Impacts of Changing Postharvest Agricultural Practices on Abundance of Waste Grain in California's Central Valley

The Central Valley of California (Figure 1) supports one of the largest concentrations of wintering waterfowl in the world despite loss of over 90% of its historic wetlands (Heitmeyer et al. 1989; Fleskes et al. 2012). The 6–7 million waterfowl that winter annually in the Central Valley rely heavily on agricultural foods (Petrie et al. 2016). Rice fields that are flooded after harvest (winter-flooded) provide an estimated 52% of all the food energy available to dabbling ducks in the Central Valley from fall through spring, while rice provides an estimated 95% of the food energy available to geese between October and January (CVJV 2020). Agricultural practices that are favorable to wintering waterfowl in the Central Valley have allowed the region to maintain its continental importance to North American duck and goose populations, despite land use changes in the region. The cost of restoring winter-flooded rice fields to intensively managed wetlands that provide an equal amount of food is estimated at two billion dollars (Petrie et al. 2014).

Conservation planning for waterfowl in the Central Valley is the responsibility of the Central Valley Joint Venture (CVJV), a partnership established in 1988 under the auspices of the 1986 North American Waterfowl Management Plan (NAWMP 1986) to help conserve the continent's waterfowl populations and habitats. Many conservation Joint Ventures, including the CVJV, rely on energetic-based carrying-capacity models to predict the number of waterfowl an area can theoretically support and project the amount of foraging habitat needed to sustain populations at desired levels (CVJV 2006, 2020; Pearse and Stafford 2014; Williams et al. 2014). Accurate estimates of the biomass of waterfowl foods in each foraging habitat type are fundamentally important to the performance of these models (Williams et al. 2014; Petrie et al. 2016). Typically, field sampling studies that measure the abundance of important food items (typically seeds from a variety of moist soil plants or waste grain left after harvest of agricultural crops such as rice or corn) provide these estimates.

The 2020 CVJV implementation plan assumes that harvested rice fields provide 52% of the food energy available for dabbling ducks, while harvested grain corn provides 4%. Estimates have determined that managed seasonal wetlands provide the remaining 44% (CVJV 2020). On average, there are approximately 541,000 acres (218,935 ha) of harvested rice fields in the Central Valley of which an estimated 341,000 acres are winter-flooded. Managed wetlands and harvested grain corn fields total 196,000 and 34,000 acres, respectively (CVJV 2020). Estimates of the amount of food these habitats provide waterfowl rely heavily on accurate food biomass values as key inputs for the bioenergetic model. However, the amount of food provided by agricultural habitats is strongly dependent on postharvest practices. Monitoring changes in these postharvest practices, and understanding how these changes may impact waterfowl food supplies, is important to future planning efforts in the CVJV and elsewhere.

The harvesting process in both rice and corn fields removes the grains and leaves the rest of the plant behind. This results in a large amount of organic matter in the fields after harvest, which farmers must remove before the next crop can be planted (Bird et al. 2000). Prior to the early 1990s, California rice farmers used burning as the principal method of rice straw disposal. However, burning was largely phased out because of air quality concerns that led to a change in California State law (Connelly-Areias-Chandler Rice Straw Burning Reduction Act of 1991). Most rice farmers quickly transitioned to winter flooding, which provided an economical alternative for decomposing rice straw and, as a result, provided tremendous waterfowl benefits (Eadie et al. 2008). Now, three decades after the burn ban, roughly 63% of all harvested rice fields are winter-flooded each year (CVJV 2020).

Although the winter flooding of rice fields is highly favorable to waterfowl, the long-term feasibility of this postharvest practice depends on reliable and affordable water supplies. Rice farmers that do not winter-flood typically incorporate straw, which is left after harvest, into the soil using postharvest practices generally described as “dry incorporation” (Miller et al. 2010). There is considerable uncertainty regarding the impact of these postharvest practices on the abundance and availability of waterfowl food. The recent California drought from 2011 to 2016 greatly restricted the availability of surface water sources for winter flooding, and highlighted the long-term uncertainty around these water supplies. As a result, an increasing number of rice farmers may turn to dry incorporation as a means of eliminating rice straw as California continues to experience water shortages (Petrie et al. 2016). Dry incorporation relies on heightened contact between residual plant material and the soil to achieve decomposition of the material. We were concerned that the mechanical manipulation needed to produce such contact would cover seeds with the soil and effectively reduce the abundance of waste rice. The postharvest treatments implemented in both rice and corn fields differ slightly but serve the same goal—to remove or promote the decomposition of rice straw or corn stalks.

To date, researchers have focused considerable efforts on evaluating the abundance of waste rice, rice harvesting methods, combine efficiency, and bird use in major wintering areas for waterfowl in California and elsewhere in North America (Eadie et al. 2008; Manley 2008; Marty et al. 2020). Few studies have evaluated the effects of postharvest straw management on waste rice abundance (Manley et al. 2004; Kross et al. 2008; Havens et al. 2009). More challenging is that the Central Valley is the site of few studies on this topic (Miller et al. 1989, 2010), and no studies include all the methods that producers currently use. At the time of this study, the assumption was that grain corn provided 25% of the food energy available to waterfowl in the Central Valley, regardless if it was flooded (CVJV 2006). However, the most recent CVJV implementation plan reduced this estimate to only 4% (CVJV 2020). The earlier estimate relied on untested assumptions about the postharvest practices used in grain corn fields in California. In addition, research done outside of California formed the basis for estimates of the amount of corn remaining after harvest, so there were no published studies on the abundance of waste corn for this state. This lack of information highlights the need to update our assessments of the abundance of agricultural waste grains in the Central Valley that result from different postharvest practices.

Our objective was to quantify the abundance of waste rice and waste corn for waterfowl in the Central Valley across multiple postharvest treatments. We hypothesized that fields with a greater amount of mechanical manipulation would provide significantly less available waste grain. In addition, we sampled fields both before and after flooding to determine if flooding impacted food abundance. Sampling of dry fields has formed the basis of most waste grain estimates, with the assumption that all of the food resources will be similar at the time of flooding. Finally, we also compared two different sampling metrics. Traditional waste grain studies have used soil cores; however, a new method has been developed using line-transect methods (Halstead et al. 2011), but this method has not been compared to soil core estimates. Differences between the methods could influence both the design of future studies and the ability to compare estimates derived from previous studies.

The study area encompassed the 10 major rice and grain corn producing counties within the Central Valley of California (Figure 1). These counties contained 98% of the rice and 53% of the grain corn harvested in the state during 2016 (National Agricultural Statistics Service 2018). Although rice and corn are grown throughout the Central Valley, other counties were excluded for two reasons: 1) the county did not produce a significant proportion, at least 1% of the total acreage of that crop in California or 2) the county was outside a major wintering area within the boundaries of the Central Valley, and thus was not relevant to waterfowl habitat planning (CVJV 2006, 2020).

All field sites were on private lands and we obtained permission to work on each of the properties. In order to generate a list of willing producers we sourced a list of names and phone numbers from the University of California Cooperative Extension. We used that list, and additional producers identified through personal communications and word of mouth about the study, as our final list of participating producers. Once we obtained the list, we selected field sites opportunistically based on harvest date, location, and postharvest treatment. We do acknowledge that our field site selection was not truly random and there could be some bias in the sampling; however, our objective was to sample fields that were representative of the typical postharvest treatments in each region. Producers often use these postharvest treatments independently or in some combination (defined in Table 1); flooding always occurs in combination with at least one of the other postharvest treatments. When reporting the postharvest treatment used in a field, we only list the last operation conducted in that field before sampling occurred. We did not consider harvester header types during field site selection or analysis. Although producers do use stripper headers in California, they often own both conventional and stripper headers and will switch between the two based on field conditions during harvest.

The sampling area for rice comprised the six major rice-growing counties in the Central Valley (Figure 1). We sampled corn in San Joaquin, Sutter, Glenn, and Colusa counties. While both Sacramento and Solano counties were also corn-producing areas, we were unable to obtain access to any farms within those regions. We selected specific fields based on harvest date and postharvest treatment. We sampled all fields as soon as possible after postharvest treatments occurred and sampled all flooded fields within 1 wk of postharvest treatments to reduce the effects of environmental factors, seed deterioration, and seed depletion. We did not control for harvest dates when sampling fields, since the time between harvest and postharvest treatment is very unpredictable and we did not expect seed depletion to be significant. We did not sample organic or wild rice fields because their collective acreages account for less than 5% of the total state production (National Agricultural Statistics Service 2018). At the end of each harvest season, we reached out to farmers to obtain yield data from the fields that we had sampled.

We collected samples from August to December in 2016 and 2017 using two methods. The modified line-transect method employs a 6-m-long by 6.35-mm-wide plastic measuring tape with alternating 10-cm bands of red and white color (Halstead et al. 2011). We used a random number generator to select three random points within each field by choosing random distances from the edge of the field. We did not consider field size when selecting fields; however, if fields were larger than 80 ha we constrained the sampling area within the field to a maximum length and width of 880 × 880 m (Halstead et al. 2011) We placed a stake at each point and extended the line-intercept perpendicular to the direction of the harvester and placed another stake at the other end. We then lowered the line to the ground, carefully removed straw, and counted all seeds directly below every other red interval (i.e., 15 sampled bands within the 6-m line). Any seeds that captured in the straw went unaccounted. However, Halstead et al. (2011) indicated that seeds remaining in the straw did not significantly influence seed estimates.

We also collected five soil cores at each of the three random locations in every field (i.e., 15 soil samples per field). Traditionally, soil core samples obtained in the United States have been 10 cm deep by 10 cm in diameter (Manley et al. 2004; Stafford et al. 2006a; Marty et al. 2020). To reduce sample processing time and storage space, we used 6-cm-diameter and 5-cm-deep soil cores. Shallow cores successfully collect most available seed, because 70–90% of seeds exist in the top 5 cm of soil (Pernollet et al. 2017), and they may be more representative of actual food availability for waterfowl based on bill lengths of most ducks and geese (Baldassarre 2014).

We sampled dry fields using two methods: modified line-transects and soil cores. The differences in sampling methods between soil cores and transects result in two different estimates. Transects only sample the surface of the soil and, therefore, will capture only waste grain from the current season. In contrast, soil cores potentially capture waste grain from both the current year's harvest and the previous harvest (if not decomposed). For this study, we were only interested in estimates of rice and corn seeds. Therefore, we ultimately discarded any moist-soil seed that the soil cores may have captured during the processing of soil. In addition to comparing two methods, we compared estimates of grain among pre and postflooding of fields. We sampled a subsample of rice fields before and after postharvest flooding. Sampling rice fields in this manner was critical because only 50–70% of the rice acreage is postharvest flooded in most years (CVJV 2006, 2020; Miller et al. 2010). The time between harvest, postharvest treatment, and postharvest flooding was not consistent between farmers, as we had no control over when farmers flooded their fields. However, we restricted sampling to within 1 wk of flooding to reduce impacts of grain decomposition or foraging by waterfowl and to maintain comparability between environmental conditions. We sampled flooded fields using only soil cores and selected these based on the timing of flood and our ability to access them within the restricted time frame. We generated three additional random points for each of the fields and collected five soil cores at each of the random locations in every field. Sampling protocols for soil cores were the same in both dry and flooded fields.

At each site, we collected, bagged, and labeled all samples individually. We immediately transported samples to a lab at University of California Davis, and froze them. Processing soil cores began by thawing and washing each sample through a series of graduated sieves, with a U.S. Standard No. 10 mesh (2.00-mm) sieve as the bottom of the series. We collected seeds from the sieves and dried them to constant mass at approximately 80°C for 48 h. After drying, we removed chaff from rice seeds and measured dry mass. The presence of sprouted seeds was infrequent but occurred with both grain types. We did not treat sprouted seeds differently in the drying process and, as a result, we often removed sprouts with the chaff before measuring dry mass. We washed corn seeds using the same process and then dried and weighed them. We also froze, dried, and weighed seeds collected from the transect as described above. Samples from each field consisted of a total of 15 soil cores and 45 transect segments (Data S1, Supplemental Material). However, samples were not independent within each of the three random locations in each field (i.e., we collected 5 soil cores and 15 transect segments from each of the three locations). Accordingly, we averaged the transect sections and the cores samples, respectively, at each of the three sample locations, yielding three independent random values for transects and for cores for each field. To provide field-wide estimates, we calculated means of the three estimates (cores and transects, separately; Data S2, Supplemental Material).

The focus of our study was to compare estimates of the abundance of waste grain among fields subjected to different postharvest treatments, using two sampling methods (transects and cores). Accordingly, we did not adopt a sampling procedure such as multistage sampling, which has been shown to provide accurate estimates of waste grain abundance (Stafford et al 2006a, 2006b; Marty et al. 2020). Multistage sampling provides unbiased estimates by clustering samples within fields, and fields within growers or regions. The proper implementation of such a sampling design requires estimates of weighted probabilities of selection (e.g., probability of selecting a rice farm, probability of selecting a field within a rice farm, and the probability of selecting a sample location within a field). We did not design our study to apply such a sampling protocol and we did not have estimates for some of the weighted probabilities required, given that we sampled fields opportunistically as they became available. Hence, we caution that our estimates of average seed grain densities may be biased. However, the focus of our study was not to provide overall estimates of waste grain, but rather to contrast the abundance of grain among postharvest treatments; since we used the same sample selection protocol for all treatments, these contrasts should be subject to similar bias and therefore comparable.

To further account for potential nonrandom clustering of samples, we used a nested mixed model design for all analyses. We did not have control of which fields we would sample, and so we treated field sites as random effects within each treatment type. Sample locations (n = 3) were nested within each field and again treated as random effects because of possible covariance among samples in the same field. We included main fixed effects of postharvest treatment, sample type (transect or core), and year (2016 or 2017) and we included an interaction term for treatment by sample type to examine for the possibility that core or transect samples might vary depending on postharvest treatment.

To test for differences in waste grain abundance among postharvest treatments in both rice and corn fields, we employed standard least squares analysis in mixed nested models using the restricted maximum likelihood method. Restricted maximum likelihood is better suited to unbalanced designs in mixed models and provides unbiased estimates, tests, and confidence intervals compared with EMS (Expected Means Squares method) to parametrize variance components. We included main fixed effects of postharvest treatment (bale, chop, burn, chisel, disc, disc and roll, and no treatment), sample type (core, transect), treatment by sample type interaction, and year (2016, 2017). Random effects included the field identity (field site) and the sample location in each field (sample location; n = 3) nested within each field site. We used post hoc tests (least squares Student's t-tests, α = 0.05) to compare pairwise differences among treatments when a treatment effect was detected and to compare between sampling types (core, transect),

We also used standard least squares analysis in mixed nested models using the restricted maximum likelihood method to test for differences in waste rice grain abundance prior to and after flooding. We included main fixed effects of postharvest treatment (chop, chisel, disc, disc and roll, and no treatment), year (2016, 2017), and flooding (preflood, postflood), and we included an interaction term for treatment by flooding to examine for the possibility that the abundance of waste rice might differ between pre- and postflooding, depending on postharvest treatment. We used post hoc tests (least squares Student's t-tests, α = 0.05) to compare pairwise differences between preflooded and postflooded estimates of waste rice grain. We conducted all analyses with JMP^® Pro, Version 16.2 (SAS Institute Inc., Cary, North Carolina, 1989–2021).

For all analyses, we calculated dry mass estimates of the abundance of waste grain as described above and we compared untransformed and log-transformed (ln + 1) data to improve normality prior to analysis. Data were fit better using ln transformation in all analyses (all delta AIC >> 100, Akaike weights = 1.0 for ln-normal vs. normal distributions fit to data). Given a large number of zeros in the data (samples containing no grain), the ln transformation did not fully normalize the distribution and, in some cases, resulted in differences in pairwise comparisons among treatments when analyzing ln-transformed vs. untransformed data. However, overall main effects were consistent and we present all statistical analyses here using ln-transformed data (analyses based on raw data are available upon request). Means and confidence intervals for each treatment were back-transformed in Tables 2 and 3 (exp[ln values] − 1).

The back-transformed values are geometric means and, unlike arithmetic means, account for skew in data when there are several very high values and many small values. Our data were clearly better fit with an ln transformation illustrating such skew. Accordingly, the geometric mean provides the best estimates of average abundance of waste grain, as reported in Tables 2 and 3. However, most previous studies have reported arithmetic means, and to allow direct comparison with previous studies we present those in Figures 2–4. We note that arithmetic means will often be larger than geometric means because of the strong influence of extreme high values. Wildlife managers may need to reconsider which is most appropriate for use in estimating food availability and carrying capacity for waterfowl. To provide a direct comparison, we present both untransformed (arithmetic) and back-transformed ln values (geometric means) in Tables S1 and S2 (Supplemental Material) and, where appropriate, we report both in the text.

We sampled 84 rice fields and 47 corn fields in falls of 2016 and 2017. The number of rice fields sampled in each of the postharvest treatments was as follows: bale (16), burn (7), chisel (18), chop (12), disc (13), disc and roll (9), no treatment (6), and stomp (3). Of these 84 rice fields, we resampled 24 after postharvest flooding. Our ability to conduct sampling within the 1-wk window from the flood date on each field restricted the number of fields that we resampled after flooding. The number of corn fields sampled in each of the postharvest treatments was as follows: bale (9), chop (16), disc (4), disc and roll (6), and no treatment (12). We sampled fields distributed across the major rice and corn growing regions (Figure 1).

We detected significant differences in abundances of waste grain among postharvest treatments in dry rice fields (P < 0.0001; Table 2A; Figure 2). There was significant variation in grain estimates among field sites (25.94% of total, P = 0.0005) and sample locations within fields (29.19% of total, P < 0.0001; Table 2A). Differences in estimates of waste rice from core sampling vs. transect were marginal (P = 0.040), although there was a significant interaction of sample type and treatment (P < 0.0001; Table 2A). There were no statistically detectable differences in estimates of waste grain among years (Table 2A).

Post hoc tests indicated that no treatment provided the greatest amount of waste rice, followed by bale (Table 3A). Differences among chop, chisel, burn, and disc and roll were lower and statistically indistinguishable (Table 3A). Disced fields provided the least amount of waste rice. The significant interaction of sample type with postharvest treatment was largely driven by very high estimates of waste rice obtained in transect samples with no treatment in 2017. Rice harvest yields can vary dramatically by field, and to protect against such a possible bias, we also calculated estimates of the abundance of waste rice as a proportion of the total yield in each field (rather than as an absolute value). Doing so did not appear to influence patterns of waste grain abundance among different postharvest treatments in rice fields. Furthermore, variation in harvest yields among fields was not significantly correlated with estimates of the amount of waste rice (R² = 0.021, P = 0.72)

We sampled a subset of rice fields before and after postharvest flooding. The time between harvest, postharvest treatment, and postharvest flooding was not consistent between farmers, as we had no control over when farmers flooded their fields. However, we restricted sampling to within 1 wk of flood initiation to reduce impacts of rice decomposition or foraging by waterfowl and to keep environmental conditions comparable. We compared the abundance of waste rice estimates in these fields obtained prior to flooding to those obtained after flooding (Figure 3). We conducted this analysis using soil core data only, as we could not obtain transect samples in flooded fields. In this analysis, there was no overall significant effect of postharvest treatment (P = 0.147; Table 2B), but there was a very strong effect of flooding (P < 0.0001; Table 2B). There was no interaction of treatment with flooding but there was a significant year effect with waste rice estimates postflooding considerably lower in 2016 than 2017 (Figure 3). There was also significant variation among field sites (20.1% of total variation; P = 0.018) but not among sample locations within field sites (5.6% of total; Table 2B).

Estimates of the abundance of waste rice grain before flooding (preflood) were significantly greater than those obtained in the same fields after flooding (postflood; Table 3C; Figure 3). The combined waste rice estimates for flooded fields, averaged across all postharvest treatments, was 297.7 kg/ha preflood (arithmetic mean based on untransformed raw data; geometric mean = 202.5 kg/ha; Table 3C and Table S1, Supplemental Material). Estimates of waste rice postflood averaged 169.1 kg/ha arithmetic mean; geometric mean = 97.7 kg/ha; Table 3C and Table S1, Supplemental Material).

Postharvest treatment of corn fields had significant impacts on the abundance of waste grain estimates (P = 0.0013; Table 2C). As was the case for rice fields, there was significant variation among field sites (18.6% of total variation, P = 0.0095; Table 2C), and also among sample types (P = 0.0009) although there was no interaction of sample type (core or transect) and treatment (Table 2C) and no differences among years or among sampling locations with fields (Table 2C). Post hoc tests indicated fields that were not incorporated postharvest (i.e., bale, no treatment, chop) contained significantly more waste corn than those fields incorporated postharvest (i.e., disc and roll, and disc only; Table 3D; Figure 4). Corn harvest yields vary dramatically by field, and to protect against such a possible bias, we also calculated waste corn estimates as a proportion of the total yield in each field (rather than as an absolute value). Similar to the results found in rice fields, patterns of waste grain abundance were not correlated to harvest yields in corn fields (R² = 0.008, P = 0.35). Analyses based on ln + 1 transformed data indicated that transect samples yielded higher estimates of waste corn than core samples (Table 3E), but analyses based on untransformed raw data indicate no difference and estimates from core samples tended to be higher (Figure 4). We attribute these differences to the effects of ln transformation on distributions with several very high extreme values and many small values as was the case with our data (see “Methods”).

We detected significant impacts of postharvest treatments on estimates of the abundance of waste grain in both rice and corn fields. However, sampling of rice fields during fall 2016 and fall 2017 provided some unexpected results. Most notably, the effects of postharvest treatments on the abundance of waste grain were not as great as anticipated. While we did observe results consistent with our initial hypothesis—that is, that more mechanical manipulation in fields would result in reduced amounts of waste grain—the reduction in incorporated fields was not as large as anticipated. We found that estimates of waste rice abundance from incorporated fields were significantly lower than the no-treatment and bale categories, while the other treatments all shared considerable overlap. Incorporation includes the following postharvest treatments: disc, disc and roll, and chisel. The average estimate (based on untransformed data) of waste rice in bale and no-treatment categories was 447 kg/ha, whereas the average estimate across the remaining treatments (burn, chop, chisel, disc, and disc and roll) was 276 kg/ha. A similar study in the Mississippi Alluvial Valley found that fields with standing stubble contained up to twice as much waste rice than any other postharvest treatment, consistent with our findings (Havens et al. 2009).

The difference observed between no treatment and chopping was unexpected as these two treatments should result in similar waste rice estimates. These were the only postharvest treatments that do not remove or incorporate any of the straw. One possible explanation for the lower estimates in the chop treatment could be due to the difference between how much time elapsed between harvest and sampling. We sampled no-treatment fields immediately after harvest as producers implement no other postharvest treatments. In contrast, we sampled chopped fields only after chopping occurred. Depending on the farm, chopping occurred as early as a few days after harvest, but in many cases did not occur for up to a month after harvest. This delay would allow ample time for seed depletions by various wildlife species.

Miller et al. (1989) reported that dry incorporation greatly reduced food abundance in rice fields; however, they based their analysis on only four plowed fields and they noted that they would need a larger sample size to provide definitive results (Miller et al. 1989). During the 2 y of their study, average waste rice estimates (arithmetic means) were 388 kg/ha in harvested fields, 276 kg/ha in burned fields, and an overall average of 332 kg/ha across both postharvest treatments (Miller et al. 1989). Kross et al. (2008) found that disced fields in the Mississippi Alluvial Valley yielded the lowest amounts of available waste rice in late autumn; however, they found no significant effect of postharvest treatment on waste rice abundance immediately after harvest. A second study in the Mississippi Alluvial Valley (Manley et al. 2004) confirmed that result. The researchers estimated available waste rice in fields with standing stubble and incorporated straw (e.g., disced) and, while they observed a trend of less available waste rice in incorporated fields, there was no significant difference in available waste rice between the two postharvest treatments (Manley et al. 2004). Ultimately, across all of these studies there is a trend toward lower availability or abundance, or both, of waste rice in incorporated fields, but the variance among fields appears to be too large to detect significant statistical differences. We found similar patterns, but unlike the previous studies, we detected significant differences among some postharvest treatments.

Estimates of the abundance of waste grain in flooded rice fields were much lower than the estimates in dry fields, given the same postharvest treatment. The average abundance of waste rice across 24 flooded fields was 169.1 kg/ha (arithmetic mean; 97.7 kg/ha geometric mean; Table 3C) compared with a value of 297.7 kg/ha (202.5 kg/ha geometric mean) in the same fields before flooding occurred. Two factors may be responsible for this large reduction. First, many farmers do not initiate postharvest flooding until they have harvested all of their fields. Ample opportunity exists for seed depredation by other species such as blackbirds, geese, and rodents during the time between harvest and postharvest flooding. Several studies have shown that the abundance of waste rice decreases rapidly after harvest (Manley et al. 2004; Stafford et al. 2006a; Kross et al. 2008). For example, Stafford et al. (2006a) found that available waste rice dropped by 71% from postharvest to late autumn, a reduction of 58 kg/ha biweekly. Second, wintering goose numbers have increased dramatically in California; in fact, population estimates have more than doubled in the past 15 y (CVJV 2020). While sampling, we regularly observed large number of geese concentrated on rice fields that were being flooded. It appeared that geese were targeting fields as they were being flooded, which could result in geese consuming large portions of the waste rice before fields became fully inundated. There has been some concern that geese may deplete available waste rice that would otherwise be available for dabbling ducks (Havens et al. 2009); furthermore, current estimates of goose and swan numbers have reached a point where their nutritional needs have surpassed the food supply and they would be capable of fully depleting all agricultural food sources in the Central Valley each winter (CVJV 2020).

The CVJV assumes that ducks only forage in harvested rice fields that are winter-flooded. However, the food biomass value that the CVJV assigns to winter-flooded rice fields has been based on sampling in the preflood period (CVJV 2006, 2020). Our results indicate that the foraging value of winter-flooded rice fields may be significantly reduced by the time postharvest flooding occurs. As a result, the CVJV may be overestimating the contribution that winter-flooded rice fields currently contribute to food energy needs of dabbling ducks in the Central Valley. It has been the assumption that winter-flooded rice fields provide nearly half of all the food available to dabbling ducks (CVJV 2006, 2020; Petrie et al. 2016); however, our results suggest that this may be an overestimation of food availability.

Estimates in corn fields followed the same predicted trend as rice fields, based on our original hypothesis; specifically, fields treated with less mechanical manipulation postharvest (i.e., no treatment, chop, and bale) have significantly greater abundance of waste corn estimates than fields incorporated postharvest (i.e., disc and disc-and-roll treatments). Several studies conducted outside of California have reported similar results for postharvest corn fields. For example, idle fields (i.e., equivalent to our no treatment) had the greatest available waste corn, followed by mulched (i.e., chopped), while tilled (i.e., disced) fields provided the lowest amount of available waste corn (Baldassarre et al. 1983; Warner et al. 1985; Sherfy et al. 2011). Baldassarre et al. (1983) found that discing reduced available waste corn by 77% and deep plowing reduced waste corn by 97% compared to fields with standing stubble. In our study, fields that were not incorporated postharvest contained 256 kg/ha, on average (based on untransformed data), whereas fields that were incorporated postharvest contained only 58 kg/ha, a 77% reduction. The overall average, across all postharvest treatment types, was 184 kg/ha. The CVJV was using a much higher estimate, 519 kg/ha, in the 2006 implementation plan. Reducing this number will better reflect the amount of food available from corn in California.

We obtained harvest yields from all of the growers to determine the relationship between yields and estimates of the abundance of waste grain in both rice and corn fields. Our original hypothesis was that fields with higher yields would result in significantly higher abundances of waste grain. We based this on the expectation that a standard proportion of the yield would be lost during harvest. If this were true, then increased yield would result in increased abundance of waste grain. An alternative hypothesis was that fields with higher yields would result in significantly lower abundance of waste grain due to harvester efficiency. In this scenario the higher yields would be directly related to more efficient harvest. Surprisingly, our results supported neither of these predictions. Rice farmers reported a wide range of yields, from 6,725 kg/ha to 12,330 kg/ha. Corn farmers reported and even larger range of harvest yields, from 3,027 kg/ha to 16,252 kg/ha. Never the less, higher harvest yields did not result in significantly higher or lower abundances of waste grain in fields. This result leads us to believe that harvester inefficiencies were static and not impacted by fluctuations in crop yields.

Although core sampling has been the standard approach to estimate the abundance of waste grain in agricultural fields, a new method has recently been employed using line-transect methods (Halstead et al. 2011). However, we are unaware of any attempt to compare the efficiency or accuracy of these two approaches, particularly in the same fields and sample locations. We found that transect and soil core sampling methods provided similar estimates of the abundance of waste grain across all of the postharvest treatments in both crop types, with one exception: the no-treatment category in rice fields in 2017 (Figure 2). It is possible to attribute the large difference observed in no-treatment rice fields to how residual rice straw affected each of the sampling methods. Cores did not penetrate through large amounts of rice straw. Therefore, when collecting cores amid dense straw we attempted to shake free any seeds in the rice straw, and gently removed the straw before taking the soil core. In contrast, with transects, we carefully picked through the rice straw and collected any seeds that were directly below the transect, including those that were both on the soil and suspended in the straw. It was possible that soil cores were biased low in the no-treatment category if they do not accurately account for seeds that remain trapped within the rice straw and removed before the sample was obtained. However, transects were likely biased high as that sampling method relies on an ocular estimate to determine if the seeds are either “within” or “outside of” the transect segment. Additionally, in fields containing more complex structure (e.g., standing stubble), we could not lower the transect to the surface of the soil, which in turn adds more potential observer bias when determining if a seed was either “within” or “outside of” the transect segment. The large difference in the no-treatment fields in 2017 likely drove the significant difference found among sample types when years and fields were included in the overall analysis after ln transformation (Table 3B)

Core samples appeared to have higher averages (untransformed) than transect samples in dry corn fields in 2016 (no treatment and chop; Figure 4) but not in 2017. After ln transformation, pairwise comparisons indicated that transect samples yielded higher estimates of waste corn than core samples (Table 3E). Overall, when comparing the two different sampling methods, postharvest treatment appeared to have more influence in rice fields (significant treatment by sampling type interaction; Table 2A) but not in corn fields (no interaction, Table 2C). Given the greater variability in transect samples among treatment, soil cores seem to provide reliable estimates, and therefore might be the preferred method of sampling.

Our study supports the hypothesis that incorporation reduces the amount of waste grain available for waterfowl, although the differences were not as dramatic as anticipated. We observed this trend in both rice and corn fields, so we recommend that growers or managers interested in providing food for waterfowl use postharvest practices such as no treatment, baling, or chopping. Ducks primarily forage in flooded fields, so they may underutilize food resources unless the fields have been flooded (Sesser et al. 2016). Therefore, we strongly encourage flooding of rice fields whenever possible to encourage use by waterfowl. We also found very rapid depletion of waste rice in flooded fields despite the fact that we sampled all fields within 1 wk of flooding. Geese may impact food abundance for ducks as wintering goose population continue to climb and competition increases over the limited agricultural food resources (CVJV 2020). We recommend managers consider decline in waste rice when developing estimates of carrying capacity and planning for nonbreeding Joint Ventures. We also found an impact from postharvest treatments on field sampling methods (i.e., transects vs. soil cores). When sampling the abundance of waste grain in the future, it will be important to note that postharvest treatment can more readily impact transect sampling than traditional soil-core sampling. Finally, we recommend that the CVJV use more conservative estimates for waste rice abundances when developing management plans for the Central Valley. Although this results in a lower estimate of the abundance of waste rice overall, we believe that it is the best representation of actual food abundance for ducks.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Data S1. Full raw data set from the waste grain sampling of rice and corn fields in the Central Valley, California, during 2016 and 2017. Definitions for column headers: year = year the sample was taken; graintype = rice or corn; designation = sampling method (core, flooded core, transect); FIDcounty = county of sample (COL = Colusa; GLE = Glenn; YOL = Yolo; BUT = Butte; SUT = Sutter; YUB = Yuba; SAN = San Joaquin); FIDfldnum = the field number (1–136); FIDlocation = one of the three random sample locations per field; FIDsmplnum = sample number at FIDlocation. Transect = 0; core = 1–5; flooded core = 6–10; treatment = postharvest treatment; wetcount = number of seeds before drying; drycount = number of seeds after drying; actualcount = number of seeds weighed; initialweight = dry weight (g) with chaff and dirt; chaffweight = weight (g) of chaff and dirt; smplweight = initialweight − chaffweight; avseedweight = smplweight/actualcount; actualsmplweight = dry weight (g) of seeds only; areasmpl = area sampled: transect = 0.017625 m², core = 0.00273 m²; lbsare conv = conversion factor from (g/m²) to (lb/acre); lbacre = (actualweight/areasmpl) × (lb/acreconv); kghaconv = conversion factor (g/m² to kg/ha); kgha = (actualweight/areasmpl) × kghaconv.

Available: https://doi.org/10.3996/JFWM-21-061.S1 (303 KB XLSX)

Data S2. Analysis data set from the waste grain sampling of rice and corn fields in the Central Valley, California, during 2016 and 2017 based on means from the lbsare and kgha columns in Data S1, Supplemental Material. For analysis we took averages of the transect sections and the cores samples, respectively, at each of the three sample locations, yielding three independent random values for transects and cores for each field. To provide field-wide estimates, we calculated means of the three estimates (cores and transects) separately. Definitions for column headers: year = year sample was taken; grain type = corn vs. rice; treatment = postharvest treatment used in field; sample type = sampling method (core, transect); flood = blank, only sampled dry, preflooded = the dry sample before flooding occurred, postflood = the second sampling event in the same field after flooding occurred; field number = the field number (1–136); field location = one of the three random sample locations per field; N rows = the number of samples within the mean. Core = 5 samples; transect = 1 sample; mean(lbsacre) = mean of each FIDlocation from the full data set in lbs/acre; log(mean[lbsare] + 1) = log of the means plus 1; mean(kgha) = mean of each FIDlocation from the full data set in kg/ha; log(mean[kgha] + 1) = log of the means plus 1.

Available: https://doi.org/10.3996/JFWM-21-061.S2 (70 KB XLSX)

Table S1. Arithmetic means and 95% confidence intervals (untransformed values) and geometric means and 95% confidence intervals (back-transformed means of ln + 1 values) of abundance of waste rice in rice fields in California in 2016 and 2017. Least square means are shown for each treatment, sample type (transect or core), and year in dry rice fields. Least square means are shown for preflood and postflood fields in fields subjected to postharvest flooding. Geometric means are lower than simple arithmetic means due to ln-transformation reducing the influence of extreme (high) values.

Available: https://doi.org/10.3996/JFWM-21-061.S3 (27 KB DOCX)

Table S2. Arithmetic means and 95% confidence intervals (untransformed values) and geometric means and 95% confidence intervals (back-transformed means of ln + 1 values) of abundance of waste corn in corn fields in California in 2016 and 2017. Least square means are shown for each treatment, sample type (transect or core), and year in dry corn fields. Geometric means are lower than simple arithmetic means due to ln-transformation reducing the influence of extreme (high) values.

Available: https://doi.org/10.3996/JFWM-21-061.S3 (27 KB DOCX)

Reference S1.[CVJV] Central Valley Joint Venture. 2006. Central Valley Joint Venture 2006 implementation plan: conserving bird habitat. Sacramento California: U.S. Fish and Wildlife Service.

Available: https://doi.org/10.3996/JFWM-21-061.S4 (16.540 MB PDF)

Reference S2.[CVJV] Central Valley Joint Venture. 2020. Central Valley Joint Venture 2020 implementation plan: Sacramento, California: U.S. Fish and Wildlife Service.

Available: https://doi.org/10.3996/JFWM-21-061.S5 (22.106 MB PDF)

We are grateful to the organizations that provided funding and made this research possible, specifically, California Department of Fish and Wildlife (California Duck Stamp Funds) and the Central Valley Joint Venture for providing funding for this project, and Ducks Unlimited for initiating the study and coordinating funding. This work was also supported by the Dennis G. Raveling Waterfowl Endowment at University of California Davis and project no. CA-D-WFB-6342-H/Accession No. 1013603 from the U.S. Department of Agriculture National Institute of Food and Agriculture (J.M.E.). We thank the anonymous reviewers and the Associate Editor of the journal for the thoughtful review of our manuscript. Additionally, we thank the following individuals for their involvement in this study. V. Getz and G.S. Yarris were instrumental in getting the project started and securing funding. B.A. Linquist played a key role in the development of this project. D.J. Smith assisted with construction of sampling tools, program troubleshooting, and statistical analysis. M. Cazzasa and J. Kohl provided sampling equipment and advice on sampling methods. Technicians P. Maleko, E. Gokce, D. Coye, C. Gorman, P. McKinney, and C. Bryant provided excellent assistance, both in the field and the lab. Finally, we thank all the growers who allowed us to work on their property. Without their generosity and willingness to participate this project would not have been possible.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.