Livestock Management During Drought in the Northern Great Plains. I. a Practical Predictor of Annual Forage Production1

The Professional Animal Scientist 23 (2007):224–233

Livestock Management During Drought in the Northern Great Plains. I. A Practical Predictor of Annual Forage Production1

R. E. Kruse,* M. W. Tess,*2 and R. K. Heitschmidt† *Animal and Range Sciences Department, Montana State University, Bozeman 59717; and USDA- ARS, Fort Keogh Livestock and Range Research Laboratory, Miles City, MT 59301

minimum temperatures were used to minimize the negative impacts of ABSTRACT develop regression equations predicting drought on rangelands and beef enter- growing season forage production at prises. This research addressed the hypothe- the Fort Keogh Laboratory and Manyb- sis that spring precipitation data can erries Substation. At Fort Keogh Labo- Key words: forecasting, primary be used to detect agricultural drought ratory, a combination of fall (October production, simulation, modeling early in the growing season. The and November) and spring (April and Rangetek range model was used to May) precipitation were predictors of INTRODUCTION simulate yearly forage data based on simulated forage yield index (P < historical precipitation and tempera- 2 0.01, R = 0.84). At Manyberries Sub- Drought is an inherent trait of ture records from the USDA-ARS Fort station, April and May precipitation most rangelands including those Keogh Livestock and Range Research were predictors of simulated forage found in the Northern Great Laboratory (Miles City, MT) and the < 2 yield index (P 0.01, R = 0.44). Us- Plains. For example, Hurtt (1951) Agriculture and Agri-Food Canada Ma- ing the actual forage data from Ma- reported drought occurred once ev- nyberries Substation (Lethbridge, AB, nyberries Substation yielded similar re- ery 5 yr in southeastern Montana, Canada). Monthly total precipitation sults, in that April, May, and June and Johnson (1985) reported and monthly average maximum and were predictors of forage production (P drought occurred once every 4 yr < 2 0.01, R = 0.50). Although the re- in southeastern Alberta. gression equation for actual forage pro- The impacts of drought on range- duction data from Manyberries Substa- 1Research was conducted under a coopera- land ecosystems are numerous and tive agreement between USDA Agriculture tion did indicate that July precipita- well documented with much of the Research Service and the Montana Agricul- tion was a significant predictor, early ecological research conducted ture Experiment Station. Mention of a pro- adding July precipitation did not in- on Great Plains rangelands being prietary product does not constitute a guar- crease the ability of the equation to de- driven by the droughts of the antee or warranty of the product by USDA, tect reduced forage production. These 1930s (Weaver and Albertson, Montana Agriculture Experiment Station, results imply that annual forage pro- 1936, 1939, 1944; Ellison and or the authors and does not imply its ap- duction can be estimated with consid- Woolfolk, 1937; Whitman et al., proval to the exclusion of other products erable confidence by July 1 and that 1943; Albertson et al., 1957). In that may also be suitable. The USDA-ARS forage produced by early July is a good general, research has shown that Northern Plains Area is an equal opportu- indicator of total growing season for- nity and affirmative action employer and drought conditions reduce both all agency services are available without dis- age production. Early season detection quantity and quality of forage pro- crimination. of drought effects on forage production duced and consumed with the re- 2Corresponding author: mwtess@ provides much-needed flexibility in de- sulting effect being a general de- montana.edu vising management alternatives to cline in animal production on both Prediction of Forage Production During Drought 225

an individual animal and per unit age yield (Rangetek) to simulate season, date of peak standing crop, area of land basis (Valentine, 1990; long periods of time representing date of end of growing season, and Weltzin and McPherson, 2003; the 2 locations (Miles City and a relative growth curve. Soil inputs Heitschmidt and Vermeire, 2005, Lethbridge). An existing large data include number of soil layers, tex- 2006). set (Lethbridge, AB) was used to ture, thickness, percent OM, per- For livestock managers a funda- evaluate the simulation approach. cent sand, percent clay, percent mental problem with drought is its Rangetek (Wight and Neff, 1983), rock fragments, bulk density, and unpredictability. Hence, livestock a modified version of the Ekalaka initial soil water. Runoff and site management strategies and tactics Rangeland Hydrology and Yield variables include longitude, lati- are often reactive rather than proac- Model (ERHYM-II; Wight and tude, elevation, slope, and aspect. tive. In this study the effects of Hanks, 1981; Wight, 1987) was Water enters the system as precip- drought were confined to effects used to simulate yearly forage yield itation and may leave the system on forage production. The broad based on historical precipitation as runoff, deep drainage, evapora- objective of this study was to iden- and temperature records from the tion, or plant transpiration. As wa- tify key climatic variables associ- USDA-ARS Fort Keogh Livestock ter accumulates in the soil, progres- ated with drought in the Northern and Range Research Laboratory sively deeper soil layers are filled to Great Plains whereby impending ef- near Miles City, MT (46° 22′ N field capacity, and excess water fects on forage production could be 105° 5′ W) and from the Agricul- drains to the next layer. Water may predicted with considerable confi- ture and Agri-Food Canada Manyb- be lost from the system as runoff dence. Previous research at this lo- erries Substation near Lethbridge, or deep drainage if soil profile is sat- cation has shown that on average AB, Canada (49° 7′ N, 110° 28′ W). urated. Soil evaporation and plant about 90% of Northern Great The 30 yr of climate data from Fort transpiration, in that order, remove Plains annual graminoid produc- Keogh Laboratory and 50 yr of data water from the soil profile, begin- tion is completed by early July from the Manyberries Substation ning with the uppermost layer. Po- (Heitschmidt and Vermeire, 2005). (Smoliak, 1986) were used to de- tential evapotranspirative demand As a result, we hypothesized that velop regression equations pre- is calculated with the Jensen and high-impact variables would most dicting growing season forage pro- Haise (1963) evapotranspiration likely be from a fairly short period duction (simulated yield index). equation. The potential demand is of time (e.g., spring) rather than an The 50 yr of forage production partitioned into potential transpira- entire frost-free growing season. data were also used from the Ma- tion and potential evaporation; val- Successfully meeting this objective nyberries Substation (Smoliak, ues are used to calculate actual tran- was prerequisite to developing the 1986) to develop regression equa- spiration and evaporation (John- cow-calf drought management strat- tions predicting actual forage son, 1985). egies presented in a companion pa- yield. Output from the model is in the per (Kruse et al., 2007). form of a yield index. The yield in- Rangetek Model dex calculated by the model is an MATERIALS AND METHODS Rangetek is a climate or water- estimate of plant growth for cur- balance model, which provides rent climatic conditions and site pa- Our approach was to develop daily simulation of soil water evap- rameters and is expressed as a frac- and evaluate predictors (multiple re- oration, transpiration, runoff, and tion of site potential yield (John- gression equations) of annual for- soil water routing for individual son, 1985). This cumulative index age yield from 3 different data sets: range sites. It can utilize real-time equals the ratio of actual transpira- one simulated based on climatic climate data to simulate ongoing tion to potential transpiration. The and soil data from Miles City, MT; processes, or it can utilize long- index, or predicted total plant one simulated based on climatic term weather records to simulate yield, is calculated on the date that and soil data from Lethbridge, AB, runoff and herbage production un- peak standing crop occurs. The Canada; and one composed of ac- der a variety of climatic conditions product of site potential yield (kg/ tual forage and climatic data col- and management practices (Wight, ha) and the yield index therefore lected at Lethbridge, AB. Prelimi- 1987). Driving variables are daily provides an estimate of cumulative nary analyses of several models re- precipitation, minimum and maxi- production (Johnson, 1985). The vealed small local data sets (i.e., mum air temperatures, and solar ra- yield index is considered a good in- limited to a few years and multiple diation, which can be simulated by dicator of the growing season cli- sites in Miles City, MT) were inade- the model. Plant variables required mate as it relates to plant growth quate to satisfactorily address our by the model include percent gram- and enables comparisons of range objectives. Hence, we used a locally inoids, percent bare ground, leaf treatments or vegetation invento- validated model of rangeland for- area index, date of start of growing ries among years or range sites by 226 Kruse et al.

accounting for a large portion of are loamy Aridic Haploborolls and Data were collated to correspond climate-induced variation in plant vegetation belongs to the Stipa-Bou- to the growing season year to en- response (Wight, 1987). teloua faciation of the Mixed Prairie able a measure of agricultural Association (Smoliak, 1986). Princi- drought, when moisture during the Fort Keogh Laboratory, pal herbaceous species are needle- growing season is inadequate to Miles City, MT and-thread, western wheatgrass, support healthy forage growth to blue grama, prairie junegrass (Koel- maturity, prevent extreme forage Herbage production data were eria cristata (Lam.) Beauv.), Sand- stress, and permit normal forage from an unpublished study con- berg’s bluegrass (Poa sandbergii Va- production (Kulshreshtha, 1989; ducted by the USDA Natural Re- sey), and threadleaf sedge (Smoliak, Thurow and Taylor, 1999). Based source Conservation Service over a 1986). The dominant shrub is on previous studies at both loca- 3-yr period (1991 to 1993) at the fringed sagewort (Artemisia frigida tions (Smoliak, 1986; Heitschmidt Fort Keogh Laboratory. Climate at Willd.). and Vermeire, 2005), the regressed Miles City is semiarid with annual The single, end-of-growing sea- 12-mo growing year was defined as precipitation averaging 34 cm. Veg- son (i.e., late September) harvest extending from the previous Au- etation at Fort Keogh Laboratory is data set spanned 53 yr (1930 to gust through the current July. grama-needlegrass-wheatgrass (Bou- 1983). The data set had been used Multiple regression methods teloua-Stipa-Agropyron) mixed grass in related study published by Smol- were used to determine the effects dominant (Kuchler, 1964). Domi- iak (1986). Site variables included of the climatic factors on range for- nate species are western wheatgrass Wardlow soil series type, north as- age production; i.e., simulated [Pascopyrum smithii Rydb. (Love)], pect, an elevation of 938 m, and a yield index at both locations and threadleaf sedge (Carex filifolia slope of 3%. Vegetation variables actual yield at Manyberries Substa- Nutt.), needle-and-thread [Hesperos- included an average yield of 388 tion. The stepwise multiple regres- tipa comata (Trin. & Rupr.) Barkw- kg/ha, 79% gramminoids, 30% bare sion procedure of SAS (SAS Inst. orth], blue grama [Bouteloua gracilis ground, and a leaf area index of Inc., Cary, NC) was used to select (H. B. K.), and Japanese brome (Bro- 0.70. General soil parameters for the most accurate prediction equa- mus japonicus Thunb. ex Murr.]. the Alberta site were found in the tions. Independent, or predictor, The dominant shrub is Wyoming Canadian Soil Series survey. variables used were monthly total big sagebrush (Artemisia tridentata precipitation and monthly average subsp. wyomingensis Beetle and Statistics maximum temperatures. Variables Young). remained in the model only at an Summarizing from above, 3 data Annual production was measured α-level of 0.05. Best predictors were sets were used: 1) 30 yr of simu- on 12 sites from a single harvest identified, first using all available lated production (i.e., yield index) near time of peak standing crop data (monthly temperature and pre- at Fort Keogh Laboratory; 2) 50 yr (i.e., mid-July; Table 1). General cipitation). A second set of pre- of simulated production (i.e., yield soil characteristics were found in dictors was identified constraining index) at Manyberries Substation; the Natural Resource Conservation the predictor variables to months and 3) 50 yr of field estimated pro- Service survey for Custer County, prior to July. Because temperature duction (i.e., forage yield) at Ma- MT. Initial soil water percentage variables contributed little to the nyberries Substation. Our hypothe- for all sites was reported by Heitsch- predictive ability of the equations, sis was that the effects of emerging midt et al. (1999). Temperature a third set of equations was devel- drought on forage production and precipitation data (NOAA, oped using only monthly precipita- could be reliably identified from 2001) were from Miles City Air- tion prior to July. port, located about 10 km from temperature and precipitation data study sites. Simulated yield index collected from a subset of the grow- values for the 12 sites were aver- ing season (e.g., before July 1). Sim- RESULTS AND DISCUSSION aged to produce a single predicted ulated data were used because ac- index value for statistical analyses. tual forage production data were Fort Keogh Laboratory, not available for the area of our pri- Miles City, MT Manyberries Substation, mary interest (Montana). The simu- Measured precipitation over the Lethbridge, AB lation model was developed and validated for this area (Wight and 30-yr period was quite variable (Ta- Climatic, site, and vegetation Neff, 1983). For the Alberta loca- ble 2). Greatest annual precipita- data were collected at the Manyber- tion both simulated and actual tion recorded was 51.5 cm in 1978 ries Substation. Climate at Leth- data were used to provide a type of and the least was 13.4 cm in 1988. bridge is semiarid with annual pre- validation to the use of the simu- Mean annual precipitation for the cipitation averaging 33 cm. Soils lated data. area was 35.2 cm. On average, 72% Prediction of Forage Production During Drought 227

Table 1. Input variables for the Rangetek model from 12 sites at USDA-ARS Fort Keogh Livestock and Range Research Laboratory, Miles City, MT

Slope Average yield2 Grass Bare Soil series Range site Aspect1 Elevation (m) (%) (kg/ha) (%) ground (%) Cabbart Shallow NNE 2,600 8 1,038 75 35 Cabbart Shallow NNE 2,640 6 985 75 35 Cambeth Thin silty ENE 2,720 15 997 90 25 Creed Claypan SSE 2,540 1 1,442 75 40 Ethridge Clayey N 2,642 1 1,270 75 20 Kobase Clayey ESE 2,424 5 1,717 70 20 Twilight Thin sandy E 2,560 15 1,094 85 27 Twilight Sandy SSW 2,660 18 1,122 85 10 Yamacall Silty NE 2,540 10 1,458 85 5 Wabek Shallow to gravel N 2,622 1 1,001 65 30 Wabek Shallow to gravel NW 2,570 10 719 65 30 Wabek Shallow to gravel SE 2,600 18 1,362 65 30

1NNE = North-northeast; ENE = east-northeast; SSE = south-southeast; N = north; ESE = east-southeast; SSW = south- southwest; NE = northeast; NW = northwest; SE = southeast. 2Average yield is the forage production data averaged over 3 yr for each site.

of the annual precipitation occurred in June, July, and August to those measured prior to July, the curred during the growing season (Table 2). High temperatures and resulting equation included Octo- (April to September) and 61% of low precipitation in late summer ber and November from the prior growing season precipitation oc- coincided with summer dormancy. calendar year, April and May precip- curred during the spring (April to Predicted and Simulated Yield itation, and April minimum temper- June). May and June had the great- Indexes. Using all available cli- ature (R2 = 0.89). The last step was est average monthly precipitation matic data, the best prediction to constrain variables to only pre- and also exhibited the greatest vari- equation for simulated yield index cipitation prior to July because tem- ation in precipitation, a fact which included April and May precipita- perature explained very little of the in itself suggests May and June pre- tion, January and July average max- variation in simulated yield index cipitation would be important for imum temperature, and May mini- (Table 3). The ﬁnal equation in- predicting drought. The warmest mum temperature (R2 = 0.87, Table cluded previous October and No- maximum daily temperatures oc- 3). Constraining predictor variables vember and April and May precipi-

Table 2. Average monthly precipitation, maximum and minimum temperatures for Miles City, MT

Precipitation Maximum Minimum Month (cm) SD temperature (°C) SD temperature (°C) SD January 1.05 1.07 −0.82 6.05 −13.75 5.03 February 1.15 1.02 3.28 6.17 −9.77 4.99 March 2.68 2.38 9.69 4.42 −4.69 2.38 April 3.54 3.18 16.97 3.94 1.22 1.76 May 6.33 4.25 22.64 3.49 7.16 1.36 June 5.55 3.68 27.95 2.95 12.39 1.61 July 3.25 2.57 31.38 1.91 15.58 1.56 August 3.25 2.66 28.49 3.54 14.89 1.89 September 3.28 2.49 20.36 3.88 8.24 1.71 October 2.67 3.33 12.28 5.03 1.72 1.32 November 1.16 1.18 3.93 4.04 −5.79 2.81 December 1.34 0.91 −1.07 4.50 −11.69 4.02 Total 35.22 11.12 — — — — 228 Kruse et al.

indexes being greater for Rangetek Table 3. Regression equations predicting simulated forage yield simulations than regression pre- index from precipitation and temperature data for Miles City, MT1 dictions.

2,3 2 2 Regression equations R MSE Manyberries Substation, 1 Y = 1.348 + 0.037 × (April Precip) + 0.018 × (May Precip) − 0.87 0.155 Lethbridge, AB 0.005 × (January Max Temp) − 0.031 × (July Max Temp) − 0.034 × (May Min Temp) Measured precipitation over the 2 Y = 0.048 + 0.014 × (October Precip) + 0.039 × (November 0.89 0.158 50-yr period was quite variable (Ta- Precip) + 0.041 × (April Precip) + 0.023 × (May Precip) − ble 5). Greatest annual precipita- 0.025 × (April Min Temp) tion recorded was 60.9 cm in 1965, 3 Y = 0.008 + 0.015 × (October Precip) + 0.044 × (November 0.84 0.186 and the least was 20.5 cm in 1943. × × Precip) + 0.041 (April Precip) + 0.020 (May Precip) Mean annual precipitation for the area was 32.8 cm. On average, 67% 1MS error (MSE) units are squared index units. of the annual precipitation oc- 2 Precip = precipitation, Max Temp = maximum temperature, Min Temp = curred during the growing season minimum temperature. 3 (April to September) and 61% of Equation 1 was developed using all available climatic variables (precipitation, growing season precipitation oc- maximum and minimum temperatures); Equation 2 was developed using all curred during spring (April to variables prior to July; Equation 3 was developed using only precipitation variables prior to July. June). May and June had the high- est average monthly precipitation, and the warmest maximum daily temperatures occurred in June, tation as predictors of simulated precipitation variables explained a July, and August (Table 5). yield index (R2 = 0.84). considerably greater amount of the Correlations between actual for- It is interesting to note that all of variation than temperature age yield and different measures of the temperature variables, whether variables. precipitation were similar to those maximum or minimum, had nega- There was a strong relationship described by Smoliak (1986) when tive coefﬁcients, and all the precipi- between the simulated yield index the climatic data were collated by tation variables had positive coefﬁ- from Rangetek and the predicted calendar year (data not shown). For cients (Table 3). This implies that yield index from regression equa- our analyses, data were collated by herbage production was enhanced tion 3 from Table 3 (Figure 1). growing season, with data from the by cool, wet weather. However, par- Trends were nearly identical, with 1942 growing season removed as tial R2 values (Table 4) showed that magnitude of annual shifts in yield outliers (recorded extremely high forage production). Therefore, due to differences in the analyses of and objectives for the data, results reported herein are different from Table 4. Partial R2 of variables for each regression equation those reported by Smoliak (1986). predicting simulated forage yield index for Miles City, MT Predicted and Simulated Yield Partial R2 for regression equations1 Indexes. The best regression equation for predicting forage yield in- Variable 1 2 3 dex included only April and May × October precipitation — 0.06 0.08 precipitation [Y = 0.320 + 0.014 November precipitation — 0.06 0.07 (April precipitation) + 0.016 × (May April precipitation 0.39 0.39 0.39 precipitation), R2 = 0.44, MS error = May precipitation 0.29 0.29 0.29 0.101]. The stepwise procedure January maximum temperature 0.03 — — yielded the same model regardless July maximum temperature 0.10 — — of level of restriction on the driv- April minimum temperature — 0.08 — ing variables. April precipitation ex- May minimum temperature 0.06 — — plained 14% of the variation and Total 0.87 0.89 0.84 May precipitation 30%. 1Equation 1 was developed using all available climatic variables (precipitation, Predicted yield indexes from the maximum and minimum temperatures); Equation 2 was developed using all regression equation and simulated variables prior to July; Equation 3 was developed using only precipitation yield indexes from Rangetek (Fig- variables prior to July. ure 2) were less similar than at Miles City (Figure 1). As with the Prediction of Forage Production During Drought 229

ing only precipitation variables, the best equation included Febru- ary, April, May, and June (R2 = 0.56). February precipitation only explained 7% of the variation in actual forage yield (Table 7); however, its biological signiﬁcance is suspect because such little precipitation is experienced during February and the soil would be frozen, pre- sumably limiting precipitation’s effect. When February precipitation was removed, and the resulting equation included April, May, and Figure 1. Simulated yield index output from Rangetek (dashed line) and the predicted June (R2 = 0.50). yield index (solid line) using the regression model developed from monthly precipitation As was found in Miles City, all of variables (Equation 3, Table 3) for Miles City, MT. the temperature variables had negative coefﬁcients (Table 6). This described the relationship between Miles City data, trends in annual Predicted and Actual Forage temperature and forage yield index shifts in yield indexes were similar, Yield. Using all available climatic to be negative and consistent be- but magnitude of shifts was consid- data, the best regression equation tween the 2 locations. Table 7 erably greater for Rangetek simula- predicting forage yield, as esti- shows the partial R2 values, which tions than regression predictions. mated from September harvests were again consistent with Miles This may be due to chance alone (Smoliak, 1986), included April, City; i.e., temperature variables ex- or some fundamental differences June, and July precipitation, plus plained only a small amount of between locations. For example, April maximum temperature and variation. July precipitation did the growing seasons at Lethbridge November and June minimum tem- have a relatively high partial R2 and Miles City are quite different, perature (R2 = 0.71, Table 6). Con- value when predicting actual for- as below freezing average monthly straining predictors to those mea- age data, suggesting July precipita- temperatures prevailed in Leth- sured prior to July yielded an equation is a key variable affecting for- bridge between October and April tion that included September, age production in Alberta. (Table 5) and in Miles City below April, May, and June precipitation, Predicted annual forage produc- freezing average monthly tempera- January maximum temperature, tion followed the same trends as ac- tures occurred only November and previous year’s August mini- tual production (Figure 3). But as through March (Table 2). mum temperature (R2 = 0.66). Us- with the simulated Rangetek yield

Table 5. Average monthly precipitation, maximum and minimum temperatures for Lethbridge, AB, Canada

Precipitation Maximum Minimum Months (cm) SD temperature (°C) SD temperature (°C) SD January 2.16 1.62 −6.93 5.88 −18.19 5.47 February 1.82 1.61 −4.02 5.44 −15.09 4.83 March 2.20 1.64 1.55 4.03 −9.88 3.28 April 2.97 2.70 11.23 3.39 −1.86 2.15 May 4.12 2.91 18.14 2.29 4.29 1.36 June 6.22 3.75 22.27 2.09 8.69 1.19 July 3.23 2.33 27.42 1.94 11.66 1.16 August 2.95 2.23 26.49 2.22 10.74 1.37 September 2.47 2.02 20.11 2.75 5.15 1.70 October 1.54 1.39 13.32 2.77 −0.85 1.63 November 1.54 1.28 3.50 3.77 −8.58 2.98 December 1.87 1.39 −2.93 4.15 −14.21 3.75 Total 32.81 8.86 — — — — 230 Kruse et al.

lated data sets revealed that spring and early summer precipitation were important determinants of level of annual forage production. This project included analyses of both actual and simulated data. Simulated forage yield was used because we did not have a data set representing our immediate region (i.e., eastern Montana) that was suf- ﬁcient to address our objectives. Ac- tual historical data (Alberta) served to provide not only a second loca- tion, but also a check on the simulation approach. Prediction equations for forage production based Figure 2. Simulated yield index output from Rangetek (dashed line) and the predicted on simulated or actual data yielded yield index (solid line) using the regression model developed from monthly precipitation variables (see text) for Lethbridge, AB, Canada. similar conclusions, providing evi- dence that our conclusions based on use of the Rangetek model in Miles City were reasonable. Our approach was to develop indexes (Figure 2), magnitude of an- to be greater for Rangetek simu- practical early predictors of forage nual shifts in production were not lated values than actual produc- production as affected by emerging well predicted in some instances. tion. This may imply that Rangetek drought. Early prediction is prereq- Figure 4 compares annual Rangetek simulations were affected more by uisite to providing livestock manag- simulated yield index values to ac- April and May precipitation and ers with the opportunity to make tual annual forage production. The less by summer precipitation than meaningful management changes correlation between simulated actual production. This in turn sug- that minimize the negative effects yield index and actual forage pro- gests caution should be exercised of drought on rangelands as well as duction was 0.59 (P < 0.01). Direc- when using Rangetek to predict for- beef enterprises (Kruse et al., 2007). tional trends were similar among age production in Alberta. Still, Our results suggest that forage pro- years, but annual variations tended analyses of both actual and simu- duction can be predicted with

Table 6. Regression equations predicting actual forage yield from precipitation and temperature data for Lethbridge, AB, Canada

Regression equations1,2 R2 MSE3 1 Y = 547.08 + 15.38 × (April Precip) + 13.74 × (June Precip) + 28.13 × (July Precip) 0.71 126,393 − 11.79 × (April Max Temp) − 8.98 × (November Min Temp) − 38.04 × (June Min Temp) 2 Y = 378.85 + 19.89 × (September Precip) + 22.91 × (April Precip) 0.66 117,055 + 18.14 × (May Precip) + 16.77 × (June Precip) − 7.00 × (January Max Temp) − 32.83 × (August Min Temp) 3 Y = 93.63 + 24.64 × (February Precip) + 26.33 × (April Precip) 0.56 151,556 + 11.94 × (May Precip) + 18.69 × (June Precip) 3b Y = 136.77 + 30.01 × (April Precip) + 12.76 × (May Precip) 0.50 180,103 + 16.68 × (June Precip)

1Precip = precipitation, Max Temp = maximum temperature, Min Temp = minimum temperature. 2Equation 1 was developed using all available climatic variables (precipitation, maximum and minimum temperatures); Equation 2 was developed using all variables prior to July; Equation 3 was developed using only precipitation variables prior to July; Equation 3b was developed the same as Equation 3, but removing February from possible variables. 3MS error (MSE) units are (kg/ha)2. Prediction of Forage Production During Drought 231

lect average temperature measure- Table 7. Partial R2 of variables for each regression equation ments; many ranchers keep track predicting actual forage yield for Lethbridge, AB, Canada of precipitation.

2 1 The majority of the research re- Partial R for regression equations ported in the last 70 yr has shown Variable 1 2 3 3b that some measure of precipitation is one of the most important fac- September precipitation — 0.06 — — tors in forage growth (see review February precipitation — — 0.07 — April precipitation 0.28 0.28 0.28 0.28 by Kruse, 2002). Sneva and Hyder May precipitation — 0.05 0.05 0.05 (1962a,b) performed similar work June precipitation 0.17 0.17 0.17 0.17 utilizing yield indices and found July precipitation 0.15 — — — that semiarid communities of na- January maximum temperature — 0.04 — — tive or well-adapted introduced spe- April maximum temperature 0.03 — — — cies fluctuated with precipitation August minimum temperature — 0.06 — — in a fairly uniform and predictable November minimum temperature 0.03 — — — manner when data were collated June minimum temperature 0.06 — — — Total 0.71 0.66 0.56 0.50 during a growing season. Wight et al. (1984) used the Rangetek model 1Equation 1 was developed using all available climatic variables (precipitation, to relate soil water and climatic pa- maximum and minimum temperatures); Equation 2 was developed using all rameters to plant growth. They variables prior to July; Equation 3 was developed using only precipitation found that two-thirds of the field- variables prior to July; Equation 3b was developed the same as Equation 3, but measured yields were within one removing February from possible variables. SD of forecasted yields for the April, May, and June forecasts using 55 yr of weather records and 12 yr of actual yield and soil water some confidence from spring (April forage conditions. Temperature and data. Hanson et al. (1982) com- through June) precipitation at both July precipitation variables in- pared long-term historical and sto- Miles City, MT, and Lethbridge, Al- creased the ability of the regression chastically generated weather re- berta. Although temperature and equations to explain the variation cords in terms of their statistical at- July precipitation added reliability seen in forage production, but the tributes and effects on herbage to the models statistically, they proportion of the variation ex- yield and runoff forecasts calcu- added complexity and decreased plained by each variable was small lated from the Rangetek model sim- the utility of the models as early, and inconsistent. To us it appears ulations. Yield forecasts were simi- practical predictors of forthcoming less practical for producers to col- lar using either historical or syn- thetic weather records. Our results suggest that predictions based on spring precipitation could be used effectively to make management decisions as early as July 1. Sneva and Hyder (1962a,b) in eastern Oregon and Smoliak (1956) in Southern Alberta declared that predictions of forage production could be utilized by July 1 as well. Although Smoliak (1986) con- ceded, as we do, that August 1 is a more reliable date than July 1 for predicting drought, we advocate July 1 because 1) it has been shown that on average 91% of pe- rennial grass forage production is completed by July 1 at Fort Keogh Laboratory (Heitschmidt and Ver- Figure 3. Actual forage production (dashed line) vs. predicted forage production (solid meire, 2005); and 2) July 1 pro- line, Equation 3b, Table 6) forage production data for Lethbridge, AB, Canada. vides one additional month of time 232 Kruse et al.

Heitschmidt, R. K., and L. T. Vermeire. 2006. Can abundant summer precipitation counter losses in herbage production caused by spring drought? J. Range. Ecol. Manage. 59:392. Herbel, C. H., J. D. Wallace, M. D. Finkner, and C. C. Yarbrough. 1984. Early weaning and part-year confinement of cattle on arid rangelands of the southwest. J. Range Man- age. 37:127. Hurtt, L. C. 1951. Managing Northern Great Plains cattle ranges to minimize the effects of drought. USDA Cir. No. 865. Jensen, M. E., and H. R. Haise. 1963. Esti- mating evapotranspiration from solar radia- tion. Amer. Soc. Civil Eng. Proc. J. Irrig. Drain. Div. 89:15. Johnson, C. A. 1985. Simulated effects of supplementary water on two grasslands, Figure 4. Simulated yield index values from Rangetek (solid line) vs. actual forage Agropyron smithii and Bouteloua gracilis. production data (dashed line) for 50 yr for Lethbridge, AB, Canada. M.S. Thesis, Montana State University, Bozeman. Kruse, R. E. 2002. Beef cattle management decisions relating to drought in the North- ern Great Plains. M.S. Thesis, Montana State University, Bozeman. to respond to drought. Still, re- ACKNOWLEDGMENTS Kruse, R. E., M. W. Tess, and R. K. Heitsch- sponding to drought in August is midt. 2007. Livestock management during certainly better than failing to re- drought in the Northern Great Plains. II. The authors gratefully acknowl- Evaluation of alternative strategies for cow- spond at all. edge the Agriculture and Agri-Food calf enterprises. Prof. Anim. Sci. 23:234. It has been shown that use of pre- Canada Manyberries Substation for Kuchler, A. 1964. Potential natural vegeta- diction equations should be lim- graciously sharing data used in par- tion of the coterminous United States. Spe- cial Publication 36. Am. Geogr. Soc., New ited to areas of similar vegetation tial completion of this project. and soil type (Cannon and Niel- York, NY. son, 1984). Differences in equa- Kulshreshtha, S. N. 1989. Agricultural LITERATURE CITED drought impact evaluation model: A sys- tions developed for Fort Keogh Lab- tems approach. Agric. Syst. 30:81. oratory and Manyberries Substation Albertson, F. W., G. W. Tomanek, and A. NOAA. 2001. National Oceanic and Atmo- appear to support this concept. Riegel. 1957. Ecology of drought cycles and spheric Administration. U.S. Department of However, for the specific purpose grazing intensities on grassland of the Cen- Commerce. http://www.noaa.gov/in- of early prediction of drought, all tral Great Plains. Ecol. Monogr. 27:27. dex.html Accessed Nov. 1, 2001. the models define spring and early Cannon, M. E., and G. A. Nielson. 1984. Es- Smoliak, S. 1956. Influence of climatic con- summer as predictors of forage pro- timating production of range vegetation ditions on forage production of shortgrass from easily measured soil characteristics. rangeland. J. Range Manage. 9:89. duction. Soil Sci. Soc. Am. J. 48:1393. Smoliak, S. 1986. Influence of climatic con- Ellison, L., and E. J. Woolfolk. 1937. Effects ditions on production of Stipa-Bouteloua of drought on vegetation near Miles City, prairie over a 50- year period. J. Range Man- IMPLICATIONS Montana. Ecology 18:329. age. 39:100. Hanson, C. L., J. R. Wight, J. P. Smith, and Sneva, F. A., and D. N. Hyder. 1962a. Esti- Ranchers can use weather records S. Smoliak. 1982. Use of historical yield mating herbage production on semiarid data to forecast range herbage production. ranges in the intermountain region. J. to reasonably forecast forage pro- J. Range Manage. 35:614. Range Manage. 15:88. duction by July 1. Practically, for- Heitschmidt, R. K., M. R. Haferkamp, M. G. Sneva, F. A., and D. N. Hyder. 1962b. Fore- age produced by early July is a Karl, and A. L. Hild. 1999. Drought and casting range herbage production in east- good indicator of growing season grazing: Effects on quantity of forage pro- ern Oregon. Oregon Exp. St. Bull. No. 588. duced. J. Range Manage. 52:440. forage production. Producers can Thurow, T. L., and C. A. Taylor. 1999. View- then detect drought-reduced forage Heitschmidt, R. K., and L. T. Vermeire. point: The role of drought in range manage- production and change manage- 2005. An ecological and economic risk ment. J. Range Manage. 52:413. avoidance drought management decision ment strategies early in the grow- Valentine, J. F. 1990. Grazing Management. support system. In: Pastoral Systems in Mar- Academic Press, Inc., San Diego, CA. ing season to forestall some of the ginal Environments. J. A. Milne, ed. p. 178. negative impacts of drought on Proc. Satellite Workshop, XX Int.Grassland Weaver, J. E., and F. W. Albertson. 1936. Ef- Congr., Glasgow, Scotland. Wageningen Ac- fects of the great drought on the prairies of range resources and livestock per- ademic Publ. Wageningen, The Neth- Iowa, Nebraska, and Kansas. Ecology formance. erlands. 17:567. Prediction of Forage Production During Drought 233

Weaver, J. E., and F. W. Albertson. 1939. Whitman, W., H. C. Hanson, and R. A. Pe- Wight, J. R., C. L. Hanson, and D. Major changes in grassland as a result of terson. 1943. Relation of drought and graz- Whitmer. 1984. Using weather records continued drought. Bot. Gaz. 100:576. ing to North Dakota rangelands. North Da- with a forage production model to forecast kota Agric. Exp. St. Bull. 320. Weaver, J. E., and F. W. Albertson. 1944. range forage production. J. Range Manage. 37:3. Nature and degree of recovery of grassland Wight, J. R. 1987. ERHYM-II: Model descrip- from the great drought of 1933 - 1940. tion and user guide for the BASIC version. Ecol. Monogr. 14:393. USDA, ARS 59, Beltsville, MD. Wight, J. R., and E. L. Neff. 1983. Soil-vege- Weltzin, J. F., and G. R. McPherson. 2003. tation-hydrology studies Vol. II. A User Changing Precipitation Regimes and Terres- Wight, J. R., and R. J. Hanks. 1981. A wa- Manual for ERHYM: The Ekalaka Rangeland trial Ecosystems: A North American Perspec- ter-balance, climate model for range herb- Hydrology and Yield Model. USDA-ARS. tive. Univ. Arizona Press, Tucson. age production. J. Range Manage. 34:307. ARR-W-29, Oakland, CA.