DEVELOPMENT OF A NUTRIENT BUDGET APPROACH AND OPTIMIZATION OF FERTILIZER MANAGEMENT IN WALNUT

Transcription

1 DEVELOPMENT OF A NUTRIENT BUDGET APPROACH AND OPTIMIZATION OF FERTILIZER MANAGEMENT IN WALNUT Katherine Pope, Theodore DeJong, Patrick Brown, Bruce Lampinen, Jan Hopmans, Allan Fulton, Richard Buchner, Joseph Grant & Emilio Laca ABSTRACT With imminent regulations stemming from findings of high fertilizer-related nitrate contamination of groundwater and many orchard management expenses on the rise, it is critical that California walnut growers are applying their fertilizer efficiently, to reduce nitrate leaching and wasted resources. But much of the information necessary to optimize fertilizer usage is lacking. The timing and scale of nutrient needs on typical soils at today s high yields have not been quantified, nor have the assessment tools (e.g. critical values) for today s cultivars. This project aims to quantify the monthly nutrient needs of walnut orchards, estimate soil nutrient losses and contributions, improve grower nutrient assessment techniques like critical values and leaf sampling, and communicate findings on opportunities for improved nutrient efficiency with a decision support mobile application, publications and presentations. In this fourth year of a four year project, nutrient content results were received and analyzed for all 2015 samples. Results for total NPK needs per ton of nuts and monthly nut requirements for are detailed below. The last piece of the whole orchard nutrient budget NPK accumulation in different perennial tree parts is pending removal of three mature Chandler trees in Tehama County in the spring of Soil nutrient loss data has been analyzed and found to have severe quality issues. These data are currently being reviewed to try to extract some useful information and lessons. Nitrogen leaf status assessment tools were revisited this year with the three years of data. The optimum number of trees from which to collect leaves and distance between trees was estimated for more accurate, less time-consuming leaf sampling. A first attempt was made at creating a model to predict mid-summer leaf nitrogen levels based on May leaf nutrient status. The best model is presented below, but should be considered a first draft, as a large project specific to this modeling was launched in 2016, to gather data from an addition ~25 orchards to improve model predictions. OBJECTIVES 1. Develop a phenology- and yield-based nutrient demand model for walnut. 2. Determine the contribution of, and losses by, soil of nutrients for tree growth. 3. Validate current leaf critical values and determine if nutrient ratio analysis provides useful information to optimize fertility management. 4. Fine-tune sampling protocols to more accurately reflect the true nutrient status of an orchard block and to enable early season tissue sampling. 5. Integrate phenology, weather and orchard-specific details into a phenology-based nitrogen budget decision support tool (online and mobile application) and Best Management Practices publications. California Walnut Board 1 Walnut Research Reports 2016

2 SIGNIFICANT FINDINGS Results indicated walnuts accumulate less N per ton of harvested nuts than previous research indicated, closer to 30 lbs than 40 lbs N per ton. This finding was fairly consistent over the three years, though 2015 harvest N was higher than 2013 and N accumulation in fruit is fairly evenly distributed over the course of the growing season. Fruit P accumulation is steady through most of the growing season but tapers in September. Fruit K accumulation is fairly steady until September, when significant amounts of K is left in the field in hulls. When sampling leaves to assess orchard nitrogen status, o 23 trees should be sampled from trees o 30 meters apart or more (roughly every 4 th or 5 th tree, based on current common spacing) o This will ensure the test results are within 0.1% of the true orchard N status 95% of the time. Early season N status is a very strong indicator of mid-summer N status. May leaf N predicted 51% of the variability in July leaf N (Model D). The best predicting model, Model B, included numerous additional predictors and barely improved model fit, with R 2 = PROCEDURES (Abridged. For more details see 2014 or 2015 proposal) 1. Nutrient Demand Model: a. Catkins, Leaf and Fruit Demand: i. Samples were taken from 10 trees from Chandler and Tulare orchards at 3 sites per cv. (near Los Molinos, Linden, Hanford) in the last 7 days of each month and analyzed for N, P, K in 2013, 2014 and 2015 ii. Catkins: At senescence; Leaves: Apr-Nov; Fruit: May-Sept/Oct in 2013, 2014 and 2015 b. Perennial Parts Demand: i. For 3 of the 10 trees from a above, perennial parts were sampled for N, P, K ii. Sampled January (dormant), April (leaf-out), May (full leaf growth), July (conventional leaf-sample time), and November (post-harvest) for one annual cycle, spanning 2013 and 2014 iii. Roots, Trunk, Scaffold, Canopy branches, 2-3 year old Branches c. Yield: The 10 sampled trees were harvested for individual yield in 2013, 2014 and 2015 d. Phenology: Nut characteristics and weight were measured at each sampling 2. Soil nutrient losses a. Soil texture analyses were performed for variability in the Delta Chandler site. b. Equipment was installed between March and April 2014 at three locations in the orchard. At each location, soil moisture, soil temperature, EC and nitrogen species were monitored within and below the root zone (depths: 30cm, 90cm, 150cm, 210cm and 280 cm). Tensiometers were installed below the root zone in order to estimate leaching (depths: 260cm and 300cm). c. Suction lysimeters were sampled once a week to every other week, depending on California Walnut Board 2 Walnut Research Reports 2016

3 rainfall and irrigation timing. Soil solution was analyzed for nitrate, ammonia and total dissolved nitrogen (TDN) content. 3. Assessment Tool Refinement: CVs and Nutrient Ratios Leaves from 10 trees from Obj 1a.iii sampled in May and July are being analyzed for S, Ca, Mg, B, Zn, Fe, Mn, Cu in addition to N, P, K. Values are being compared with true individual tree yield, as well as PAR- and LAIadjusted individual tree yield from Obj. 1c to validate or revise critical values 4. Assessment Tool Refinement: Sampling Protocol Leaves from 30 trees (20 additional plus 10 from Obj 3) were sampled in May and July for N, P, K. Spatial statistics are being applied to nutrient content from the 30 trees to quantify inter-tree and intra-orchard variability. May leaf nutrient content will be compared with July leaf content to build a predictive model. 5. Lessons and Tools Dissemination: The majority of this work will be completed in 2016 when data has been analyzed, models have been developed and conclusions have been drawn. See proposals for more details. RESULTS 1. Nutrient Demand Model: A. Annual Nutrient Content per Ton of Harvested Nuts Nuts were collected at harvest, as well as stuck hulls. Nutrient content of nuts at harvest were received as dry weight percent nutrient content. This amount was transformed to pounds of nutrient per ton of in-shell nuts at 8% moisture. Most nuts lacked hulls when harvested. The weight of all hull at harvest was estimated as the difference between average late August fruit weight (shell, meat and hull) and average harvest nut weight (shell and meat). This was multiplied by stuck hull nutrient content to estimate pounds of NPK in the hulls at harvest for every ton of nuts. Because some hulls are left in the field, the Harvest NPK columns in the tables below represents NPK use of the fruit, and is slightly higher than NPK removed from the field. This is mostly pertinent for K. The N and P content of the hulls is quite low. Table 1, 2 and 3 show the N, P and K removed, respectively, from a given orchard in one ton of harvested in-shell nuts and associated hulls in the Harvest column. Total nitrogen use varied from lbs in 2013, lbs in 2014 and lbs in 2015 at harvest per ton of in-shell harvested nuts. B. Monthly Nutrient Allocation Fruit Fruit (kernel, shell and hull) were collected monthly, weighed and analyzed for nutrient content. Monthly NPK accumulation per ton of eventually harvested nuts was estimated as follows: The average weight of the nuts at harvest for each site-year was used to estimate the number of nuts in a harvested ton. Number of nuts per harvested ton was then multiplied by the average weight of fruit in a given month to get, for example, the total May weight of the 73,004 nuts that would eventually produce a ton of nuts at the Central Chandler site in This in-month weight per California Walnut Board 3 Walnut Research Reports 2016

4 eventual ton of harvested nuts was then multiplied by the percent NPK content in that month to give the pounds of NPK allocated to the fruit in a given month (see Tables 1, 2 and 3). June and July 2015 nut weights were measured on a faulty scale and thrown out. Weights were estimated as the site-month average for 2013 and Blank values are missing because of orchard entry problems. July and August %N results for many sites was much higher in 2015 than previous years. This is reflected in the higher numbers shown in Table 1. These high samples (40 of 120) were re-analyzed twice. The %N did not change significantly under repeated testing. Figure 1 shows this accumulation on a monthly basis, averaged over all sites, cultivars and years for nitrogen. Nitrogen accumulation in the fruit for every eventually harvested ton ranged from lbs in May, lbs in June, lbs in July, lbs in August and lbs at harvest. N accumulation tracked with latitude, cultivar and management style, with southern sites that leafed out earlier generally accumulating more nitrogen early in the season, and southern and central sites, which were more aggressively managed, accumulating more nitrogen over-all. Phosphorus accumulation in the fruit for every eventually harvested ton ranged from lbs in May, lbs in June, lbs in July, lbs in August and lbs at harvest. Potassium accumulation in the fruit for every eventually harvested ton ranged from lbs in May, lbs in June, lbs in July, lbs in August and lbs at harvest. California Walnut Board 4 Walnut Research Reports 2016

5 Table 1. Nitrogen allocated per month to the fruit for every 1 ton of harvested nuts. Nitrogen (N) (Lbs Per End-of-Season Harvested Ton) Year Region Cultivar May June July Aug Harvest 2013 North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare Chandler Mean Tulare Mean Grand Mean California Walnut Board 5 Walnut Research Reports 2016

6 Table 2. Phosphorus allocated per month to the fruit for every 1 ton of harvested nuts. Phosphorus (P) (Lbs Per End-of-Season Harvested Ton) Year Region Cultivar May June July Aug Harvest 2013 North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare Chandler Mean Tulare Mean Grand Mean California Walnut Board 6 Walnut Research Reports 2016

7 Table 3. Potassium allocated per month to the fruit for every 1 ton of harvested nuts. Potassium (K) (Lbs Per End-of-Season Harvested Ton) Year Region Cultivar May June July Aug Harvest 2013 North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare North Chandler Central Chandler South Chandler North Tulare Central Tulare South Tulare North Chandler Central Chandler * South Chandler * North Tulare Central Tulare * South Tulare * 41.9 Chandler Mean Tulare Mean Grand Mean *These outlier numbers are being re-examined. California Walnut Board 7 Walnut Research Reports 2016

8 Figure 1. Nitrogen accumulation by month for every ton of eventually harvested walnuts. Average of , Chandler and Tulare, 3 sites per cultivar Nitrogen Accumulation Per Month ( , Chandler & Tulare) Previous Accumulation In-Month Accumulation Lbs N per Ton Nuts May June July Aug Harvest [2. Soil Nutrient Losses and 3. Critical Values and Nutrient Ratios continue to be analyzed] 4. Assessment Tool Refinement: Leaf Sampling Protocol A. Within Orchard NPK Leaf Variability i. Distance Between Trees to Sample To answer the question of how far apart sampled trees should be when collecting leaf samples, the spatial covariance (also called semivariance) of the average tree leaf %N were analyze. The minimum distance necessary between trees for them to be considered statistical independent can be estimated by analyzing the semivariance of nutrient concentrations in the trees. If two sampled trees are not independent, no new information is gained by sampling both of them. The typical semivariance graph shows an increasing semivariance as distance increases. The distance at which the curve (almost) reaches an asymptote is called range and it is the distance beyond which leaf N is no longer correlated between trees (i.e. trees have independent %N leaf California Walnut Board 8 Walnut Research Reports 2016

9 values). The sampling grid had all trees at least 30 meters apart, based on similar research done in almonds, thus this analysis cannot say whether trees closer than 30 meters would be independent. It can only say whether trees at 30 meters or great distance from each other are independent. An exponential variogram without nugget was used for the data. Figure 2 shows that in most cases it was clear that the range was longer than the shortest lag measured. Figure 2. Variograms for walnut leaf N content in May (filled circles and solid line) and July (triangles and dashed line). California Walnut Board 9 Walnut Research Reports 2016

10 ii. Number of Trees to Sample The variances of nitrogen values from leaves at each site were analyzed to estimate how many trees need to be sampled from an orchard to accurately estimate the nitrogen status of that orchard. More specifically, leaves from a 30 trees grid sampled in May and July were used to estimate how many trees have to be sampled and pooled to estimate the mean N concentration within 0.1% with 80%, 90% and 95% confidence (Table 4). In other words, how many trees need to be sampled for the lab result to be within 0.1% of the true tree orchard nitrogen status 80% of the time, 90% of the time and 95% of the time, if the same orchard were sampled over and over and over. In the site-month-year combination with the highest variance (a measure of variability in leaf %N among the 30 trees), Central Tulare in May 2015, a sample size of 23 trees was necessary to calculate the mean May leaf N within 0.1 pph with 95% confidence. This result was corroborated using a parametric simulation to simulate 100,000 sampling events. In this analysis, a sample size of 20 trees resulted in the desired margin of error with 95% confidence. The same approach was used to estimate the sample size necessary for 90% confidence and 80% confidence. Repeating the calculations using the site-month-year combination with the highest variance, a sample size of 16 is necessary to calculate the mean May leaf N within 0.1 pph with 90% confidence. This was corroborated with 100,000 sample simulations which showed with sampling 14 trees resulted in the desired accuracy of 0.1 pph with 90% confidence. Corresponding sample sizes for 80% confidence are 10 trees based on the variance and 8 trees based on simulation of 100,000 sampling events. Table 4. Sample size to achieve different confidence levels in %N leaf analysis results Confidence Level Tree Sample Size Necessary 80% 10 90% 16 95% 23 B. Early Season N Predictive Model Multiple statistical models were constructed and tested using May leaf nutrient values (N, P, K, S, Ca, Mg, B, Zn, Fe, Mn, Cu) to predict July leaf nitrogen levels. The first model (Model A) including nitrogen, phosphorus and potassium concentrations in May as well as all two-way interactions between nutrients and nutrient quadratic fixed effects. Effect of soil series and the interactions of soil with nutrients were also included. Random effects were year, orchard, and orchard by year. Once the model was fitted with the data and unsupported variables were dropped, the final model included only the linear fixed effects of each element concentration in May, plus random effects for year, orchard and orchard by year combination. The coefficient of determination for this model was R 2 = The second model (Model B, Figure 5) started with all the variables in the first model and added cultivar interaction and days between sampling as potential variables. Once the model was fitted California Walnut Board 10 Walnut Research Reports 2016

11 with the data and unsupported variables were dropped, the final model included the linear fixed effects of N, P and K in May, cultivar, days between May and July sampling, P-K interaction, and interactions between N-cultivar, P-cultivar and K-cultivar, plus random effects for year, orchard and orchard by year combination. The coefficient of determination for this model was R 2 = This was the best fitting of all models. Figure 5 shows the bootstrap distribution of the predictions using this model, by site, along with a green line that shows the actual measured value at that site and year. It can be seen that the model predicts leaf July %N well for most sites and year, but poorly for Central Chandler 2013 (dc_13) and North Chandler 2014 and 2015 (nc_14, nc_15). The third model (Model C) attempted to predict change in %N from May to July, as opposed to the July %N value using the linear fixed effects of each P and K in May, plus random effects for year, orchard and orchard by year combination. Once fitted, only the intercept and the random effects were found to be significant. The coefficient of determination for this model was R 2 = The fourth and final model (Model D) attempted started with all the nutrients measured in May (N, P, K, S, Ca, Mg, B, Zn, Fe, Mn, Cu) with potential cultivar interaction, ratios of N:Ca and N:P, and random effects as in other models. Once the model was fitted with the data and unsupported variables were dropped, the final model only included N in May. The coefficient of determination for this model was R 2 = The data used to fit these models are being included in a larger project launched in 2016 funded by CDFA FREP, which will include an addition sites. Additional analysis to find more accurately predicting models will continue with those additional data. California Walnut Board 11 Walnut Research Reports 2016

12 Figure 3. Bootstrap distribution of prediction based on Model B. The vertical green lines show the observed leaf %N in July. The bars show the density of distribution of predictions under repeated simulations of the model and the data. California Walnut Board 12 Walnut Research Reports 2016

13 DISCUSSION 1. Nutrient Demand Model: A. Annual Nutrient Content per Ton of Harvested Nuts Observed total N content in one ton of nuts (Table 1) is lower than that observed by Weinbaum et al. (1991) by about one quarter. At some sites, some nuts approach as much N accumulation as was previously reported, indicating that our larger sample size may show more of the variability than was captured by the six trees in Weinbaum s study. Additionally, our figures do not yet include N allocation for perennial tissue growth. It is likely that the upper range of N use in our sites will overlap with Weinbaum s figures once perennial tissue growth is added to the total nitrogen budget. We anticipate reporting these numbers in the Walnut Research Reports in 2017, though we have not requested funding to support the analysis. Total P content (Table 2) of walnuts have not been the topic of much research to date, so there is little information to compare our findings against. Phosphorus content is small in comparison with nitrogen and potassium, and have been stable over the two years of data analyzed so far. Total K needs were researched by Olson (1991). They found highly variable K content in the hulls, as we have observed to date. B. Monthly Nutrient Allocation Fruit Nitrogen accumulation in fruit was observed to be fairly evenly distributed over the course of the growing season for all cultivars and sites (Table 1). This is in keeping with research previously done in almonds in California, and research done on walnuts in Europe (Drossopoulos et al., 1996). This indicates that an even division of nitrogen application over the growing season may be a simple, straightforward approach to increasing nitrogen use efficiency, rather than asking growers to keep higher or lower percentages of use in mind for different months when dividing their applications. P accumulation was observed to be steady through most of the growing season but tapered in September (Table 2). K accumulation was observed to be fairly steady until September, when significant amounts of K are left in the field in hulls (Table 3). 4. Assessment Tool Refinement: Sampling Protocol A. Within Orchard N Leaf Variability i. Distance Between Trees to Sample Based on the variogram analysis, it appears that spatial correlation between trees is minor for lags greater than 30 m, and thus, effective sample size will be almost equal to number of trees sampled if trees are more than 30 m apart. Figure 2 shows that in most cases it was clear that the range was longer than the shortest lag measured (i.e. trees were independent at greater than 30 m apart). The only potential exceptions were the Central Chandler site in 2015 (dc_15) and the Northern Tulare California Walnut Board 13 Walnut Research Reports 2016

14 site in 2013 (nt_13). Overall, these data indicate that spatial correlation is limited between trees that are more than 30 m apart. In other words, when sampling leaves to assess orchard nitrogen status, trees should be 30 meters apart or more - roughly every 4th or 5th tree, based on current common spacing. ii. Number of Trees to Sample Variance of leaf N within orchards and years was very small. Very high accuracy and precision to estimate the mean for an orchard can be achieved with sample sizes of 23 trees for a given orchard or block, assuming fairly uniform growing conditions. In other words, when sampling leaves to assess orchard nitrogen status, sampling 23 trees will ensure the test results are within 0.1% of the true orchard N status 95% of the time. Six terminal leaflets were collected from the periphery of each tree at a height 6-8 feet. If there are significant difference in soil texture within an orchard block, the lessons from this analysis are not applicable and two separate groups of samples should be taken. B. Early Season N Predictive Model Numerous combinations of variables associated with each orchard site were used to attempt to predict mid-summer leaf %N using samples taken in May. Variables used in testing included May leaf N, P, K, S, Ca, Mg, B, Zn, Fe, Mn, Cu, cultivar, soil type, time between May and July sampling and interactions among these variables. It was evident from Model D that early season N status is a very strong indicator of mid-summer N status. May leaf N predicted 51% of the variability in July leaf N (Model D). The best predicting model, Model B, included numerous additional predictors and barely improved model fit, with R2 = This modeling work was done with only 16 sets of data (3 years x 6 sites, minus 2 because we didn t have southern sites in May 2013). A rough rule of thumb for building complex multivariate models those necessary for these predictions is to use a minimum of 30 sets of measurements. The model building in this project has been leveraged to inform a large predictive model project that started in 2016 with CDFA FREP funding in which a larger numbers and diversity of sites will be sampled. REFERENCES Drossopoulos J.B., Kouchaji G.G., Bouranis D.L. (1996) Seasonal dynamics of mineral nutrients by walnut tree fruits. Journal of Plant Nutrition 19: DOI: / Olson B. (1991) The effects of various potassium levels on Chandler walnut trees, yield and nut quality, Walnut Marketing Board. Weinbaum S.A., Muraoka T.T., Catlin P.B., Kelley K. (1991) Utilization of fertilizer N by walnut trees, Walnut Marketing Board. pp. 18. California Walnut Board 14 Walnut Research Reports 2016