Study area

The study area encompassed the known nesting range of Marbled Murrelets in California and southern Oregon (based on Meyer 1999 and Meyer et al. 2002). This area extended from Coos Bay, Oregon, south to Monterey Bay, California, and inland up to 40 km from the coast. Because 98% of landscapes occupied by murrelets have been within fog-influenced vegetation zones (Meyer et al. 2002), we studied only those zones (Fig. 1). They included the redwood (Sequoia sempervirens) zone in California and the western hemlock (Tsuga heterophylla) and Sitka spruce (Picea sitchensis) zones in Oregon (Franklin and Dyrness 1973). Douglas-fir (Pseudotsuga menziesii) is the dominant tree species in the western hemlock zone and is the species most likely to be used as a nest tree, whereas, in California, redwood trees are used most often (Hamer and Nelson 1995). Based on Meyer et al. (2002), the criteria used to delineate the nesting range were further defined by elevation and distance to key marine features (see footnote in Table 1). Within this nesting range, average maximum temperature and annual precipitation were 25° C and 203 cm in California and 27° C and 297 cm in southern Oregon, respectively.

Sampling design and murrelet indices of use

Our study is a retrospective analysis of available survey data, where the sampling design varied from area to area. Surveyors completed 17,145 surveys at 9,326 stations from 15 April to 15 August 1991–1997 (Fig. 1). Forty-six percent of the survey stations were randomly or systematically (in a grid) placed within potential murrelet nesting habitat that contained large platform-like branches; except for a few mature stands surveyed in Oregon, these were mostly found in stands of old-growth forest (hereafter referred to as stands). The remaining survey stations were placed in such habitat near proposed timber sales. The 2-hr surveys began 45 min before dawn according to a standardized protocol for intensive murrelet surveys (Ralph et al. 1994).

Because murrelets rarely used stands outside the fog-influenced vegetation zone, only the 7,616 surveys at 4,158 stations that were within that zone were included in our analyses. From that set of stations, 349 were selected as “central stations” (Fig. 1), each ≥800 m apart (n = 133 in Oregon and 216 in California). A central station was defined as the one nearest the center of the cluster of stations located within a small (<50 ha) stand or within a 50-ha section of a large stand (Fig. 1).

To determine occupancy, we classified a central station as either occupied or unoccupied based on murrelet behaviors indicative of nesting. A central station was considered occupied if any of the following behaviors were observed at any survey station within a 400-m radius (50 ha) of the central station: a bird flew below the canopy, circled above the canopy, landed in the canopy, was stationary in a tree, or broken eggshells were found (Ralph et al. 1994). Central stations with no murrelets observed or heard at any station within a 400-m radius were considered “unoccupied.” We deleted from the analysis any central stations where birds were seen or heard at a station within the 400-m radius, but occupancy status could not be verified (i.e., stations had birds “present” but no occupying behaviors were observed). To ensure 91% confidence that the remaining unoccupied areas with no birds seen or heard did not have murrelets “present” in the surveyed year, at least four surveys must have been conducted in 1 year within the 400-m radius area (based on an analysis in Evans Mack et al. 2003, Appendix A, p. 40) before the central station was assigned “unoccupied” status (otherwise it was deleted from the analysis). This confidence rate does not consider that the status of a site may change between years. Most of our stands were not surveyed in more than 1 year because the survey protocol initially did not require it. During the later years of this study, there was a change in the recommended number of surveys and years. To retain our ample data collected during the early years of our study, we used the early survey protocol and accepted that some of the unexplained variance in our models would be due to some unoccupied central stations actually being occupied in years they were not surveyed. Henceforth, the use of the term “station” in this paper will refer to the central station and its associated stand characteristics within the 400-m radius circle surrounding the central station (Fig. 1).

Our index of abundance was the total number of standardized bird detections observed during each survey averaged over all surveys conducted within 400 m of the central station. Each visual or auditory observation of a single bird or group of birds was considered one “detection.” In each state, the total number of detections was standardized by applying a multiplier to adjust for seasonal variations over time. On average, detections peak in July in both California (O'Donnell et al. 1995) and Oregon (Jodice and Collopy 2000). For the 12 consecutive 10-day periods starting with 15 April, the multipliers were 1.14, 1.81, 1.22, 1.36, 1.12, 1.40, 1.29, 0.93, 0.81, 0.59, 0.61, and 1.14. These multipliers are the ratio of the mean number of detections for the entire survey season divided by the mean for each 10-day period obtained from pooled surveys at three long-term (1989–1995) monitoring sites in northern California (Miller and Ralph 1995).

On average, six surveys contributed to the final mean number of murrelet detections assigned to a central station. Because differences between observers, daily survey conditions, or daily murrelet activity can cause murrelet survey results to vary, even when averaged over six surveys (Jodice et al. 2001), we assigned means to four levels of detections: zero, low (<2), medium (2–15), and high (>15), and used the categories, rather than the actual means, in our analyses. Such category boundaries are useful for showing differences in landscape variables (Meyer et al. 2002).

We assumed that the abundance metric was positively correlated to the number of nesting birds in a stand or section. To what extent this is true is unclear. The number of birds detected may be biased by the size of the forest canopy opening (Rodway and Regehr 2000), skill of the observer, weather, the flight path of the birds from the ocean to the nest site, or by the high variability in number of birds detected during each survey at a station (Jodice and Collopy 2000, Buckland et al. 2001, Jodice et al. 2001). Although the occupancy metric also can be biased by these factors, such problems probably would be associated with the abundance metric more than with the more robust occupancy metric (Rodway and Regehr 2000). Thus, we compared the interpretation of the results of the two methods to assess whether such biases might have occurred.

Forest stand variables

We recorded elevation, slope, aspect (using 90-m digital elevation models), and distance to nearest stream or major road (identified with 1:100,000 digital line graphs) from each central station. To create linear variables from a circular distribution, aspect was divided into an east-west component using the sine of the aspect and a north-south component using the cosine of the aspect (in radians). Sine and cosine values range from 1 to −1, where positive values represent eastern and northern aspects and negative values represent western and southern aspects (Briggler and Prather 2003).

The annual precipitation and maximum temperature for the summer, averaged over the survey period (1991–1997), were measured at each station using PRISM (Parameter-Elevation Regressions on Independent Slopes Model) at a 4-ha resolution (Spatial Climate Analysis Service, Oregon State University,  http://www.ocs.orst.edu/prism). This analytical model interpolates station data for monthly and annual climate, accounting for orographic, coastal, and hillslope exposure effects, and distributes such data onto a regular grid across the landscape (Daly et al. 2002).

The estimated mean dbh of canopy conifers in each stand was obtained for each station in northern California only (Point Reyes National Seashore to northern border) from a GIS database in ARC/INFO developed by the California Timberland Task Force (1993). Estimates of mean tree dbh for each stand on a 4-ha resolution GIS map were derived using remote sensing techniques (spectral signatures compared to ground observations) from LANDSAT Thematic Mapper (TM) imagery. Unfortunately, the expected accuracy of the mean dbh for each stand in our study was not available. However, this remote satellite approach was effective in identifying the size class >91 cm dbh (82.8% accuracy; California Timberland Task Force 1993), and the dbh datasets appeared reasonable given our knowledge of some areas. The quadratic mean diameter at breast height (QMD) was used to estimate the mean stand dbh. QMD is a measure applicable to the Marbled Murrelet because it emphasizes the dbh of large trees. The formula weights dbh by the percent tree cover (cover) contributed by that size tree:

Statistical analyses

We used either non-parametric Kruskal-Wallis tests or t-tests with unequal variances (α = 0.05; Sokal and Rohlf 1981) to detect differences in mean ranks or means of forest stand variables (1) between occupied and unoccupied stations, and (2) among the abundance categories of occupied stations. Both tests identified almost the same set of variables as significant (one exception, shown in Table 1), even though many variables had highly skewed distributions (thus, the Kruskal-Wallis test was used for most comparisons). To evaluate whether abundance provided useful information beyond that provided by occupancy, stations with zero abundance (unoccupied stations already used in the occupancy analysis) were not included in the comparison of abundance categories.

We used multiple logistic regression to predict murrelet use and assess the effect of habitat variables in combination. Binary logistic regression was used to relate occupancy to habitat variables. Ordinal logistic regression was used to relate murrelet abundance (zero to high categories) to the variables. Ordinal logistic regression creates three regressions with parallel slopes but different intercepts, where each category is contrasted with the highest abundance category. For the ordinal logistic regressions, we added the zero category to the other three abundance categories to include the full range of abundance levels.

For each regression analysis, we developed 15 candidate a priori model combinations that were the most biologically meaningful. We calculated Akaike's Information Criterion (AIC_c, corrected for sample size) for each model in SAS (SAS Institute, Inc. 1990). The models were ranked by ΔAIC_c, and the models with ΔAIC_c < 2 (which had the highest Akaike weights) were selected as the best models (Burnham and Anderson 2002). Multicollinearity was not a problem (Neter et al. 1989) because no two variables were highly correlated (all r < 0.6); models also met the assumption of linearity (Neter et al. 1989).

The predictability of the logistic regressions was assessed by comparing Somers' d among the models. To calculate Somers' d, all possible pairs of stations are compared, where each in the pair is in a different category. For example, in the occupancy models, the proportion of such pairs in which the occupied station has a higher predicted probability of occupancy than the unoccupied station is recorded as the proportion of concordant pairs. This proportion is adjusted to range from −1 to 1. Zero is no correlation and 1 or −1 is perfect positive or negative correlation between observed and predicted occupancy (Harrell 2001).

Tree size

In California, dbh was greater in occupied stands than in unoccupied stands (dbh data were unavailable for Oregon; Table 1). Only 54% of stations in the fog zone with mean dbh between 100 and 140 cm were occupied, whereas almost all (91%) stations with larger tree sizes were occupied (Fig. 2A). The majority (79%) of stations in the fog zone with mean tree dbh below 100 cm were not occupied. Mean dbh was much larger inside than outside of the fog zone (where occupied behaviors suggestive of nesting generally do not occur; Fig. 2B). At the occupied stations, dbh did not significantly differ across murrelet abundance categories (Table 1).

Elevation

Occupied stations averaged 188 and 168 m lower in elevation than unoccupied stations in California and southern Oregon, respectively (Table 1). Elevation was lower at high-abundance stations in California (P < 0.001) but not in Oregon (P = 0.79; Table 1). The difference in elevation between stations with zero and high abundance (>15 detections per survey) was significant in both states (Kruskal-Wallis, P < 0.001).

Location in drainage

Slope, aspect, and distance to nearest stream provide insight into locations within a drainage potentially used for nesting. In southern Oregon, location in the drainage did not differ between occupied and unoccupied stations, except that occupied stations were more west-facing than unoccupied stations (Table 1). In contrast, slope in California differed between occupied and unoccupied stands and across the three abundance categories; unoccupied stations had slopes (12°) that were three times steeper than the slopes of high-abundance stations (4°). Although occupied stations were significantly closer to streams (291 m) than unoccupied stations (359 m) in California, there was no difference among abundance categories in distance to stream (Table 1). In contrast to Oregon, aspect in California did not differ between occupied or abundance categories.

Distance to roads

California stations with high bird abundance were significantly closer to roads than were stations with medium or low abundance, but distance to roads did not differ between occupancy categories. Road distance was not significantly related to occupancy or abundance in Oregon (Table 1).

Climate

In California, occupied and high-abundance stations had cooler maximum summer temperatures than unoccupied stations (Table 1). Temperatures were 2° C cooler at occupied stations, and 3° C cooler at high-abundance stations. In southern Oregon, occupied stations were significantly cooler than unoccupied stations but temperatures did not differ among abundance categories. Precipitation results were less clear. Precipitation was not significantly related to either occupancy or abundance in southern Oregon or to abundance in California. However, occupied stations in California were significantly drier than unoccupied stations.

Variables in combination

The best binary logistic regression model in California (predicting occupancy) included tree size, maximum temperature, elevation, and precipitation (first model; Table 2). Based on Akaike weights, this model had about twice the support for being the best model compared with the two next-best combinations of variables (evidence ratios of the weights between two compared models ranged from 1.8 to 2.1); these models had the same variables but also included slope (second model) or distance to road (third model). All three models (ΔAIC_c < 2; Table 2) indicate murrelets occupied stands on low-elevation, cool, moist, gentle slopes with large trees in areas farther from roads.

The best ordinal logistic regression (abundance) in California also included dbh, temperature, elevation, precipitation, and slope (first model; Table 3). Based on the evidence ratio, this model had 2.7× the support of the next-best model. Similar to the occupancy models, these models indicate that murrelets were most abundant in low elevation, cool, moist stands with gentle slopes and large trees. The predictability (Somers' d) of the occupancy models was higher than that of the abundance models (Tables 2 and 3).

The best binary logistic regression model for southern Oregon included elevation, east-west aspect, precipitation, and distance to stream, indicating that murrelets occupied sites that were often on low-elevation, high-precipitation, west-facing slopes far from streams (first model; Table 2). The next three best models of occupancy (ΔAIC_c < 2) had about one-third less support for being the best model (Table 2).

The first four ordinal logistic regression models (abundance) for Oregon included elevation, east-west aspect, distance to stream, or precipitation in various combinations; Akaike weight did not strongly separate these models (Table 3). The fifth-best model, which included distance to road, had a lower Akaike weight but still had ΔAIC_c < 2; thus, it is also a comparatively good model. Overall, the Oregon models had lower predictability than the California models (Somers' d; Tables 2 and 3).

Tree size

As expected based on previous research, murrelets in California most consistently used stands with the largest trees. As tree size decreased, stands were often unoccupied, especially those with mean dbh ≤100 cm. Other studies in Oregon, Washington, and Alaska have shown similar trends (Rodway et al. 1993, Grenier and Nelson 1995, Hamer 1995, Kuletz et al. 1995). Larger trees often have more potential nest platforms than smaller trees (Hamer 1995, Naslund et al.1995), although, as Hamer (1995) postulated, epiphytes on some smaller trees can effectively thicken branches, thus providing platforms for nest sites.

Slope position

As predicted, murrelets in California most often used cool, moist, forests on gentle, low-elevation slopes near the bottom of drainages where the largest trees grow. Similarly, birds in southern Oregon used cooler and moister low-elevation areas, but not as low as in California. Unlike California murrelets, the Oregon birds tended to use west-facing slopes farther from streams, and the steepness of the slope was not important. The difference in slope position and distance from stream may be due to differences in dominant tree species used for nests between the two states. Unlike redwood trees, which grow largest along alluvial flats in California (Agee 1993), coastal Douglas-fir in Oregon is not adapted to flooding and does not grow well adjacent to streams (Stone and Vasey 1968). Thus, in Oregon, the best nest platforms may be in the largest Douglas-fir trees, which are farther from streams and higher on the hillside.

Data from other studies support our finding that forests with mostly flood-intolerant tree species may have more murrelet occupancy higher on the hillside—in contrast to forests with flood-tolerant tree species that may have the best nesting sites at lower elevation, on gentler slopes, and in flood-prone areas. In central Oregon just north of our study area, where the more flood-intolerant Douglas-fir was the dominant tree used by murrelets, occupied stands were mostly on the gentler slopes of middle and upper portions of hillsides (Grenier and Nelson 1995). In a Washington study that included areas dominated by flood-intolerant Douglas-fir trees, as well as other tree species used by murrelets that were more tolerant of flooding [western red cedar (Thuja plicata) and Sitka spruce; Duddles and Fitzgerald 1998], the probability of murrelet occupancy of a stand was highest on lower to middle portions of the hillside, and increased as slope increased (Hamer 1995). A study of 45 actual nest sites across the Pacific Northwest showed that, on average, nests were located on the lowest slopes in California, on the highest in Oregon, and on intermediate slope positions in Washington (Hamer and Nelson 1995), corresponding to the proportion of the dominant tree species that are flood-tolerant. Although there may have been a bias because areas were not randomly searched to locate nests, the trend in that nest study is suggestive.

One study does not support our finding that areas with flood-tolerant tree species have more murrelet use near the bottom of drainages. In a recent study of 157 actual nest sites in British Columbia forests near Desolation and Clayoquot sounds, murrelets avoided flat areas (F. Huettmann unpubl. data) even though the forests had many flood-tolerant species. It is difficult to compare our study area with British Columbia because slopes in our area are not as steep and the large avalanche chutes that are commonly found in British Columbia do not occur in Oregon and California. More research is needed to evaluate the relationships between tree size and number of potential nesting platforms in relation to position on slope and tree tolerance to flooding.

Elevation

The apparent preference of Marbled Murrelets for nesting in relatively low-elevation sites (not necessarily at the bottom of valleys) appears to be fairly consistent across its range from Alaska to California (Rodway et al. 1993, Burger 1995, Grenier and Nelson 1995, Hamer 1995, Kuletz et al. 1995, Miller and Ralph 1995, Manley 1999). One might argue that lower-elevation sites tend to be closer to the coast, and, thus, increased use could be due to closer proximity to marine habitat. However, studies in California and southern Oregon have shown that low elevation is important even when proximity to marine habitat has been taken into account (Meyer and Miller 2002, Meyer et al. 2002).

For studies in which low elevation was found to be important to murrelet use, only Rodway et al. (1993), Miller and Ralph (1995), and this study clearly demonstrate that the elevation effect is related to slope position, with murrelets more often occupying stands at the bottom of local major drainages rather than occupying ridge tops. On a more regional scale, murrelets also may be responding to less favorable vegetation conditions that occur as elevation increases (Burger 1995, Hamer 1995). For example, in Washington and British Columbia, high-elevation trees such as silver fir (Abies amabilis) and mountain hemlock (Tsuga mertensiana) may be smaller and have fewer platform branches than trees that occur at low elevations—such as western hemlock, western red cedar, Douglas-fir, and Sitka spruce (Rodway et al. 1993, Hamer 1995). However, murrelets readily nested in high-elevation (>800 m) forests composed of mountain hemlock and yellow cedar (Chamaecyparis nootkatensis) in British Columbia (Manley 1999, Bradley 2002), even though they marginally preferred forests at elevations of <800 m (F. Huettmann unpubl. data). Although they may prefer lower elevations, murrelets readily use high-elevation forests.

Murrelets probably are responding to factors correlated to elevation, rather than to elevation itself. High elevations along ridge tops may be used less because branch growth needed to develop nest platforms may be less optimal (Daniel 1942), or wind damage may be higher. Also, fog—which provides cooler conditions for nesting birds and obscures a predator's view of nests—burns off more quickly on ridge tops.

Climate

As predicted, murrelets used relatively cool forest stands with high amounts of precipitation (mostly rain). Although we hypothesized that such areas would be used by a coldwater-adapted seabird (Ralph et al. 1995), more use could also result if cool, moist environments promote more epiphyte growth on branches (i.e., suitable nest sites). In California, birds often occupied drier areas (Table 1), but once tree size was taken into account, the areas of greater precipitation had greater occupancy. In regard to temperature, other studies have shown that murrelets avoid warm areas, even where suitable nest sites are available (Dillingham et al. 1995, Hunter et al. 1998). Our study is the first to show that, even within the fog zone, murrelets appear to prefer moist areas with the coolest temperatures. Laboratory studies on the physiological tolerances of murrelets are needed to help validate this apparent preference.

Distance to roads

When elevation was taken into account (Tables 2 and 3), regression analyses showed that sites used by murrelets were often farther from roads than sites without murrelets. We obtained this result even though birds might be detected more easily at survey stations near roads. In a previous study, wherein the level of fragmentation in old-growth forest was taken into account, murrelets were most abundant in sites farther from roads (Meyer et al. 2002). These results suggest that human disturbance and noise along roads may reduce murrelet use of an area.

Potential biases in indices based on audio-visual surveys

We used occupancy as an index to murrelet nesting, and abundance as an index to number of nesting murrelets in a stand. These metrics may be poor proxies for nesting if they are biased by survey station placement. Rodway and Regehr (2000) showed that audio-visual survey sites near streams often have larger canopy openings than sites farther from streams, making it possible to detect more birds and more below-canopy flight or circling, two major behaviors used to classify a site as occupied. Consequently, detections may be upwardly biased in low-elevation valley bottoms with gentle slopes, even when the birds have no preference for using such areas. Moreover, murrelet abundance may be inflated if the birds use streams as travel corridors (Rodway and Regehr 2000). We detected higher occupancy and abundance in valley bottoms in California, possibly due to this bias; however, results in Oregon did not show the same trend. The lower predictability of our abundance models suggests that such biases and the high daily variability in number of detections made it more difficult to detect a significant relationship between abundance and habitat variables than between occupancy and the same variables. More actual nest sites need to be identified in our study area using unbiased techniques, so that our results for occupancy and abundance, both of which are affected by detectability, can be compared with results of other studies.

In summary, our results suggest that recovery efforts for murrelets in our study area should focus on protecting cool, moist, low-elevation stands of old-growth forest with the largest-dbh trees, and that these areas should be far from roads. In California, redwood stands along alluvial flats adjacent to streams should be given high priority, and, in southern Oregon, the low-elevation Douglas-fir stands higher on hillsides should be given priority. This information, along with landscape and regional-level results from other studies, can help managers prioritize recovery efforts. More research is needed to establish the link between occupancy, abundance, and nest density to verify our conclusions.