Target species: Q. douglasii, Q. kelloggii, Q. wislizeni, Q. chrysolepis

Other species: none

Life stages: seedlings, saplings, trees

Origin: natural

Situation: range

Locations: Madera, Tulare, Fresno, and Kern counties

Overview: The authors of this study used logistic regression to correlate various site and stand factors with the presence or absence of oak seedlings or saplings within plots. Plots were located along elevational transects. The particulars of the sampling scheme are not presented so possible sampling biases cannot be ruled out. Although four oak species were included in the study, Q. chrysolepis and Q. wislizeni were lumped together. Site management history and environmental variables were not adequately represented in the regressions. Plots that were rated as grazed were less than half as likely as nongrazed plots to have Q. douglasii seedlings. For most oak species, seedling and sapling presence were correlated with levels of canopy cover of the same species. The relevance of other reported significant predictor variables is questionable.

Methods: The authors established 10 elevational transects that extended from the lower limit of Q. douglasii savannah (180 - 485 m) to the oak woodland - mixed conifer forest transition zone (920 - 1525 m). The basis for selecting these elevational transects and the nature of the transects (e.g., total length or orientation) are not described. Presumably, the transects were not randomly distributed, so any biases associated with transect selection would be reflected in the data and the transects may not be representative of the overall study area. Actual survey plots (192 total) were established 60 m north or south of the elevational transects. Although the number of plots per transect is not mentioned, we may infer from the total that the number varied. The authors do not describe how study plot location was randomized along the transects. Plots were rejected if they did not occur in "oak woodland vegetation type", although the definition they used for this type is not given. Survey data were collected on linear strip transects 30.5 m by 3.7 m (0.011 ha in area) laid out in randomly chosen directions.

Oak trees less than 0.3 m in height were classed as seedlings, and those between 0.3 and 3 m in height were classed as saplings. Each of these size classes could include age ranges that span several decades. For a given species, the plot was considered "stocked" with seedlings or saplings if one individual in that size class was observed in the plot (i.e., the binary outcome variable STOCK=1). The authors concede that this variable is a "somewhat crude" measure of regeneration, which may be an understatement. The authors used logistic regression to determine which of the environmental factors measured (shrub cover, canopy cover, solar radiation, elevation in meters, grazing (plus or minus, apparently based on site observations at the time of survey), number of woody species on site, forage residual dry matter [RDM]) were related to sapling or seedling presence for each species. Data were collected over three years, July and August of 1987, 1988, 1989. Because small Q. douglasii seedlings may defoliate or die back by midsummer, seedling presence may be underestimated in surveys made this late in the growing season. Data for Q. chrysolepis and Q. wislizeni were combined, although no reason for this was given.

Logistic regression model limitations. The authors do not note how many plots were used in each logistic regression model. It would have been invalid to use all 192 plots for all models because, for instance, plots beyond the altitude range of a given species do not provide meaningful information for that species. Because data were collected in 3 separate years, herbaceous RDM data are not meaningful. Year to year differences in RDM would be confounded with differences in RDM between plots (note: RDM was not significant in any regression model). The grazing variable (presence or absence of grazing in the year of the survey) is a very poor measure of either historical or current grazing impacts. The regression includes no predictor variables that meaningfully describe cultural practices (grazing, fire, cutting) during the period of seedling/sapling establishment. Soil and topographic predictor variables are conspicuously absent from the study. Elevation and solar radiation are the only environmental predictor variables used, and these do not fully incorporate effects of rainfall, temperature, and potential evapotranspiration. Correlations exist between some predictor variables (e.g., for Q. douglasii, the authors note that elevation and the number of woody species are correlated). Such correlations can lead to difficulties in interpreting the logistic regression coefficients which are used to indicate the strength and direction (positive or negative) of effects. Finally, the "percent right predictions" row in table 2 is meaningless as presented. Logistic regression provides an estimate of the probability that the outcome (STOCK) is observed; it does not predict which plots will or won't have a specific outcome.

Findings / assertions

Presence of seedlings and saplings. The percentages of plots that had at least one seedling or sapling of oak species present in the plot overstory were as follows:

The data illustrate that seedlings are more common than saplings for all species. However, the data do not provide estimates of seedling or sapling densities or the status of regeneration either within the plots or in the study area as a whole. For Q. douglasii, percentages are based on 131 plots; the remaining species were reportedly present on "about one-third" of the 192 plots (i.e. about 60 plots).

Effects of grazing. Even though the grazing and outcome variables were crude at best, the authors demonstrated a significant negative impact of grazing on seedling presence for Q. douglasii. Given that the variables used were not especially sensitive, we can conclude that the negative effect of grazing was very strong. Plots shown in Figures 1 and 2 suggest that currently nongrazed plots were more than twice as likely to have one or more seedlings than currently grazed plots.

Q. douglasii sapling presence was positively correlated with the number of woody species per plot. Although the authors relate this variable to rainfall and soil conditions, low diversity of woody species can also reflect chronic impacts of livestock browsing (Swiecki, Bernhardt and Drake 1993). Hence, the data are consistent with the hypothesis that grazing adversely impacts Q. douglasii sapling presence.

The lack of significant grazing effects for the other species could be solely related to the poor sensitivity of the variables and/or lower numbers of data points for these species. Alternatively, the data may reflect real effects, such as the fact that Q. chrysolepis and Q. wislizeni are less preferred by livestock than Q. douglasii.

Stand factors. For all species, seedling presence was positively correlated with the level of tree canopy of the same species, which is logical. Tree cover was also correlated with sapling presence for Q. chrysolepis, Q. wislizeni, and Q. kelloggii, which are more tolerant of understory conditions than is Q. douglasii.

Site factors. Coefficients for site variables must be interpreted with caution due to the likelihood that many of these are also correlated with stand factors. High correlations between different variables in the regression can reverse the apparent direction of the factor. Hence, the apparently positive correlation between live oak seedlings and high solar radiation could be spurious, but insufficient data are presented to test this possibility. Likewise, although Q. douglasii seedling presence and Q. kelloggii sapling presence were positively and negatively correlated with elevation, respectively, the reported direction of these effects could also be spurious.