Continuous versus rotational grazing, again: Another perspective from meta-analysis A quantitative analysis of Briske et al. 2008 Kristina Wolf, U.C. Davis, Restoration Ecology Marc Horney, Cal Poly San Luis Obispo
68th Society for Range Management Annual Meeting February 5, 2015
Outline Background • Grazing systems – what’s all the fuss? • 2008 synthesis paper methods and results
Meta-analysis • Investigate results and contributing factors o Is there a difference? o What factors are driving observed patterns?
• Meta-analysis methods • Results • Conclusions & recommendations 2
Outline Background
• Grazing systems – what’s all the fuss? • 2008 synthesis paper methods and results
Meta-analysis • Investigate results and contributing factors o Is there a difference? o What factors driving observed patterns?
• Meta-analysis methods • Results • Conclusions & recommendations 3
Rotational Grazing (RG) • many different systems • more ecologically and economically sustainable?
Continuous Grazing (CG)
from Anderson 1988 4
Rotational Grazing (RG)
RG
• many different systems • more ecologically and economically sustainable?
CG Continuous Grazing (CG)
from Anderson 1988 5
Outline Background • Grazing systems – what’s all the fuss?
• 2008 synthesis paper methods and results Meta-analysis • • • •
Goal: investigate results and contributing factors Meta-analysis methods Results Conclusions & recommendations 6
• Synthesis of 47 papers comparing RG to CG • 3 responses 1) Animal Production (kg gain/head) AP / head 2) Animal Production (kg gain/ha) AP / ha 3) Plant Production (kg DM/ha) PP / ha
7
• Synthesis of 47 papers comparing RG to CG • 3 responses 1) Animal Production (kg gain/head) AP / head 2) Animal Production (kg gain/ha) AP / ha 3) Plant Production (kg DM/ha) PP / ha
CG equal to or better than RG most of the time
“These experimental results conclusively demonstrate that rotational grazing is not superior to continuous grazing across numerous rangeland ecosystems.” 8
2008 Method: Vote-Counting • A simple, straight-forward comparison of RG to CG • Tallies the number of “significant” studies as a percent of total studies to determine if there is an effect AP / head: 19/38 studies found no significant difference = 50%
Why do something else? • Non-significant results still hold information o 19 studies with p-values of 0.265 fail to find evidence of an effect o Combination gives a p-value < 0.000001 effect!
• Meta-analysis can tap into non-significant results to: o Is CG > RG, or RG > CG, and when? o Tease apart influential variables across a range of ecosystems and grazing regimes o May reveal previously obscured patterns and important factors 9
Borenstein et al. 2009
Outline Background • Grazing systems – what’s all the fuss? • 2008 synthesis paper methods and results
Meta-analysis • Goal: Investigate results and influential factors • Meta-analysis methods • Results • Conclusions & recommendations 10
Meta-analysis Methods • • • •
Each experiment treated as a sample Removed studies with no quantitative data Removed studies counted 2+ times Data from text, tables, graphs, figures Three response variables: 1) 2) 3)
Animal Production (kg/head) AP / head Animal Production (kg/ha) AP / ha Plant Production (kg DM/ha) PP / ha
• Generated index of effect size = Response Ratio Response Ratio (RR)
RG CG Borenstein 2009, Hedges et al. 1999, Tummers 2006
11
Eleven Predictors Tested Climate Annual Precipitation (mm) Seasonality (Precipitation) Seasonality (Temperature)
Time Years of Treatment Length of RG Rest Periods (days) Length of RG Graze Periods (days) Length of Grazing Season (Year, Season)
Stocking Rate
Experimental Design
Stocking Rate for RG > CG Stocking Rate (low, moderate, high)
Study Size (Hectares, “Scale”) Number of Replicates
12
Statistical Analysis for Three Responses 1) Is there a difference between RG and CG? Test for µ of RG/CG = 1 non-parametric t-test RG > 1 RG higher CG < 1 CG higher
2) What factors influence the response? Generalized linear model for three responses – Regression w/ Gaussian or Gamma (link = “log”) distribution – Final model • Meets assumptions • Best “fit” • Lowest AIC for model selection • Most parsimonious (least complex) 13
R Core Team 2012
Outline Background • Grazing systems – what’s all the fuss? • 2008 synthesis paper methods and results
Meta-analysis • Goal: Investigate results and influential factors • Meta-analysis methods
• Results: 1) AP/Head 2) AP/Ha 3) PP/Ha • Conclusions & recommendations 14
Continuous > Rotational Grazing for Animal Production / Head 0.93
CG
RG
Test that RG = CG p < 0.0001
0.93
CG outperforms RG by ~ 7%
15
Animal Productivity (Head) AP / Head ~ Hectares + Seasonality (Temp) + Grazing Season + Replicates Top 4 Predictors * Inclusion of reps reduces variation to reveal other trends * Decrease variation and better mean estimates * More reps = higher RR
p-value
Effect on Response Ratio
0.03
Increases with size
Seasonality (Temperature)
< 0.001
Increases with seasonality
Grazing Season (Season/Year)
0.07
Increases under yearlong systems
Reps*
0.01
Increases with reps
Size of Study (Hectares)
16
Better performance of CG is more evident in smaller pastures RG may perform better at larger scales, need more replicates Hectares: p = 0.03
17
Better performance of CG is stronger in more constant environments RG may perform better in more variable environments Seasonality: p < 0.0001
18
Better performance of CG is stronger in more constant environments RG may perform better in more variable environments Seasonality: p < 0.0001
19
Continuous > Rotational Grazing for Animal Production / ha CG
0.95
RG
Test that RG = CG p < 0.0001
CG outperforms RG by ~ 5%
20
Animal Productivity (Ha) AP/ha ~ Hectares + Stocking Rate RG>CG + Reps + Days Grazing + Precipitation Predictor
p-value
Effect on Response Ratio
Stocking Rate (RG:CG)
0.01
Equal > Higher
Days Grazing
0.02
Increases with length of RG graze period
Precipitation
0.008
Decreases with more rainfall
Hectares
0.006
Similar to AP/Head
< 0.004
Similar to AP/Head
Reps
21
0.8
0.9
Stocking Rate RG:CG p = 0.01
0.7
RG:CG for Animal Productivity/Ha (kg)
RG performs better relative to CG when the two treatment stocking rates are the equal
Equal
Stocking Rate Equal or Higher for RG:CG
Higher 22
Better performance of CG relative to RG is more evident when RG graze periods are shorter Graze Period: p = 0.02
23
Better performance of CG relative to RG is more evident when RG graze periods are shorter Graze Period: p = 0.02
24
Better performance of CG is stronger in wetter environments RG may perform better in more arid environments Precipitation: p < 0.001
25
Better performance of CG is stronger in wetter environments RG may perform better in more arid environments Precipitation: p < 0.001
26
Better performance of CG is stronger in wetter environments RG may perform better in more arid environments Precipitation: p < 0.001
27
Outline Background • Grazing systems – what’s all the fuss? • 2008 synthesis paper methods and results
Meta-analysis • Goal: Investigate results and influential factors • Meta-analysis methods • Results
• Conclusions & recommendations 28
Summary of Conclusions • In general, not a major deviation from conclusions of the 2008 Synthesis Paper Across all Stocking Rates: Response Variable
Synthesis Paper 2008
Wolf and Horney 2015
AP (kg head-1)
50% CG = RG Evidence for CG > RG 42% CG > RG Estimate < 1 8% RG > CG
• Greater size (ha/scale) • Higher seasonality (temp) • More reps
AP (kg ha-1)
50% CG = RG Evidence for CG > RG 34% CG > RG Estimate < 1 16% RG > CG
• • • • •
PP (kg DM ha-1) 83% CG = RG Evidence for RG > CG 4% CG > RG Estimate > 1 13% RG > CG
Significant Predictors (+ effect)
SR for CG < RG Increased days of grazing Less precipitation Greater size (ha/scale) More reps
• BUT no significant predictors • Not likely a good estimate of plant productivity • Often estimated utilization 29
Suggested Future Direction • Large number of studies at small scales in constant environments – Results of this analysis suggest the possibility of higher productivity for RG at large scales in drier, more variable climates worth investigating!
• Meta-analysis “from the ground up” – Broaden search method • Increase spatiotemporal representation • Seek unpublished data to reduce publication bias – Inclusion criteria stricter, more explicit • Explicit experimental design and methods • Mean, SE, and n for all years weighted estimate – Larger sample size for more even distribution of studies across climates and scales to improve estimates and reduce number of confounding factors 30
Borenstein et al. 2009, Harrison 2011, Hedges et al. 1999
“Continued advocacy for rotational grazing as a superior strategy of grazing on rangelands is founded on perception and anecdotal interpretations, rather than an objective assessment of the vast experimental evidence.” (2008 Synthesis Paper) • Strong evidence for what happens at small scales • Ranching happens at large scales and must consider additional factors – Long-term range quality – Floral and faunal diversity – Economics, Cultural considerations
1) New meta-analysis
2) More research?
Need to know what is happening under real world conditions 31
[email protected] KristinaMWolf.com 32
References • •
• • •
•
• •
Borenstein, M., L. V. Hedges, J. P. T. Higgins, and H.R. Rothstein. 2009. Introduction to meta-analysis. John Wiley & Sons, Ltd. Hoboken, NJ. Briske, D. D., J. D. Derner, J. R. Brown, S. D. Fuhlendorf, W. R. Teague, K. M. Havstad, R. L. Gillen, A. J. Ash, and W. D. Willms. 2008. Rotational grazing on rangelands: reconciliation of perception and experimental evidence. Rangeland Ecology & Management 61:3-17. Harrison, F. 2011. Getting started with meta-analysis. Methods in Ecology and Evolution 2:1-10. Hedges, L. V., J. Gurevitch, and P. S. Curtis. 1999. The meta-analysis of response ratios in experimental ecology. Ecology 80:1150-1156. Quantum GIS Development Team. 2013. Quantum GIS Geographic Information System Version 1.8.0. Open Source Geospatial Foundation Project. Available at http://qgis.osgeo.org. R Core Team. 2012. RStudio: Integrated development environment for R (Version 0.97.312) [Computer software]. Boston, MA. Retrieved Jan 20, 2013. Available from http://www.rstudio.org/. Tummers, B. 2006. DataThief III. Available from http://datathief.org/. WorldClim 2013. Bioclim 2.5 arcminutes resolution generic grids. WorldClim Global Climate Data, Version 1.4 Release 3. Available at http://worldclim.org/. 33
Meta-analysis Studies • • • • • • • • •
• • •
Anderson, D. M. 1988. Seasonal stocking of Tobosa managed under continuous and rotation grazing. Journal of Range Management 41:78-83. Barnes, D. L. and R. P. Denny. 1991. A comparison of continuous and rotational grazing on veld at two stocking rates. Journal of the Grassland Society of Southern Africa 8:168-173. Biondini, M. E., and L. Manske. 1996. Grazing frequency and ecosystem processes in a northern mixed prairie, USA. Ecological Applications 6:239-256. Cassels, D. M., R. L. Gillen, F. T. McCollum, K. W. Tate, and M. E. Hodges. 1995. Effects of grazing management on standing crop dynamics in tallgrass prairie. Journal of Range Management 48:81-84. Derner, J. D. and R. H. Hart. 2007a. Grazing-induced modifications to peak standing crop in northern mixed-grass prairie. Rangeland Ecology & Management 60:270-276. Derner, J. D. and R. H. Hart. 2007b. Livestock and vegetation responses to rotational grazing in shortgrass steppe. Western North American Naturalist 67:359-367. Fisher, C. E., and P. T. Marion. 1951. Continuous and rotation grazing on buffalo and tobosa grassland. Journal of Range Management 4:48-51. Fourie, J. H., D. P. J. Opperman, and B. R. Roberts. 1985. Influence of stocking rate and grazing systems on available grazing in the northern cape. Journal of the Grassland Society of Southern Africa 2:24-26. Fourie, J. H., E. A. N. Engels, and B. R. Roberts. 1986. Herbage intake by cattle on the Tarchonanthus veld in the Northern Cape as affected by stocking rate and grazing system. Journal of the Grassland Society of Southern Africa 3:85-89. Gillen, R. L., F. T. McCollum, K. W. Tate, and M. E. Hodges. 1998. Tallgrass prairie response to grazing system and stocking rate. Journal of Range Management 51:139-146. Hart, R. H., M. J. Samuel, P. S. Test, and M. A. Smith. 1988. Cattle vegetation and economic responses to grazing systems. Journal of Range Management 41:282-286. Heady, H. F. 1961. Continuous vs. specialized grazing systems: a review and application to the California annual type. Journal of Range Management 14:182-193. 34
Meta-analysis Studies •
• • • • • • •
•
• • •
Heitschmidt, R. K., S. L. Dowhower, and J. W. Walker. 1987. Some effects of a rotational grazing treatment on quantity and quality of available forage and amount of ground litter. Journal of Range Management 40:318-321. Hepworth, K. W., P. S. Test, R. H. Hart, J. W. Waggoner, and M. A. Smith. 1991. Grazing systems, stocking rates, and cattle behavior in southeastern Wyoming. Journal of Range Management 44:259-262. Hirschfeld, D. J., D. R. Kirby, J. S. Caton, S. S. Silcox, and K. C. Olson. 1996. Influence of grazing management on intake and composition of cattle diets. Journal of Range Management 49:257-263. Holechek, J. L., T. J. Berry, and M. Vavra. 1987. Grazing system influences on cattle performance on mountain range. Journal of Range Management 40:55-59. Hubbard, W. A. 1951. Rotational grazing studies in western Canada. Journal of Range Management 4:25-29. Jacobo, E. J., A. M. Rodriguez, J. L. Rossi, L. P. Salgado, and V. A. Deregibus. 2000. Rotational stocking and production of Italian ryegrass on Argentinean rangelands. Journal of Range Management 53:483-488. Kothmann, M. M., G. W. Mathis, and W. J. Waldrip. 1971. Cow-calf response to stocking rates and grazing systems on native range. Journal of Range Management 24:100-105. Kreuter, U. P., G. M. Brockett, A. D. Lyle, N. M. Tainton, and D. I. Bransby. 1984. Evaluation of veld potential in east Griqualand using beef cattle under two grazing management systems. Journal of the Grassland Society of Southern Africa 1:5-10. Manley, W. A., R. H. Hart, M. J. Samuel, M. A. Smith, J. W. Waggoner, and J. T. Manley. 1997. Vegetation, cattle, and economic responses to grazing strategies and pressures. Journal of Range Management 50:638646. McCollum, F. T., R. L. Gillen, B. R. Karges, and M. E. Hodges. 1999. Stocker cattle response to grazing management in tallgrass prairie. Journal of Range Management 52:120-126. Owensby, C. E., E. F. Smith, and K. L. Anderson. 1973. Deferred-rotation grazing with steers in KansasFlint-Hills. Journal of Range Management 26:393-395. Wood, M. K. and W. H. Blackburn. 1984. Vegetation and soil responses to cattle grazing systems in the Texas Rolling Plains. Journal of Range Management 37:303-308. 35
Geographic Scope 24 randomly selected papers from Briske et al. 2008
QGIS 2013, WorldClim 2013
36
Temporal Scope
37
Stocking Rate does not significantly impact the ratio of RG:CG
Isn’t Stocking Rate Important? Response Ratios of RG:CG for Hypothetical Animal Production over Three Stocking Rates
weight gain (kgs head-1)
85
80 RG CG
75
70
Response ratio of RG:CG weight gains
Hypothetical Animal Production over Three Stocking Rates
1.1
1.0
0.9
Response Ratio
0.8
65
Stocking Rate
Stocking Rate
Illustration of stocking rate as significant and non-significant factor for production and response ratios. Hypothetical illustration to show that stocking rate may impact animal production, but may not significantly alter response ratios for animal production. 39 RR = ratio of weight gain for RG systems to CG systems; CG = continuous grazing, RG = rotational grazing
Studies counted 2 + times • One study spanned 25 years, but was reported as 4 separate studies totaling 48 years o Derner & Hart 2007a o reported values for PP only
o Hart et al. 1988 & Manley et al. 1997 o stated no differences between CG and RG for AP and PP, but no values reported
o Hepworth et al. 1991 o reported on animal behavior only
Studies counted 2 + times • One study spanned 25 years, but was reported as 4 separate studies totaling 48 years o Derner & Hart 2007a • Gillen et al. 1998 and Cassels et al. 1999 – both o reported values for PP only reported on PP for 1989-1993 o Hart et al. 1988 & Manley et al. 1997 • Hubbard 1951 reported on AP for 1949-1950, and o stated no differences between CG and RG for AP and PP, but no Smoliak 1960 reported on AP for 1949-1957 values reported o Therefore, o Hepworth et al. 19913 experiments were counted 8 times o reported on animal behavior only
*Some* Assumptions & Operational Definitions Stocking Rate •
Assumed SR categories of “low”, “intermediate”, and “high” were appropriate to the site.
•
For studies with > 1 SR level – but that did not use “low” and “high”, used the lowest SR as “low” and the highest SR as “high” (e.g., if used intermediate and very high SRs, assigned intermediate = “low” and very high = “high”.
•
For studies with treatments at only one SR, these data were included only in one-sample t-tests and/or regression analyses.
•
If multiple SRs were used, but only the mean response across all SRs was reported, the SR was categorized as “intermediate” unless authors stated it was of a different average SR category
•
If SR category was not assigned by the author(s), research was conducted of the literature and/or internet to determine the SR category based on grazing research and established methods in similar biomes/areas.
•
If no SR information was provided (category or densities), it was assumed to be an “intermediate” SR and treated as such.
•
For studies in which the RG SR was greater than the CG SR, the two grazing systems were compared at the higher SR level .
Animal Productivity •
If multiple weights were taken over a year, the final weight at end of each grazing or growing season were used. 42
Continued assumptions and definitions Plant Productivity • • • • •
If multiple plant measurements were taken over a year, these were averaged. If plant materials were separated out by condition (live, dead, previous year’s growth, etc.) these were totaled. If condition of plant material was not reported, assumed the reported values were for “total”. If multiple range “types” were included in a study (e.g., lowland vs. upland), productivity values for each were added and averaged across the number of types within each treatment. In general, “weeds” were omitted from productivity means in all studies; if plant types not reported, it was assumed that “weeds” were not included in measurements.
Other • • • • • •
Assumed weather (esp. rainfall) and topography were sufficiently similar across treatments within a study to not have significantly different effects on the dependent variables. Means were averaged across all years for each study; if the average for all years was not provided, data from the final year of treatment were used. If data from all years were not reported (or data was not collected for a year(s) within a study), only those years for which data were reported were used to calculate the mean response(s). If studies were duplicated (same site, treatments, etc.), study with the most complete dataset was used; if no values were reported from duplicate studies, that paper was not used in the meta-analysis. If 2+ RG categories were compared to one CG treatment, these were treated as two separate samples with individual response ratios (RRs) with case ID’s a,b,…,n. If the study was included in the Briske et al. (2008) paper but did not provide sufficient data (or any data) to allow for calculation of RRs, it was dropped from this analysis. If Briske et al. did not evaluate the response for a particular RG system, but data were provided in the study, it was included in this analysis.
43
Continuous > Rotational Grazing for Animal Production / ha CG
0.95
RG
Hypothesis test that RG = CG p < 0.0001 Test that RG = CG CG > RG p < 0.0001
0.92
CG > RG
44
2008 Synthesis Results AP / Head
AP / Ha
PP / Ha
No difference
50%
50%
83%
RG > CG
8%
16%
13%
CG > RG
42%
34%
4%
How you group data may alter the interpretation
Plant Productivity / Ha • 87% equal or greater PP/ha for CG
45
2008 Synthesis Results AP / Head
AP / Ha
PP / Ha
No difference
50%
50%
RG > CG
8%
16%
83% 13%
CG > RG
42%
34%
4%
How you group data may alter the interpretation
Plant Productivity / Ha • 87% equal or greater for CG • 96% equal or greater for RG • The “equal” (non-significant) result holds power and information not tapped by vote-counting
46
Better performance of CG is more evident with smaller plots RG may perform better at larger scales, need more replicates Hectares: p = 0.006
47
Rotational > Continuous Grazing for Plant Production / ha CG
RG
Test that RG = CG p < 0.0001
RG > CG 1.02
48
Rotational > Continuous Grazing for Plant Production / ha CG
RG
Test that RG = CG p < 0.0001
RG > CG 1.02 • No significant predictors for regression “Productivity” in response to the treatments rarely actually measured o Most often measured residual DM = utilization o Clipped at end of season o Permanent exclosures clipped; never impacted by the grazing treatment • Cattle gain more (per head and ha) under CG less plant biomass left behind • Doesn’t tell us much about treatment effects on plant productivity
49