data over fields during the Phase I period to determine the feasibility of using phenological development stages, as determined from computer vision analysis, to improve the forecasting capability. We will compare forecasted to actual harvest dates and yields for a set of representative crops grown in California and Arizona.The proposed capability will enable producers to know with high accuracy a field's different levels of maturity and when they can expect most of each field to be ready for harvest. This intelligence will significantly increase their ability to manage their supply in the face of environmental variabilities, ultimately reducing waste and increasing their margins We plan to sell the forecasting capability as a subscription service.
Animal Health Component
40%
Research Effort Categories
Basic
40%
Applied
40%
Developmental
20%
Goals / Objectives
The major goals of this project are 1) to validate that harvest date forecasts for one or more of the targeted crops can be improved significantly beyond current capabilities by incorporating some combination of temperature, radiation and remotely sensed imagery data, and 2) point to specific, in-depth experiments to be performed and capabilities to be developed during a follow-on Phase II effort (e.g. experiments in controlled growing environments and with additional crop types) that will improve the forecasting accuracy and robustness to a level that will make it commercially feasible and scalable. The specific objectives towards these ends are the following:Determine how much of the representative crops' growing periods can be explained solely by a degree day (DD) model, by analyzing historic weather and growing period data.Assess the accuracy effects of introducing photosynthetically active radiation (PAR) monitoring into the harvest maturity models.Assess the overall complexity of the observed relationship between temperature and PAR effects.Identify which phenological growth stages for each representative crop can be detected reliably with current aerial imaging capabilities.Assess the effectiveness of incorporating growth stages to serve as supplemental indicators of plant maturity in our forecast modeling framework.Test growth stage discrimination feasibility against functional image acquisition and processing requirements: e.g. image spatial resolutions, collection frequencies, and individual image frames versus field-level mosaics.
Project Methods
The following scientific methods and associated efforts will be used to evaluate the feasibility of the proposed forecasting system:1) Analyze Historical Weather and Growing Duration Data:The goal of this effort is to determine how much of each crop's growth can be explained solely by temperature effects, modeled as degree days (DDs).2) Collect Imagery, Temperature and PAR Data for Test Fields:We will begin the planning and execution of data collection early during Phase I, since it involves field work and equipment.3) Test Effectiveness of Temperature and PAR in Forecast Models:We may request that our partner-customer Bonipak grows one or more test beds with larger spacings between plants, to reduce or avoid occlusion of neighboring plants to test the impact.We will use historical PAR data and in-field PAR data from light sensors to assess the extent to which typical lighting conditions in the targeted geographies cause saturation effects for the crops of interest.4) Use Imagery Analytics to Derive Growth Stages:We will select a set of image classifiers to detect and classify the representative crops, train image classifiers to detect a preliminary set of growth stages for each representative crop, test feasibility of classifying growth stages after crop canopies have closed, using a segmentation technique, and test growth stage classification accuracy against different image spatial resolutions.5) Integrate Growth Stage Analysis into Forecast Models:We will select the growth stages identified in aerial imagery that occur consistently at about the same number of EDDs after planting, and from this set of growth stages, select specific stages for each test field and compare their EDDs to the fraction of total EDDs the field should have accumulated based on the forecast model. If possible, we will then correct ("reset") the forecast model at each stage and compare the new forecast to the original.6) Prepare Final Report:A final report will be prepared, detailing project objectives, work performed, results obtained, and estimates of technical feasibility.