What ranchers often want to know from research information is, what would happen if I did these different things on my ranch which has different soils, slopes, vegetation, cattle and rainfall? We are developing and testing the SPUR computer model to answer such questions for specific ranches. Much research does not and cannot deal with problems on different ranches or at the ranch scale. Models have the potential of being able to integrate much component research so the relevance and economic consequences for individual ranches, using different management combinations can be assessed at the ranch scale. This would be impossible to research with actual ranches.
However, before a model can be used for such purposes, it has to give reliable results when measured against production and biological measurements obtained in the field. To test how well the model compares with the real world we are testing the output from the SPUR model with production and biological measurements made on 3 Experimental Ranches in Texas; the Texas Experimental Ranch near Throckmorton and the Y and Waggoner Experimental Ranches, near Vernon.
Once calibrated for a particular location, the model can be run for more than one hundred years to predict the outcome of different management and different weather sequences, to assess best management strategies and combinations of management practices (Baker et al. 1993). The output from SPUR can be selected to include rainfall runoff, soil loss, grass production, soil organic matter, forage harvested by livestock and wildlife, animal weight and gain, and estimates of net economic return. It has the advantage of being able to assess the results of managing in a certain manner for decades, rather than just 2-3 years which happens with most research. When combined with cost data from commercial ranches in this area, this tool will enable us to assess the economic and biological consequences of different management strategies and practices in a much more precise manner than is currently possible.
In this paper SPUR2.4 output will be compared to observed values for one soil and one site on the Texas Experimental Ranch, Throckmorton to illustrate how well we can currently predict the outcome of climate, physical environment and different management, in this environment. Hydrology, soil organic matter, vegetation and cow-calf production are presented. Differences between SPUR2.4 and SPUR91 are emphasized.
SPUR2.4 integrates all previous versions of SPUR . The forage submodel (Baker et al. 1992) from SPUR2 predicts forage intake and diet selection by grazing herbivores, and the cow-calf submodel (Hanson et al. 1992) simulates all individuals of a herd for their entire life cycle based on defined genetic traits. The changes from SPUR91 (Carlson and Thurow 1992; 1996) make the model more applicable to Texas and include specification of rooting depth for each plant species. Finally, the three-component soil organic matter and nitrogen submodel from CENTURY (Parton et al. 1992) aim at improving soil carbon and nitrogen cycling identified by Carlson and Thurow (1992) as a weak component in the original SPUR .
The hydrology and plant production components of the model are compared with lysimeter data collected on one soil by Carlson et al. (1990) at the Texas Experimental Ranch. Very little data exists on rangeland soil carbon. To parameterize this portion of the model, unpublished litter biomass and plant standing crop field data from the Texas Experimental Ranch were used. The work of Burke et al. (1989) was used to evaluate this portion the model. It characterizes soil carbon levels in the Great Plains rangeland according to long-term mean annual precipitation and mean daily temperature. Livestock production from the model is compared with data collected in grazing experiments conducted by Heitschmidt et al. (1990) at the Texas Experimental Ranch and unpublished information from the same venue (TAES 1985).
Monthly soil water content (mm) was over-estimated by SPUR91 and SPUR2.4 (Figure 1) . Although the data set only covers from October 1988 through July 1989, the correlation between observed and simulated soil water content to 100 cm was high for both models (SPUR91, r=0.861; SPUR2.4, r=0.972). Monthly runoff (mm) was also over-estimated by both models (Figure 2) , although monthly wet year runoff (1986) was more highly correlated (r=0.757 in both models) than monthly dry year runoff (1988)(SPUR91, r=0.678; SPUR2.4, r=0.643).
Total simulated runoff for 3 years was twice the observed runoff (Table 1). Simulated monthly potential evapotranspiration (mm) was equivalent from both models, but the correlation to observed data was low (r=0.491) (Table 1). The simulated actual evapotranspiration for these 3 years was also less than observed (Table 1), while deep percolation was over-estimated by both SPUR91 and SPUR2.4. Sediment production was under-estimated by both models (Table 1). It was under-estimated in 1986 but reasonably accurate with both models in the following 2 years (Figure 3) .
Using the long-term mean annual precipitation and mean daily temperature, the formula of Burke et al. (1989) predicts a soil carbon level of 4009 g.m-2 for the Texas Experimental Ranch. This same formula predicts a soil carbon level of 3767 g.m-2 at this location for the period of simulation (1986-1995). The observed soil carbon content in the topsoil for rangeland in this area is 2900 g.m-2 (Natural Resource Conservation Service, Temple Texas).
The published parameter values set in the one-component submodel of SPUR91 (Carlson and Thurow 1992) predicts very low soil carbon levels (1200 g.m-2 ) (Table 2) . When this model is re- parameterized using litter mass and plant data from the Texas Experimental Ranch (Dowhower, unpublished data), SPUR91 predicts higher soil carbon levels (2900 g.m-2 ) (Table 2) . This is somewhat low compared to the levels predicted by Burke et al. (1989) but concurs with the NRCS soils data. The simulated soil carbon from the one-component submodel remains stable over the period of simulation using both sets of parameters (Figure 4) .
When the three-component soil carbon submodel from CENTURY is used, the initial soil carbon value was set at 4009 g.m-2 (after Burke et al. 1989, see above). The simulated mean for the three- component SPUR2.4 model is 3545 g.m-2 compared to the mean level of 3767 g.m-2 at this location for the period of simulation, using the Burke et al. (1989) equation (Table 2) . Over the length of the simulation period, soil carbon predicted by the three-component SPUR2.4 model declines slightly from the initial figure of 4009 g.m-2 (Figure 5) . This concurs with the slight overall downward trend shown by the fluctuating levels of soil carbon calculated using annual rainfall and temperature data with the Burke et al. (1989) equation (Figure 5) . The SPUR91 one-compartment model output increases slightly over the simulation period from the starting value of 2900 g.m-2 to same ending level (3500 g.m-2 ) as the three-component SPUR2.4 model (Figure 5) .
To be of value the model needs to adequately simulate the growth of key individual species through a growing season and simulate growth of these key species over as many years as possible. SPUR was designed to run for up to 100 years. However, most plant data sets exist for 2-4 year periods, based upon the graduate student cycle.
Regarding growth through the growing season, the original set of parameters with both SPUR91 and SPUR2.4, over-estimated live aboveground annual grass and midgrasses and under-estimated Texas wintergrass and shortgrasses (Table 3) . To improve model output the plant growth parameters were adjusted until the correspondence between observed and simulated monthly plant production for each species was as good as possible. This was an iterative process between adjusting decomposition proportion constants and adjusting plant growth parameters until as many species as possible reached maximum live peak standing crop and the correlation between observed and simulated monthly plant production was highest for as many species as possible. Initial values for litter, live roots, dead roots, and soil organic carbon were set too low in the original parameter set. Litter averages around 250 g.m-2.yr-1 (Steve Dowhower, unpublished data), while total roots for a dry shortgrass-to-mixed prairie ecosystem range from 490 to 1600 g.m-2 (Webb et al. 1983).
In the final set of parameters SPUR2.4 simulates live peak standing crop accurately for each species except shortgrass, which is under-estimated. SPUR91 under-estimates live peak standing crop for annual grass and accurately simulates all other plant species (Table 3) . The monthly live biomass of annual grasses, shortgrasses and midgrasses were all simulated reasonably well, while the Texas winter grass was slightly under-estimated (Figure 6) . When total aboveground biomass (live + dead) are considered, the same patterns are observed but estimates are less accurate (Figure 7) .
Regarding the yearly variation, persistence of plant species over time was very poor with the original set of parameters. Midgrasses soon dominated the sites and accounted for most of the total biomass (Figure 8) . Using the final set of parameters developed with the iterative process described earlier, all plant species persist for at least ten years in concurrence with field observations . However, Texas wintergrass is still under-estimated by both SPUR91 and SPUR2.4. This winter perennial is not adequately simulated by SPUR . Since SPUR does not simulate this very important component of the ecosystem at the Texas Experimental Ranch, live aboveground biomass and total (live+dead) aboveground biomass (g.m-2)is under-estimated for all species.
The original SPUR and SPUR91 only have a steer growth model. The addition of a cow-calf simulation capability by adding the Colorado Beef Cattle Producers Model (CBCPM)(Hanson et al. 1992) expands the utility of SPUR considerably. There is no experimental data for steer performance from the Texas Experimental Ranch but rates of gain per day range from 0.45 to 0.81kg based on data from this location (Heitschmidt et al. 1982a,b; TAES 1985).
Using the original steer growth submodel, when small- to moderately-sized steers (509 kg mature weight) are grazed continuously, yearlong at a moderate stocking rate for 10 years, simulated average daily gain (kg) from both sets of parameters and both SPUR91 and SPUR2.4 (0.521-0.868 kg.d-1) is within range for this part of Texas (TAES 1985). When average daily gain is plotted over the ten year cycle, the original set of parameters simulates steer gain somewhat higher than the final set of parameters. Carlson and Thurow (1996) hypothesized that SPUR91 under-estimated steer gain because nitrogen cycling was inadequate in SPUR91. When CENTURY is used in SPUR2.4, nitrogen levels and steer gain are increased using both sets of parameters. The other result of using CENTURY is that steer gain becomes more variable. In 1992 and 1993, steer gain per day falls below the minimum expected value of 0.45 kg per day.
When the CBCPM is used to estimate steer growth over summer (day 72 to day 222), average daily gain was 0.84 kgs per day (Table 4) . Mature weight for the steers was 655 kgs; well above the upper limit for the original steer model (large steer mature size = 585 kgs). When simulated Charolais x (Angus x Hereford) steers were grazed after weaning through winter until the following fall (day 291), they gained 0.66 kgs per day (Table 4) . Both of these estimates are satisfactory.
The animal performance data from the Texas Experimental Ranch was used to evaluate the cow-calf simulation capability of CBCPM in SPUR2.4. The most abundant soil at this location (Nuvalde clay- loam) was used to generate plant growth using the same weather data used in the plant production simulations above. Charolais bulls were bred to Angus x Hereford cows. The model slightly over- estimates mean cow weight (r=0.975) and slightly under-estimates mean calf weights (r=0.992) over this period (Figure 9) .
When the model was used to simulate cow-calf performance in the different stocking rate and grazing system trials conducted at the Texas Experimental Ranch (TAES 1985), kilograms of beef per cow and per acre are estimated reasonably accurately. Supplement fed was overestimated (Table 4) and replacement rates are over-estimated while conception rate is under-estimated (Table 5) .
The changes made to SPUR in creating first SPUR91 and then SPUR2.4 have improved the accuracy of the model considerably. The model is now able to do more than just predict general trends of management responses. The potential for aiding in the assessment of various management strategies and practices is probably now possible in limited areas, but much work remains to be done to expand this capability. The hydrology component in particular is not nearly accurate enough and this is one of the areas that the model needs to be strong in. The comments made by Carlson and Thurow (1996) are pertinent here.
The addition of the soil carbon and nitrogen submodel from CENTURY has improved model output. The soil carbon levels simulated by the model are within the range of what little data there is and the output is stable over the simulation period behaving qualitatively in line with expectation. However, the lack of good field data to test model output is a major drawback. Sites with different long-term grazing histories need to be identified and sampled to quantify long-term grazing effects on soil carbon and nitrogen.
SPUR91 made significantly improved model output of individual plant productivity and SPUR2.4 has improved this area of output even more. Both within and between season plant growth and long- term persistence of the key species simulated is good. However, the growth of Texas wintergrass needs to be improved. The inability to accurately predict Texas wintergrass growth is probably not too serious when considering livestock performance because this grass becomes very unpalatable when it matures and is almost totally avoided when mature.
Although the data set to test the steer model is very limited output of this submodel appears good with both SPUR91 and SPUR2.4. Carlson and Thurow (1996) hypothesized that SPUR91 under-estimated steer gain because nitrogen cycling was inadequate in SPUR91. When CENTURY is used in SPUR2.4, nitrogen levels and steer gain are increased.
The cow-calf submodel in SPUR2.4 gave very satisfactory initial results when compared with the data set from the Texas Experimental Ranch. A more complete test using this data is required, specifically simulating plant and cow-calf performance using the mixture of soils at the Texas Experimental Ranch under the different grazing management that was conducted at this venue for many years.
Looking to the future, it will be important to see how well model output concurs with field data from different locations using the same parameter values. The utility of the model will be greatly enhanced if a single set of parameter values can be used for a vegetation region. As mentioned above, there is a pressing need to improve the hydrology component in the model. Brush is also a significant factor in the southern Great Plains. Brush growth and population dynamics must be added to SPUR2.4 to address brush related issues. One of the major simplifying assumptions made in SPUR is that vegetation use over a grazing unit is uniform. While this may be a reasonable approximation for the relatively small areas used for research purposes, one of the major areas of concern at the ranch scale is the differential use of vegetation patches and areas around points of attraction, such as water points. Such uneven distribution of animal impact has significant impact on the range resource, hydrology and water quality. A spatial capability is needed to deal with these matters. A link up with a GIS, or incorporation of a pseudo-GIS, will be necessary to achieve this.