Final Report for GW11-012
Project Information
Herbicide resistant weeds are an ever-worsening problem in California rice systems. By integrating cultural and chemical control of weeds, alternative stand establishment approaches have been shown to reduce herbicide resistant weed populations while simultaneously reducing herbicide inputs in direct seeded rice. However, management decisions are more complicated and time-sensitive when growers use alternative establishment methods for weed control.
The overarching goal of the proposed research was to develop a geographically and temporally sensitive decision support tool that predicts the minimum time required to achieve control of watergrass (Echinochloa spp.) and smallflower umbrella sedge (Cyperus difformis) populations using alternative establishment management methods in direct-seeded rice. The project was successful in developing and validating regional-scale emergence models for both weed species, quantifying the spatial and temporal variability in emergence, and making site-specific, real-time and historical emergence data publicly available via a web-based interface.
Coupled with effective outreach, this tool will facilitate the planning and execution of weed control via cultural methods and enable wider adoption of alternative establishment approaches. It will also serve as a platform for further efforts to translate management-related modeling outputs to rice growers in the California region.
Introduction
California rice has among the highest number of herbicide resistant weeds in the United States (Heap, 2012) due, in part, to a reliance on herbicides for weed control and soils that offer limited options in terms of crop rotation (Hill et al., 1994; Pittlekow et al., 2012). As a response to the growing problem of herbicide resistant weeds (Fischer et al., 2000; Osuna et al., 2002), the use of stale seedbeds and drill-seeded systems with intermittent early-season flooding has been investigated in recent years (Pittlekow et al., 2012). These establishment systems attempt to diversify the weed recruitment environment and improve the timing and efficacy of herbicide applications, thereby reducing the overall usage of herbicides.
In both systems, ground is broken in the early spring, as soon as the fields are sufficiently dry from the rainy season to permit tillage. Tillage generally includes passes with a chisel plow, disc, tri-plane and roller. In the stale seedbed system, the tilled, rolled field then experiences two to four “flushes” of irrigation over the course of several weeks. The flushes completely saturate the seedbed and last for one to two days each, depending on the size of the field, the soil type and the design of the irrigation delivery system. Water is managed similarly in drill seeded fields; however, rice will have already been planted prior to the flushes. In either case, the management objective is a moist, primarily aerobic seedbed that rapidly recruits problematic weeds such as watergrass (Echinochloa phyllopogon and E. oryzoides) and smallflower umbrella sedge (Cyperus difformis), which are resistant to the greatest variety of herbicide formulations and cause the most economic damage in the California rice system (Fischer et al., 2000; Osuna et al., 2002).
In the stale seedbed system, once the weed population has reached a critical growth threshold, a broad-spectrum herbicide (such as glyphosate) is then applied to the field. The field is subsequently flooded, and rice is planted aerially without further seedbed disturbance. The drill seeded system primarily recruits watergrass, which is controlled via propanil and/or pendimethalin applied to the unflooded seedbed after a critical growth threshold is reached and prior to the permanent flood (CRPW, 2011). Both systems have shown promising results as alternatives to the conventional establishment systems by controlling weeds while reducing herbicide use and maintaining yields (Pittlekow et al., 2012).
Despite the promise of these systems, their efficacy is sensitive to the timing of herbicide applications. Also, in the case of stale seedbeds, planting of rice is delayed until at least as long as is required for the weeds to reach their critical growth threshold and be eliminated. This affects variety choice and introduces late season risk of blanking (Board et al., 1980). As a result, the timing of weed emergence and early growth in alternative stand establishment systems is relevant and useful information for farmers to plan for and implement effective weed management.
We undertook a two-year study to develop and validate regional temperature-based emergence models for smallflower umbrella sedge and watergrass in stale seedbeds and drill-seeded rice systems. We then merged these models with a regional temperature dataset available at fine-scale temporal (daily) and spatial (4km2) resolutions to quantify the predicted emergence and its variability across the period of interest (April 15 to June 1). To communicate the resulting information to land managers in the region, we constructed a web-based interface that reports real-time and historical weed emergence predictions and also presents the models that underpin this decision support tool. We expect this website to be a platform from which these and other temperature-based plant growth models can be translated to site-specific management information, thereby improving the precision and efficacy of rice management in the region.
Objective 1: Produce an empirical model that predicts the emergence of watergrass (E. spp.) and smallflower umbrella sedge (C. difformis) as a function of air temperature in rice fields with early-season irrigation flushes.
Objective 2: Determine the spatial and temporal resolution at which air temperature can be predicted across the Sacramento Valley between April 15th and June 1st based on available historical data.
Objective 3: Validate the accuracy of the model's emergence predictions across heterogeneous spatial and temporal environments.
Objective 4: Implement a web-based tool that communicates spatially- and temporally-sensitive emergence predictions to growers.
Objective 5: Communicate management-related results of research to growers via UC extension networks, field days and UC websites and publications.
Objective 6: Publish a peer-reviewed scientific paper that reports on the accuracy and efficacy of a spatio-temporally sensitive weed emergence model in alternative establishment rice systems.
Cooperators
Research
In 2010 and 2011, emergence of smallflower umbrella sedge and watergrass was observed in three rice fields where their presence had been confirmed during the previous growing season. One of the fields (location 1: 39 33’51”N, -122 4’14”W) was managed as a stale seedbed, while the other two fields (location 2: 38 53’43”N, -121 ;43’43”W; and location 3: 39 00’35”N, -121 42’29”W) were drill-seeded fields. The soils in two of the three fields were classified as Mollisols and the other as a Vertisol. The sand, silt and clay contents ranged from 17-27%, 30-43%, and 30-53%, respectively.
Following pre-season tillage, four to eight main plots per field were established in portions of the fields likely to maximize the absolute number of observable weeds (based on their historical occurrence as reported by the growers). Main plots were 5x10m in size and were bisected in the 10m direction by an elevated 6m wooden plank that served as an observation platform to prevent soil surface disturbance within the plot. Air temperature was recorded at the center of the main plot at 15 minute intervals via shielded Hobo U23 Pro v2 External Temperature Data Logger (Onset) located 1.5m above the soil surface. Each main plot contained four subplots 0.3x0.3m in size.
Emergence observations were made daily in 2011 and every one to two days in 2010, beginning from the first day that irrigation was brought onto the field (the “start date”) until no further increases in emergence had occurred for at least three days. “Emergence” was counted as soon as the plants were visually identifiable by species. Generally, this was the first day of emergence for smallflower umbrella sedge (<0.5cm in shoot height) and one to two days after initial emergence (approximately 1 cm in shoot height) for watergrass. Daily subplot totals were recorded on an absolute basis and then calculated as a proportion of the overall total number of plants to emerge in the subplot during the entire observation period. Proportional emergence for the main plot was calculated as the mean of the four subplots.
In one field (location 1) observations were made in both 2010 and 2011, while in each of the other two fields observations were made for a single season (location 2 in 2010 and location 3 in 2011), resulting in a total of four site-year and 63 site-day observations per species. The distance between fields was 80km in 2010 and 69km in 2011.
A validation dataset was collected in 2010 at a site located 33-62km from the other fields (39 27’3”N, -121 43’9”W). Observations were made as described above but with less frequency (five to seven days) in two basins for a total of five site-day observations for smallflower umbrella sedge and eight site-day observations for watergrass.
Using the ‘nlme’ package (Pinheiro et al., 2011) in R (2010) 2.11.1, a non-linear, mixed-effects model was fit to proportional emergence and base-temperature modified air temperature using the sigmoidal function (Equation 1):
E= 1/(1+ e^(- (TU – (TU50%))/Erate ) )
where E = emergence; TU = cumulative thermal units; TU50% = cumulative thermal units at 50% emergence; and Erate = rate of emergence.
Thermal units (TU) were determined by subtracting a base-temperature (BT) from the air temperature measurements (T) and summing the results cumulatively over time (t, days) (Equation 2):
TU=sum(i,n): (max[(Ti-BT),0]?ti)
Base temperatures were initialized at 17.5C for smallflower umbrella sedge (Pedroso, 2012) and 9.3;C for watergrass (Boddy, 2012). The resulting TU accumulation was matched to plot-site-day specific proportional emergence. The fixed parameters TU50% and Erate were fit iteratively via maximum likelihood and an autoregressive correlation structure to account for the repeated nature of the measurements. The effects on TU50% of site-year were designated as random.
Subsequently, base temperatures were incrementally modified 0.05C in both positive and negative directions. The resulting models were iteratively compared for significant differences using a 1 degree of freedom chi-squared distribution test on the log-likelihood difference. A range of base temperature values resulting in models that were not significantly different (P<0.025) from the best fit was determined. The midpoint of this range was designated as the base temperature and used to fit a final model for each weed species. Independently distributed errors and normality were determined graphically.
Using spatially and temporally specific maximum and minimum daily air temperature interpolations accurate to 4km2 (Hart et al., 2009; COMET, 2012), thermal units were calculated according to the double triangle method (Sevacherian et al., 1977; Roltsch et al., 1999) for the dates and location where the validation dataset was collected. The parameters from the aforementioned models were combined with the spatiotemporally specific thermal units to produce emergence predictions for comparison with the validation dataset.
Subsequently, the root mean square error (RMSE) and modeling efficiency index (EF) of the predicted versus observed values were calculated as in Loague and Green, (1991) such that a perfect model fit would result in a RMSE=0 and an EF=1.
Using the species-specific parameter results from the aforementioned thermal unit models and the double triangle thermal unit calculation method (Sevacherian et al., 1977; Roltsch et al., 1999), the number of days to emergence was predicted for each location on a 4km2 grid in the Sacramento Valley rice growing region where COMET (2012) air temperatures were available (n= 193), for all years from 2003 to 2011 (n=9), using all dates from 4/15 to 6/1 as start dates (n=48). This totaled 83,376 spatiotemporally specific emergence predictions per species.
Spatiotemporal variability was characterized using the ‘lm’ function of the base package and the ‘lme’ function of the ‘nlme’ package (Pinheiro et al., 2011) in R (2010) 2.11.1. First, a simple linear model was fit to the species-specific simulation results by apportioning the variance of the spatiotemporally-specific days-to-emergence responses according to the grouping factors “start date,” “year” and “location” to account for the intra-annual, inter-annual, and spatial variability, respectively. After confirming the significance (P<0.001) of the grouping factors via one-way ANOVA, for each start date from 4/15 to 6/1, a linear, mixed-effects model was fit to the data such that the variance in days-to-emergence was explained by the fixed effect of “location” and the random effect of “year.” The effect of “location” incorporated a spherical correlation structure with a lag distance (distance at which two randomly chosen locations within the population are significantly different after taking spatial autocorrelation in to account) initialized at 50km based on variogram plots of the mean spatial variability across all start dates and years. In addition to the models fit to each start date during the period of interest, for each location, a mean days-to-emergence for all start dates from 4/15 to 6/1 was calculated. These data were analyzed with the same mixed effect model as the individual start dates (fixed effect = spatially autocorrelated location; random effect = year) to determine the overall proportional variance attributable to location during the period of rice establishment. Independently distributed errors and normality were determined graphically.
With the same models and data source used in the statistical analysis, a webtool was created using JavaScript that translates the models to percent emergence (real-time) or days to emergence (historical) on a web-based interface.
Under the saturated but predominantly aerobic soil conditions which emergence observations were recorded, a base temperature range of 15.10-16.00C provided the best fit to the sigmoidal (Equation 1) for smallflower umbrella sedge (Table 1). Meanwhile, a range of 8.50-9.25C resulted in the best fit for watergrass (Table 1). In addition to the higher base temperature, smallflower umbrella sedge required fewer cumulative thermal units to achieve 50% emergence (TU50%=28.02 ; 3.85) than watergrass (TU50%=106.47 ; 6.91) and exhibited a more rapid rate of emergence as well (Table 1; Figure 1). Additionally, the proportional variability of the year-specific site effect was greater for smallflower umbrella sedge than for watergrass (standard deviation of TU50% = 7.20/28.02 and 8.16/106.47, respectively) (Table 1). Finally, the modeled emergence for both species largely agreed with the observed emergence in the independently collected validation dataset, resulting in RMSE and EF values of 0.21 and 0.80 and 0.14 and 0.88 for smallflower umbrella sedge and watergrass, respectively (Table 2).
For both smallflower umbrella sedge and watergrass, the simulated days to emergence were significantly affected by intra-annual, inter-annual and spatial variability according to a simple linear model of the effects (P<0.001). Holding intra-annual variability constant, modeling inter-annual variability as a random effect, and accounting for spatial autocorrelation also resulted in a significant effect of location on the number of days to emergence (P<0.001). Across all starting dates from 4/15 to 6/1, the mean days to emergence among locations ranged from 19.6 to 8.8 days and 31.9 to 20.4 days for smallflower umbrella sedge and watergrass, respectively (Figure 2, Table 3). The standard error for the effect of location ranged from 2.4 to 1.0 days and 1.8 to 0.6 days, and the standard deviation for the random effect of year ranged from 6.8 to 2.7 days and 5.1 to 1.6 days for smallflower umbrella sedge and watergrass, respectively (Figure 2, Table 3).
Temporal effects explained the greatest proportion of the variability (Figure 2, Table 3). Nonetheless, the effect of location on the number of days to emergence explained approximately 11% of the modeled variability in smallflower umbrella sedge emergence (F-value = 14.23(location)/129.23(intercept)) and 4.5% of the modeled variability in watergrass emergence (F-value = 21.01(location)/468.8(intercept)) across the entire period of rice establishment. According to the mixed model results, the spatial auto-correlation among locations varied intra-annually with a lag distance averaging 60.8 km and ranging from 45.0 to 161.4 km for smallflower umbrella sedge and averaging 45.0 km and ranging from 24.1 to 79.1 km for watergrass across the 125x70km region of interest (Table 3). The spatial auto-correlation and lag distances were further confirmed via Moran’s I (P<0.001), maps of the spatial variability (Figure 3), and variogram plots with a lag distance of 54.5 and 50.8 km for smallflower umbrella sedge and watergrass, respectively (not shown), when location means were taken across all start dates and years.
A wide range of base temperatures have been reported for various early growth stages of Echinochloa ssp. (Wiese and Binning, 1987; Steinmaus et al., 2000; Swanton et al., 2000; Gardarin et al., 2009; Masin et al., 2010). However, only recently have base temperature estimates been made for smallflower umbrella sedge and watergrass biotypes that are specific to the California rice growing region (Pedroso, 2012; Boddy, 2012). Using these estimates to initiate the model fitting procedure narrowed the maximum likelihood region (Lindstrom and Bates, 1990) and resulted in base temperature ranges for the field observations of both smallflower umbrella sedge and watergrass emergence (Table 1) that agreed with the base temperatures developed at the laboratory scale by Boddy (2012) and Pedroso (2012) when the range of errors across the studies are accounted for. Doing so also served to ground the field-based observations reported here in more physiologically detailed studies, improving confidence in the accuracy of the models.
Although soil temperatures are what seeds experience and are predominantly used to model weed emergence (Forcella et al., 2000), the use of physiologically-relevant (base temperature modified) air temperatures in the empirical models reported here was also effective in explaining large portions of the variability (Figure 1). This approach resulted in agreement with the independently collected validation data (Table 2) and further confirms the general utility of the thermal time approach to empirical weed emergence modeling (Bradford, 2002; Grundy, 2003; Leguizamon, 2005, Masin et al., 2010). While the results reported here clearly demonstrate that temperature is a determining environmental variable for weed emergence, osmotic potential and the availability of oxygen have also been shown to influence seed germination (Benvenuti and Macchia, 1995, Forcella et al., 2000; Bradford, 2002), including for smallflower umbrella sedge (Chauhan and Johnson, 2009) and Echinochloa ssp. (Boyd and van Acker, 2004) specifically. While osmotic potential and oxygen were not measured in the fields or accounted for in the models reported here, they were controlled for to an extent by selecting fields for observation where early-season water was managed similarly (periodic flushes to ensure moist but predominantly aerobic conditions).
Nevertheless, it is likely that portions of the unexplained variability in these emergence models are due to differences in osmotic conditions and/or oxygen availability between sites and years. Further, it should be noted that the empirical models reported here would not apply directly to rice systems with different water management approaches (eg. a continuous, early-season flood).
The relative differences in emergence parameters between the two weeds are indicative of their distinct competitive strategies and help to explain differences in the proportional variability between the two species. Smallflower umbrella sedge is a sedge with the ability to complete its vegetative and reproductive stages in roughly one third of the time required by rice (Chauhan and Johnson, 2009); watergrass is a grass whose phenology from emergence to heading closely mimics that of the rice plant (Yamasue, 2001). The rapid life-cycle and narrow temperature range for growth would tend to exacerbate temperature-related variability in smallflower umbrella sedge emergence. In contrast, the slower rate of watergrass development would provide a larger temporal interval over which to integrate temperature-related variability relative to smallflower umbrella sedge. Indeed, such differential effects of temperature-related variability between the two species are demonstrated by the greater proportional variability observed for the time to 50% emergence (Figure 1, Table 1) and the year-specific effect of site for smallflower umbrella sedge (Table 1). The simulation results, which held the error from the emergence models constant and only considered the variable effects of temperature over space and time, illustrate this difference in absolute terms (number of days) as well. Both the means and the maxima of the ranges of variability due to the effects of year and the location were greater for smallflower umbrella sedge than for watergrass (Figure 2, Table 3).
Additionally, the lag distance was more variable for the simulated smallflower umbrella sedge emergence across the period of rice establishment (Table 3).
Interspecific differences aside, the simulation results clearly demonstrate that spatiotemporally specific temperature inputs greatly improve the accuracy of weed emergence predictions when applied at the regional scale. Between the earliest and latest start dates considered, based solely on changing the intra-annual start date from 4/15 to 6/01, the nine-year average regional emergence predictions changed by as much as 10.8 and 11.5 days for smallflower umbrella sedge and watergrass, respectively (Figure 2, Table 3). In addition, during this same period of interest, the standard deviation for the inter-annual variability was as high as 6.8 and 5.1 days (Figure 2, Table 3). Although the spatial variability was small in proportion to the temporal variability, it was highly significant nonetheless (P<0.001), with the standard error for the effect of location as high as 2.4 and 1.8 days (Figure 2, Table 3). Further, the lag distance estimates indicated that, on average, emergence predictions were significantly improved by spatially explicit temperatures when locations were at least 61 and 45 km apart for smallflower umbrella sedge and watergrass, respectively (Table 3). In management terms, each effect modeled by the simulation (location, year and start date) could improve the accuracy of weed emergence predictions by days on its own and by weeks when combined. For weed management strategies that depend on precision timing for weed control, such improvements could greatly influence the utility of decision support tools and, by extension, the success or failure of a weed control strategy (see Figure 3).
As reported above, there is a strong quantitative case to be made for the utility of merging site-specific, real-time temperature data with management-oriented modeling. Making such results available to land managers in an accessible format, such as the web-based decision support tool created by this project, may help them make more precise, timely decisions.
Research Outcomes
Education and Outreach
Participation Summary:
Results of the project were presented to growers at the annual rice field days in 2011 and 2012 (posters attached). The results presented in this report are in the process of peer review and will be submitted to Agricultural Systems before the end of 2012. Further, the web tool is now housed on the Department of Plant Sciences web server and will be transferred to the rice extension site once the research has completed the process of peer-review. Currently, the web tool can be accessed at:
http://informatics.plantsciences.ucdavis.edu/rice_extension/
Information Products
- Site-specific temperatures improve weed emergence predictions in California rice: introducing a web-based decision support tool
- Predicting Weed Emergence in California Rice
- Predicting Weed Emergence in California Rice
- Site-specific, real-time temperatures improve the accuracy of weed emergence predictions in direct-seeded rice systems