Progress 09/01/20 to 08/31/22

Outputs
Target Audience:The project results are targeted to help growers and grower-processors operate more efficiently, reducing waste and saving on their production costs in a number of direct ways: First, by determining earlier if fields will not produce sufficient yield, or within the needed time window, to justify additional expenditures; then by helping them allocate worker resources more cost-effectively by directing harvest crews where they are most needed and when; and finally, by helping them plan and manage their supply against fluctuations related to weather, pests, weeds or disease outbreaks. Our overall strategy for commercializing the project results has been to demonstrate the improved forecasting to our current customers, who use our web and mobile applications for forecasting based on plant population counts we derive from aerial imagery. We demonstrated our new capabilities by comparing our forecasted harvest dates and yields to their actual schedules and yields. In the course of developing the forecasting capabilities in cooperation with partner-customers, we identified several additional but related customer needs we are addressing with additional development. As a result, we are providing a set of integrated services that help make the fresh produce production process more efficient and therefore profitable. As we further develop the targeted capabilities, we will begin sharing the concepts with additional candidate customers in the fresh produce industry, in line with our commercialization objectives. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We contracted interns to perform in-field data collection. One of the interns became a UAV pilot and provides us aerial imagery collection. ? How have the results been disseminated to communities of interest?We have shared customer-specific results to date with the associated partner-customers: Bonipak Produce, Bonduelle Fresh Americas, the celery partner-customer that prefers to remain anonymous, Cal Giant and Taylor Farms. ? What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? The project developed precision crop maturity forecasting models for three key fresh produce crops: iceberg lettuce, celery and strawberries. Current approaches to forecasting harvest dates are based on historical planting and harvest information and do not capture the significant variabilities from local weather, soil conditions and farmer practices. Our approach was to supplement forecasted weather data with unique modeling techniques and computer vision analysis of aerial imagery collected over fields during the crop growing cycle. During the second phase of the project, we refined the methods that were proven for iceberg lettuce during the first phase, acquiring more data and validating the approach for different varieties and geographic locations. In addition, we extended the techniques from lettuce to celery and strawberries, which were identified by our grower partners as priorities for improved forecasting. Task 1 - Phase II Data Collection for Iceberg and Other Crops Subtask: Arrange with our grower partners and other growers to obtain additional historical planting and harvest data. We began Phase II with the partner-customer with which we worked during Phase I, Bonipak Produce. During Phase II we also added several other partner-customers. These include Bonduelle Fresh Americas for iceberg and romaine lettuce grown to be harvested by weight for processing, a major industry producer of celery that prefers remaining anonymous, and Cal Giant for strawberries. Subtask: Collect ground truth measurements of physical plant properties. We contracted interns to perform substantial in-field data collection. Subtask: Coordinate the timing and locations of ground truth plant measurements with aerial imagery acquisition. Coordination of locations for ground truth measurements and imagery acquisition was achieved by marking sections of fields where GT measurements were taken using colored PVC pipes that were visible in the aerial imagery. Subtask: Evaluate the use of additional tools to assess crop height, including digital surface models derived during the photogrammetric process. We identified plant canopy height as an important variable to include in our forecasting model, and we used photogrammetric processing of aerial photos to derive canopy heights. We also evaluated UAV-mountable rangefinders as an alternative to photogrammetry for canopy height measurements. Task 2 - Test and Refine Prototype Iceberg Forecasting Models Subtask: Accumulate EDDs using historical weather data for additional, historical grower planting and harvest data from Task 1 and evaluate values for different geographic regions, different iceberg varieties and different growers. We focused on the growing regions of Bonduelle Fresh Americas, for iceberg lettuce harvested by head weight; not head size. Subtask: Use additional ground truth plant data from Task 1 to plot against accumulated EDDs and evaluate goodness of fit. We collected GT and weather data for bulk romaine for Bonduelle, which we used to develop a preliminary bulk romaine forecasting model. Subtask: Develop an equivalent model for forecasting head weight. For forecasting by weight, we focused on bulk romaine and we collected several plant weight measurements, including whole plant fresh weight, heart weight (having stripped excess leaves), maximum and minimum harvestable weights, and radial circumferences of the whole plant and core. We employed a grower-specified average head weight per field to estimate total yield weight per field from imagery-derived plant population counts. Task 3 - Operationalize Iceberg Forecasting Subtask: Identify tools and infrastructure needed to scale up forecasting. One area we addressed to reduce image acquisition costs was improved flight logistics. We refined the command and control system developed during the first year of the project to handle scalable scheduling and processing of aerial imagery for many fields. We also made improvements to our mobile app for in-field data collection to incorporate adjustments to harvest yield estimates based on reductions from impacts of pests, weeds and diseases. Subtask: Integrate preliminary toolset and infrastructure into GeoVisual's processing pipeline to support lettuce forecasting operationally. We integrated the iceberg forecasting capabilities developed to date with this project into our processing pipeline, with results available to our customers in web and mobile applications. Subtask: Standardize some customer inputs (e.g. formats for field boundaries; planting data). One of the key challenges we faced working with the fresh produce industry has been a lack of standards in how data is stored and shared, particularly for data associated with fields and planting. Subtask: Update GeoVisual's SeedGreen web and mobile applications with additional GUI components for customer data input and decision making based on forecasting. We developed and refined a number of different views on the data and the analytics we developed under this project. Task 4 - Extend Forecasting to Other Customer-Priority Crops Subtask: Accumulate EDDs using historical weather data for historical grower planting and harvest data and evaluate values for the priority crops, ideally for different geographic regions, varieties and growers.For celery and strawberries, EDDs were found less effective than for iceberg lettuce, but we operationalized its use for determining when to collect aerial imagery for consistent crop maturity estimates. Subtask: Use ground truth plant measurements from Task 1 to plot against accumulated EDDs and evaluate goodness of fit to assess which measurements provide reliable model parameters and have strong correlations to measurements that can be made from aerial imagery. For celery, we acquired field measurements of several plant characteristics we considered to be of possible relevance for yield forecasting. For strawberries, we selected several growth stages for observation over time. Subtask: Develop models using a methodology similar to iceberg lettuce for the additional priority crops: combine values, GAC and ground truth measurements and other imagery-derived measurements to develop a preliminary capability to forecast optimal harvest dates and associated yields. Our focus for celery was a capability to forecast a celery field's harvest yields for a specified harvest date. We found the forecasting of such optimal harvest dates to be challenged not only by the inconsistencies observed in accumulated EDDs values, but also by significant variability in the days between planting and harvest as reported by our partner-customer. ?Our approach for strawberries involved three separate models. Model 1 takes as input UAV imagery and derives flower counts. Model 2 takes as input imagery dates, field location, plant variety and weather data and predicts when flowers will mature into later growth stages. The results of Model 1 and 2 are fed into Model 3 to derive a yield estimate.

Publications

Progress 09/01/20 to 08/31/21

Outputs
Target Audience:The project results are targeted to help growers and grower-processors operate more efficiently, reducing waste and saving on their production costs in a number of direct ways: First, by determining earlier if fields will not produce sufficient yield, or within the needed time window, to justify additional expenditures; Then by helping them allocate worker resources more cost-effectively, by directing harvest crews where they are most needed and when; Finally, by helping them plan and manage their supply against fluctuations related to weather, pests, weeds or disease outbreaks. Our overall strategy for commercializing the project results is to demonstrate the improved forecasting to our current customers, who use our web and mobile applications for forecasting based on plant population counts we derive from aerial imagery. We will demonstrate our new capabilities by comparing our forecasted harvest dates and yields to their actual schedules and yields. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided? Nothing Reported How have the results been disseminated to communities of interest?We have shared customer-specific results to date with the associated partner-customers: Bonipak Produce, Bonduelle Fresh Americas, the celery partner-customer that prefers to remain anonymous, Cal Giant and Taylor Farms. What do you plan to do during the next reporting period to accomplish the goals?We will continue our work on forecasting for lettuce harvested by weight, celery and strawberries. The principal crop of one of the partner-customers using the service for iceberg is celery, so we anticipate developing the celery forecasting capabilities into an operational service as well. In the course of developing the forecasting capabilities in cooperation with partner-customers, we identified several additional but related customer needs we are addressing with additional development. Our strategy is to provide a set of integrated services that help make the fresh produce production process more efficient and therefore profitable. As we further develop the targeted capabilities, we will begin sharing the concepts with other candidate customers in the fresh produce industry, in line with our commercialization objectives.

Impacts
What was accomplished under these goals? This project is developing precision crop maturity forecasting models for several key fresh vegetable crops. Current approaches to forecasting harvest dates are based on historical planting and harvest information and do not capture the significant variabilities from local weather, soil conditions and farmer practices. Our approach is to apply forecasted weather data with a unique modeling technique and supplement with innovative computer vision analysis of aerial imagery collected over fields during the crop growing cycle. We are successfully commercializing the forecasting capability for iceberg, for the partner-customers with whom we are collaborating on the project. In addition, we are extending the techniques from lettuce to other crops that have been identified by our grower partners as priorities for improved forecasting: romaine, celery and strawberries. The proposed innovation will help growers and grower-processors produce more efficiently, reducing crop waste and saving on their production costs in a number of direct ways: first by determining earlier if fields will not produce sufficient yield, or within the needed time window, to justify additional expenditures; then by helping them allocate worker resources more cost-effectively, by directing harvest crews where they are most needed and when; and finally by helping them plan and manage their supply against fluctuations related to weather, pests, weeds or disease outbreaks. We are integrating results into our web and mobile applications, which our partner-customers currently use for production management and with less precise forecasting based on plant population counts. Task 1 - Phase II Data Collection for Iceberg and Other Crops ? Subtask: Arranged with our grower partner and other growers to obtain additional historical planting and harvest data. We began Phase II with the partner-customer with which we worked during Phase I, Bonipak Produce. During Phase II we also added several other partner-customers. These include Bonduelle Fresh Americas for iceberg and romaine lettuce grown to be harvested by weight for processing, a major industry producer of celery that prefers remaining anonymous, and Cal Giant for strawberries. ? Subtask: Collect ground truth measurements of physical plant properties. Principally we utilized university students as interns to perform in-field data collection. ? Subtask: Coordinate the timing and locations of ground truth plant measurements with aerial imagery acquisition. Coordination of locations for ground truth measurements and imagery acquisition was achieved by marking sections of fields where GT measurements were taken using colored PVC pipes that were visible in the aerial imagery. ? Subtask: Evaluate the use of additional tools to assess crop height, including digital surface models derived during the photogrammetric process. This subtask ended up being very relevant for our work on celery forecasting. We identified plant canopy height as an important variable to include in our forecasting model. Task 2 - Test and Refine Prototype Iceberg Forecasting Models For each of the planned subtasks listed as bullets below, we describe the progress made towards them during the reporting period. ? Subtask: Accumulate EDDs using historical weather data for additional, historical grower planting and harvest data from Task 1 and evaluate values for different geographic regions, different iceberg varieties and different growers. We focused on the growing regions of Bonduelle Fresh Americas, with the difference being that the forecasting model we focused on for Bonduelle is for iceberg lettuce harvested by head weight; not head size. ? Subtask: Use additional ground truth plant data from Task 1 to plot against accumulated EDDs and evaluate goodness of fit to assess which physical plant measurements provide reliable model parameters and correlate strongly with measurements derivable from aerial imagery. We collected GT and weather data for bulk romaine for Bonduelle, which we used to develop a preliminary bulk romaine forecasting model. ? Subtask: Develop an equivalent model for forecasting head weight. For forecasting by weight, we focused on bulk romaine and we collected several plant weight measurements, including whole plant fresh weight, heart weight (having stripped excess leaves), maximum and minimum harvestable weights, and radial circumferences of the whole plant and core. Task 3 - Operationalize Iceberg Forecasting ? Subtask: Identify tools and infrastructure needed to scale up forecasting. One area we addressed to reduce image acquisition costs was improved flight logistics. Our pilots made several minor improvements in the field to reduce the drone and sensor setup and teardown times between flights. ? Subtask: Integrate preliminary toolset and infrastructure into GeoVisual's processing pipeline to support lettuce forecasting operationally. We have integrated the iceberg forecasting capabilities developed to date with this project into our processing pipeline, with results available to our customers in web and mobile applications. ? Subtask: Standardize some customer inputs (e.g. formats for field boundaries; planting data). One of the key challenges we have faced working with the fresh produce industry has been a lack of standards in how data is stored and shared, particularly for data associated with fields and planting. ? Subtask: Update GeoVisual's SeedGreen web and mobile applications with additional GUI components for customer data input and decision making based on forecasting. We have developed and refined a number of different views on the data and analytics we are deriving under this project, based on feedback from our partner-customers about what they perceive as most valuable for improving their operational efficiency. Task 4 - Extend Forecasting to Other Customer-Priority Crops ? Subtask: Accumulate EDDs using historical weather data for historical grower planting and harvest data and evaluate values for the priority crops, ideally for different geographic regions, varieties and growers. We investigated the use of EDDs as a harvest forecasting technique in our development of forecasting capabilities for celery and strawberries. ? Subtask: Use ground truth plant measurements from Task 1 to plot against accumulated EDDs and evaluate goodness of fit to assess which measurements provide reliable model parameters and have strong correlations to measurements that can be made from aerial imagery. For celery, we acquired field measurements of several plant characteristics we considered to be of possible relevance for yield forecasting. For strawberries, we selected several growth stages for observation over time. ? Subtask: Develop models using a methodology similar to iceberg lettuce for the priority crops: combine values, GAC and ground truth measurements and other imagery-derived measurements to develop a preliminary capability to forecast optimal harvest dates and associated yields. Celery Yield Forecasting Our focus for celery was on the development of a preliminary capability to forecast a celery field's harvest yields for a specified harvest date, rather than attempting to also forecast the date of optimal yield. We found the forecasting of such optimal harvest dates to be challenged not only by the inconsistencies observed in accumulated EDDs values, but also by significant variability in the days between planting and harvest as reported by our partner-customer. Strawberry Yield Forecasting Our approach involves three separate models. Model A takes as input UAV imagery and derives flower counts. Model M takes as input imagery dates, field location, plant variety and weather data and predicts when flowers will mature into later growth stages. The results of Model A and Model M are fed into Model Y to derive a yield estimate.???

Publications