Source: CERES IMAGING, INC. submitted to NRP
REMOTE SENSING OF OPTIMAL WATER AND NUTRIENT MANAGEMENT FOR CROP PRODUCTION
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
Reporting Frequency
Annual
Accession No.
1009676
Grant No.
2016-33610-25442
Cumulative Award Amt.
$100,000.00
Proposal No.
2016-00921
Multistate No.
(N/A)
Project Start Date
Aug 1, 2016
Project End Date
Jul 31, 2017
Grant Year
2016
Program Code
[8.4]- Air, Water and Soils
Recipient Organization
CERES IMAGING, INC.
1155 INDIANA ST
SAN FRANCISCO,CA 94107
Performing Department
(N/A)
Non Technical Summary
Precision irrigation and fertilization management are critical to maximizing productivity while minimizing resource use. Insufficient irrigation and fertilization reduces yields and quality, while excess irrigation wastes increasingly scarce water, and excess fertilization can lead to contamination of groundwater, soil and air. Optimizing these management decisions is difficult given the tools typically used to manage large farms. Most decisions about irrigation and fertilizer application on farms are made on the basis of field management experience and a small number of soil and leaf samples. While soil/leaf samples can be accurate, they usually have high operational cost and limited spatial coverage, leading to sub-optimal irrigation and fertilizer application for different areas within the field. Existing commercial aerial photography efforts can provide much better spatial resolution, but are difficult to translate into recommendations for field management.We aim to develop a method for using aerial imagery at select wavelengths to provide quantitative recommendations on amounts of resources to apply, and to estimate how crop yields will be affected. To develop this method, we will start by focusing on almonds, and collect both ground and aerial data during the course of the California almond growing season over a field with a variety of irrigation and fertilization treatment blocks. We will study how our aerial measurements relate to both water/fertilizer input, as well as year end yields. This method provides an avenue for inexpensively measuring water and nutrient needs over a large area, and offers potential for maintaining or increasing yields while decreasing resource use, and reducing associated environmental impacts.
Animal Health Component
80%
Research Effort Categories
Basic
(N/A)
Applied
80%
Developmental
20%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
10212121020100%
Knowledge Area
102 - Soil, Plant, Water, Nutrient Relationships;

Subject Of Investigation
1212 - Almond;

Field Of Science
1020 - Physiology;
Goals / Objectives
Our goal is to quantify water and nutrient status of plants via aerial imagery, which we will then use to provide clear, quantitative and high spatial resolution recommendations on water and fertilizer application to farmers. Throughout next year's growing season, we will simultaneously collect tissue samples, soil moisture measurements, and on-the-ground plant water status measurements, in conjunction with flyovers using our imaging system. With these data, we will develop an approach to quantify in-season water and nutrient needs from our aerial imagery, and integrate our crop chlorophyll and water content measurements to develop accurate models for real-time yield monitoring.Objectives:For a minimum of 10 dates in the 2016 California growing season, collect aerial data simultaneously with the following ground data for test almond fields in Kern County:soil moisture measurements using neutron probesstem water potential, using a pressure bombtree growth monitoring, using dendrometersFor a subset of 4 of the 10 dates in which we are collecting aerial data, we will also collect the following ground data:leaf nutrient content, including nitrogen, phosphorus, and potassium, using leaf tissue analysisleaf chlorophyll content, measured using a SPAD chlorophyll meterFully process all aerial data, including sensor calibration, atmospheric correction, mosaicking, georegistration, masking, and segmentation of canopy and soil.Extract tree level estimates of plant reflectances in selected visible/NIR bands, and temperature from thermal band, for all trees for which ground data was collected.Produce model relating visible/NIR reflectances, temperature, and water and nutrient needs, as determined from ground data.Use end of season yield data, and yield from past years, to build model relating aerial data and yields.
Project Methods
Data for the experiment will be collected simultaneously in the air and on the ground throughout the 2016 California almond growing season over an experimental field operated by the UC Cooperative Extension (UCCE). Ground measurements, which include soil moisture, stem water potential, tree diameter, leaf chlorophyll content, and leaf nutrient content (NPK), will be conducted in the standard way for tens of trees within the study orchard. Aerial measurements will be collected using our imaging system. Our system includes narrow spectrum imaging in the visible/NIR (10 nanometer bandpasses centered at 480, 550, 670, 700, and 800 nanometers), designed to capture the chlorophyll signature of trees; and a thermal camera that captures over a 7.5-13 micron band that is sensitive to plant temperature. Our aerial data collection is novel. The wavelength coverage of the system provides sensitivity to both water and nutrient stress, at a resolution smaller than a tree canopy. This allows comprehensive aerial assessment of plant status at a plant level, which can then be compared with measurements obtained on the ground. At the end of the growing season, we will collect tree level yields (and will have access to tree level yields for each year going back to 2008), as well as treatment block level yields for the field. Additionally, we will collect aerial data through the season for a different experimental almond field. For this second field, we will obtain end of year yields within individual blocks within the field, but will not have ground measurements during the season.From this season long data collection, we will have a combined ground/aerial dataset with high spatial and temporal resolution, well beyond any that has previously been analyzed in the academic literature. For the primary field, we will construct a table in which each date/tree combination will be a row (e.g., 50 trees sampled at 10 different dates will produce 500 rows), and in which columns will be soil moisture, stem water potential, tree diameter, leaf chlorophyll content, leaf nutrient content, canopy level reflectance for each of the 5 visible/NIR bands, and canopy temperature from the aerial thermal measurement. We will have an additional table of yields, with a row for each tree, and a column for each year.We will use these tables as training data for multiple multivariate analyses. The goal of these analyses will be prediction of ground stress measures using the aerial data alone, as well as year-end yield. In these, we will use only aerial data as independent variables, and will use various ground measurements as dependent variables to be predicted by the model. We plan to explore various approaches for modeling the data and performing regression, which we will evaluate based on their success at fitting these training data. We will do an additional analysis using the ground measurements of water status and nutrient status as independent variables to fit the observed year-end yield data.Based on conversations with customers, relative predictions that are good to +/- 300 pounds/acre and that would highlight causes of variation in yield would be an attractive product. To evaluate the success of our modeling efforts, we will compare the explanatory power of the aerial data as compared to the ground data in predicting yields at a tree level. As we collect tree level data for the entire orchard with our imagery, we can also use generate yield predictions for each treatment block within the field (of which there are 160), and compare those predictions to actual yields. Yields depend on a number of factors beyond inputs to the field, including winter rain, nighttime temperatures, pollination success, and pests. These factors are likely to not be apparent at a small scale level, however. Thus we will focus this year on modeling yield anomalies (difference from mean field yields) within the field. For this, we will aim to do better than +/- 300 pounds/acre in predicting differences in yield from the mean at the block level. We will also use observed yields from the second almond field (without in-season ground measurements of water or nutrient status) as a test dataset for our model relating aerially measured visible/NIR/thermal data to yield, and aim to achieve the same level of accuracy in this second field.Through our partnership with the UCCE, we will share the results of the trial at monthly workshops that the UCCE hosts, explaining the methodology, as well as applications of our research.

Progress 08/01/16 to 07/31/17

Outputs
Target Audience:We organized a small conference, held in February 2017, to share the results with some members of the University of California affiliated almond research community, as well as solicit feedback and suggestions for next steps. We also sharedresults withcommerical growers at the Almond Board Conference in December, 2016. The improved understanding of how to relate measured temperatures and reflectances to management decisions like applied water were used to improve the quality oftheproducts we were already providing to customers, most importantly maps that show water stress across whole orchards. These maps have been provided over ~500,000 unique paid acres over the past year. Changes/Problems:Our research design depends upon experimental almond orchards managed for varying levels of irrigation and fertilizer application, and is premised on the assumption that yields will respond to inputs like water or fertilizer. However, this is not the case for the Tehama County experimental orchard, as we later learned. Avery uncommonly used rootstock on the nonpareil trees impacted the relationship between inputs to the crop and yield, and over multiple years now (up until 2017), the yield in this field has borne no relation to inputs. Thus, we were not able to use this field to test a model predicting yields based on water stress. We also learned of a malfunctioning pressure regulator leading to over watering and over fertilization in the eastern half of the one of the nutrient experimental plots. The excess nitrogen over half of the field distorted the correlation between plot averaged yield and plot averaged nitrogen or chlorophyll values, both for the plot, and consequently for the collective model predicting yields from chlorophyll indices. What opportunities for training and professional development has the project provided? Nothing Reported How have the results been disseminated to communities of interest?Our preliminary results were presented at the Almond Board Conference (held December 6-8, 2016), and elicited an enthusiastic response from growers and researchers in the California almond community. We held a small conference with researchers in the University of California system who work on almonds to discuss our results and get feedback on plans for future work. We plan to present the key findings from the report here at the 2017 Almond Board Conference. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? I. Impacts: Our goal is to quantify water and nutrient status of plants via aerial imagery, in order to ultimatelyprovide clear, quantitative and high spatial resolution recommendations on water and fertilizer application to farmers.To accomplish this, wefocused on California almonds, and collect both ground and aerial data during the course of the 2016 growing season over three experimentalfields with a variety of irrigation and fertilization treatment blocks. By comparing our aerial measurements to year-end yields and on-the-ground measurements of water/fertilizer inputs and status, wedemonstrate improved accuracy of aerial estimates of water status and yield over labor intensive and costly field measurements. The 2016 seasonproduced valuable results from our University of California division of Agriculture and Natural Resources (UCANR) and UC Cooperative Extension (UCCE) partners. We are able to demonstrate correlations between our aerial measures and the ultimate yield in almond production from these fields, as well as with traditional field measures of water availability and plant vigor used to manage water and fertilizer application. In fact, we found our aerial measure of water stress were a better predictor of yield (r2 = 0.50) then field measure of stem water potential (r2 = 0.13) or tissue nitrogen, and the relationship between applied water and conductance--a measure of water stress--were stronger still (r2 = 0.73).These experiments were performed in three fields located in the California Central Valley, a major center of agriculture production, and encompassed a range of weather conditions, almond tree varietals and age, and irrigation and fertilization treatments. Accurate estimates of crop physiological status are essential for growers to make optimal irrigation and fertilization decisions. Ceres Imaging currently offers two products designed to address this need. Both of these utilize band ratios of visible and near-infrared light to monitor relative leaf chlorophyll content.To further validate the accuracy of our products, we correlated our aerial measurements with leaf tissue samples of nitrogen content and total plot yield. We are able to accurately map spatial patterns of plot productivity based on these indices. However, there is considerable variability in the correlations between aerial measured chlorophyll content and yield. The water stress and chlorophyll maps we currently provide farmers highlight, on a plant-by-plant basis, areas of sub-optimal applications of water and fertilizer. However, we (nor anyone) are not yet able to quantify recommendations on specific amounts of resources to apply, or estimate how crop yields are being affected using imagery alone. Over the past year, however, we have made several advances towards this goalin the areas we initially identified as being key to success: accuracy, automation, flexibility and actionability. II. Outcomes: Below, we summarize our accomplishments for the 2016 research period. 1.We conducted 25 flights over experimental almond and walnut orchards in Tehama County, 11 flights over a Kern County almond experiment, and 3-5 flights over an additional 6 almond orchards also in Kern County. These flights allowed the acquisition of thermal and visible + near infrared (VNIR) imagery and coincided with ground data collectionby our UC partners. Ground data included stem water potential, total yields measured as both kernel weight and nut size, and leaf tissue concentrations. 2. Our partners collected leaf tissuefor analysis of nitrogen, phosphorus and potassium twice during 2016--in April and July--for a subsample of the Kern County experimental plots. April aerial imagery was used to identify 'strong' and 'weak' corners of each experimental block to identify appropriate sampling sites for composite tissue sampling. Leaf chlorophyll content was also sampled using a SPAD meter. 3. A key outcome of this research was the development and implementation of an atmospheric correction method that synthesizes publicly available meteorological data, standard atmospheric modeling, and simple physical laws. We also tested the calibration of our visible to near-infrared calibrationby flying a dark calibration tarp at regular intervals during the 2016 season. With our new atmospheric correction method and sensor calibration, we measured within 96% and 98% of the ground signal in the near-infrared and red, respectively. We also improved classification accuracies by refining our approach to separate tree canopy from soil by leveraginga deep learning technique well suited to analysis of imagery. We were able to apply the algorithms to images of full fields, successfully identify centers of tree canopies with a false positive rate of ~0.1% and a false negative rate of ~1% for almond orchards with typical plant spacing and density. This approach greatly improved the accuracy of automatically extracted physical values for every single tree in an orchard. 4.We used the improvements to extraction/classification mentioned above to not only automate extracting tree level temperature/reflectances for sample trees, but extended the method to accurately extract those data for all trees in a field. This greatly improved the quality of analysis we were able to provide, particularly those related to zonal analyses, such as average conductance within a certain irrigation zone in a field. 5.Tree-level measures of chlorophyll content and water stress were averaged over each plot and correlated to ground measures of stem water potential, applied water, leaf tissue nitrogen concentrations and the final yield from the 2016 growing season. Our analysis shows a strong relationship between the amount of applied water and our aerially measured stomatal conductance over the course of the year, as well as with field measures of stem water potential. Chlorophyll indices extracted for corresponding sample trees at the time of sampling however, only show a weak relationship with tissue nitrogen concentrations. 6. We are able to model year-end yield using conductance measurements from the Kern County experimental irrigation fields with reasonable accuracy (r2 = 0.50). However, the predictive power of applyingthis model to the nutrient management fields is minimal.

Publications


    Progress 08/01/16 to 07/31/17

    Outputs
    Target Audience: Nothing Reported Changes/Problems:We have been delayed in research by the time it has taken our partners, who collected all ground data, to compile their observations and share with us. The Tehama County field shows a very weak relationshipbetween inputs (water, in particular) and yield. Thus, even though our aerial measurements show good correspondence with applied water, the relationship with yield is also quite weak. As this was the field for which we had the densest sampling of data through the year, it is a setback to building a yield model, but we are hopeful that it is a setback we can overcome through eventually obtaining yield figures for other fields we flew during the 2016 season. 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 early results with researchers we are working with, and also presented a couple of results at the Almond Board Conference, which is attended by manyalmond growers and researchers. What do you plan to do during the next reporting period to accomplish the goals?As we continue to receive ground data for additional fields we flew in 2016, we will use the SPAD and leaf tissue samples to assess how those factors related to year end yields. Those data will be crucial to understanding howVNIR reflectance data can be used to assess leaf chlorophyll content and nutrient status. Once we have all ground data for the 2016 season, we will attempt to build a model for assessing applied water to date that can be applied to any field, and test that on other fields for which we have applied water data, with the end goal of usingthat to inform quantititave irrigation recommendations. We will also explore a yield model built from data for a subset of the fields we flew in 2016, and test it on the remainder of the fields. Mostly, we plan to execute the plan laid out in our project initiation. One challenge is understanding the yield results in the Tehama County orchard. In part to discuss this issue, and in part to discuss experimental design for data collection in 2017,we have scheduled a small conferencethat includes the agronomists with whom we are working, and various other members of the UC-affiliated almond research community.

    Impacts
    What was accomplished under these goals? 1. We achieved obtained aerial data in conjunction with ground measures of water stress over >10 dates, for one field in Kern County and one field in Tehama County. 2. We are still awaiting final compliation of tissue samples and SPAD readings. We have readings for two dates for one field in Kern County, but the agronomists were working with will be able to provide additional results. 3. All aerial data obtained has been processed according to the planned outline. 4. We have extracted tree level data for all trees in the main Kern County orchard and Tehama County orchard. We are awaiting locations of sample trees for other orchards in Kern County for which we will eventually receive ground data. 5. Using the available measures of water stress, we have begun work on relating thermal/VNIR data to more concrete assessments of water needs. We are still awaiting leaf tissue samples that would allow work towards nutrient management. 6. Preliminary work on building a yield model using in-season aerial temperature and VNIR measuremements has shown promise. Year end yields for one of the almond orchards showed a strong relationship between temperature, NDVI, and yield. The Tehama County almond orchard we studied showed very little relationship between yield and any aerial measurements (or between yield and applied water, for that matter), and we're working to understand why. We are also awaiting additional yield data for other almond orchards we observed during the 2016 season.

    Publications


      Progress 08/01/16 to 01/31/17

      Outputs
      Target Audience: Nothing Reported Changes/Problems:We have been delayed in research by the time it has taken our partners, who collected all ground data, to compile their observations and share with us. The Tehama County field shows a very weak relationshipbetween inputs (water, in particular) and yield. Thus, even though our aerial measurements show good correspondence with applied water, the relationship with yield is also quite weak. As this was the field for which we had the densest sampling of data through the year, it is a setback to building a yield model, but we are hopeful that it is a setback we can overcome through eventually obtaining yield figures for other fields we flew during the 2016 season. 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 early results with researchers we are working with, and also presented a couple of results at the Almond Board Conference, which is attended by manyalmond growers and researchers. What do you plan to do during the next reporting period to accomplish the goals?As we continue to receive ground data for additional fields we flew in 2016, we will use the SPAD and leaf tissue samples to assess how those factors related to year end yields. Those data will be crucial to understanding howVNIR reflectance data can be used to assess leaf chlorophyll content and nutrient status. Once we have all ground data for the 2016 season, we will attempt to build a model for assessing applied water to date that can be applied to any field, and test that on other fields for which we have applied water data, with the end goal of usingthat to inform quantititave irrigation recommendations. We will also explore a yield model built from data for a subset of the fields we flew in 2016, and test it on the remainder of the fields. Mostly, we plan to execute the plan laid out in our project initiation. One challenge is understanding the yield results in the Tehama County orchard. In part to discuss this issue, and in part to discuss experimental design for data collection in 2017,we have scheduled a small conferencethat includes the agronomists with whom we are working, and various other members of the UC-affiliated almond research community.

      Impacts
      What was accomplished under these goals? 1. We achieved obtained aerial data in conjunction with ground measures of water stress over >10 dates, for one field in Kern County and one field in Tehama County. 2. We are still awaiting final compliation of tissue samples and SPAD readings. We have readings for two dates for one field in Kern County, but the agronomists were working with will be able to provide additional results. 3. All aerial data obtained has been processed according to the planned outline. 4. We have extracted tree level data for all trees in the main Kern County orchard and Tehama County orchard. We are awaiting locations of sample trees for other orchards in Kern County for which we will eventually receive ground data. 5. Using the available measures of water stress, we have begun work on relating thermal/VNIR data to more concrete assessments of water needs. We are still awaiting leaf tissue samples that would allow work towards nutrient management. 6. Preliminary work on building a yield model using in-season aerial temperature and VNIR measuremements has shown promise. Year end yields for one of the almond orchards showed a strong relationship between temperature, NDVI, and yield. The Tehama County almond orchard we studied showed very little relationship between yield and any aerial measurements (or between yield and applied water, for that matter), and we're working to understand why. We are also awaiting additional yield data for other almond orchards we observed during the 2016 season.

      Publications