Philip C. Stouffer Jason A. Zoller. LSU School of Renewable Natural Resources Final Report 30 June 2006

Transcription

1 Use of the Maurepas Swamp by Migrating Birds Determined by Radar Detection Objectives Philip C. Stouffer Jason A. Zoller LSU School of Renewable Natural Resources Final Report 3 June 26 The objective of this part of the project is to use NEXRAD radar and GIS landcover maps to understand how migrating birds use the wetlands surrounding Lake Maurepas. Our focus is on the millions of birds that pass through southeastern Louisiana in spring and fall as they move between their breeding grounds in North America and wintering grounds in the Caribbean, Central America, and South America. We document bird abundance and location with NEXRAD radar images from just after dusk. These images show nocturnal migrants as they resume migration (at >4 m) after spending the day resting and foraging. We match the locations where we detect birds in the air with landcover below them to examine how bird abundance varies with variation in wetland vegetation. We were most interested in migrant use of the three main classes of swamp forest that have been identified from previous work in the Maurepas. In our earlier surveys of birds in the swamp (Zoller 24), we found comparable abundance and diversity of breeding Neotropical migrants in Intermediate and Sustainable swamp forest, but Marsh supported almost no breeding migrants. The area of our sample includes little bottomland hardwood forest, a habitat type known to be important for migrating birds. To compare the relative use of baldcypress-tupelogum swamp with bottomland hardwood forest, we also evaluated abundance of migrants in a representative area of the Pearl River basin along the Louisiana-Mississippi border. Methods Radar identification of migrants We used radar to sample birds because it allowed us to monitor a much larger area than we could by working on the ground. The nocturnal migration of most songbirds makes them particularly amenable to sampling with weather radar (Gauthreaux and Belser 23, Larkin 24). At dusk, birds that will resume migration fly up from where they spent the preceding day. As they enter the radar beam, they are detected, and their density is reflected by the amplitude of the signal returning to the radar (just like rain on the familiar radar images used to describe weather). We archived radar images from the Slidell airport NEXRAD (also called New Orleans or KLIX in some presentations) daily in the spring and fall of 24 from Weathertap ( We archived both the base reflectivity (the

2 typical radar map displayed to describe weather) and the radial velocity, which measures the speed of signals. We archived several hours of images each night. The long series of images was important for distinguishing birds from other signals; for each night, we only saved the data if there was unambiguous evidence for moderate to heavy bird migration. Usually these were nights with tailwinds. For each night that we analyzed, we used a single image from within 3 minutes after sunset. Selecting this time represents a compromise; more birds would be in the air slightly later, but the longer they are in the air, the farther they displace from where they spent the day. Using the image just after dusk allows us to assume that birds are detected directly above where they spent the day. We only choose images for which the outline of the lake was still clearly visible (Figure 1). Figure 1. Sample radar reflectivity image from spring 24 on a night of heavy bird migration. Hot colors indicate higher dbz. Notice that the effect of Lake Maurepas remains visible; birds have not yet moved far enough from their stopover locations to fill the airspace over the lake. Several features allowed us to distinguish birds from rain, insects, or other signals. Rain bands or thunderstorms have a characteristic shape to their radar image. In general, rain signals can appear constant even when they are over 1km from the radar, while bird signals fade out before this distance even when they are very strong at 4-6 km from the radar. This occurs because the upward projection of the radar beam, combined with the curvature of the earth, samples an ever increasing altitude with distance from its source. Bird abundance drops considerably above about 1m, so the beam passes over birds by about 8km from the radar. The characteristic sudden appearance of signals right at dusk also easily allowed us to avoid misinterpreting rain. Insects can also enter the atmosphere at dusk and be detected by radar. Our main technique for avoiding misinterpreting insects as birds was to evaluate the speed of the signals compared to wind speed. We determined signal speed from the radial velocity data, and wind speed from balloon soundings from the Slidell airport (available at We only used images with signals moving > 1 knots faster than wind speed (Larkin 1991). Our dataset included 13 nights in spring 24 between 9 April and 19 May and 12 nights in fall 24 between 14 August and 15 October. We combined these images to form single composite image for each season, with each 1 km 2 pixel representing the mean dbz scores of all the nights in the sample.

3 Landcover mapping We added the composite radar image for each season to a landcover GIS assembled by Gary Shaffer and Jason Zoller. That image classified the area of interest around Lake Maurepas into six cover categories that are explained in more detail in their report: marsh, intermediate (or relic) swamp forest, sustainable swamp forest, NORTH NORTHEAST WEST EXPECTED SPRING FALL EAST SOUTH Figure 2. Observed and expected seasonal use of five locations around Lake Maurepas by migrating birds. The expected values come from the overall difference in birds detected between spring and fall. This figure shows only the proportional difference in use at each site, not the differences in bird abundance among sites. The base map includes the main landcover types: marsh (red), transitional or intermediate swamp forest (yellow), sustainable swamp forest (green), water (blue), and unclassified + urban (both brown). A few small patches of bottomland hardwood (dark green) appear within sustainable swamp forest.

4 bottomland hardwood forest, urban (such as roads, houses, and parking lots) and water (Figure 2). To analyze the combined dataset we overlaid a grid of hexagons 2.48 km on a side (~16 km 2 ). Within each hexagon, we summed the pixels for both landcover and birds to get a total radar score and the proportional landcover of each category. Analysis We began with a qualitative description of bird distribution in the Maurepas wetlands. To test for overall abundance differences between spring and fall, we used a Wilcoxon signed-rank test for paired data. We evaluated the difference in bird abundance around the lake by first assigning a location to all hexagons with ( % urban + % water) < 5% (Figure 3). Location was then used as a categorical variable in a repeated-measures analysis of variance with dbz from spring and fall as the repeated measures. We used an information-theoretic approach to identify the landcover variables that best explain the variation in bird abundance. This analysis compares models based on their information content, which is a balance of the fit weighted against the number of parameters in the model. For this analysis, we used all hexagons with urban <5% and water < 9%. This slightly broader number of acceptable hexagons increased the variance in landcover variables among samples. Even so, we had little variation in sustainable swamp forest (-19.7%) or bottomland hardwood forest (-1.2%) among hexagons. We combined sustainable swamp forest with intermediate swamp forest, and did not include bottomland hardwood forest in our analysis. We also wanted to evaluate the effect of distance from the radar, so we included a variable corresponding to the column of hexagons, with 1 on the east side, closest to the radar, and 8 on the west side. Finally, we returned to the reduced dataset used in the analysis of location for a second information-theoretic analysis to consider whether location around the lake, landcover, or a combination of the two best explained bird abundance. Results We detected more birds in fall than in spring, based on a paired comparison of spring and fall values for each hexagon (paired T-test, T=2.264, P=.3). This analysis excludes one hexagon from the northeast corner of our sampling area with dbz > 28 in fall. In addition to the overall difference in bird abundance, we also recorded different patterns of use around Lake Maurepas between the two seasons. South and west of the lake, bird abundance was higher in the spring than in the fall. East and northeast of the lake, more birds were detected in fall than in spring. To the north of the lake, birds exhibited no pronounced preference (Figure 2).

5 2 15 Figure 3. Spring and fall dbz scores for 37 hexagons (excludes one outlier). The diagonal line shows equal dbz scores in spring and fall. Spring dbz Fall dbz Repeated-measures ANOVA revealed significant season, location, and season x location effects (season F 1,24 = 28.34, P<.1; location F 4,24 = 8.68, P<.1; season x location F 4,24 = 16.43, P<.1). These results show that birds avoided the north and northeast corners of the lake in the spring, but strongly favored the northeast in the fall. All locations differed between seasons except the south (Figure 4). Mean dbz + SE b BC a A b BC a B Spring Fall a C Figure 4. Mean dbz values by location and season. Locations with the same lowercase letter do not differ in the spring; locations with the same uppercase letter do not differ in the fall (least-square means, P>.5). All locations except south differed between spring and fall. N NE E S W Bird association with landcover differed between spring and fall, as might be expected from the strong differences within locations between seasons. In spring, the strongest effect was avoidance of marsh (Σw i =.78, Figure 5). The positive relationship between forest (sustainable + intermediate) and dbz was less well supported (Σw i =.41, Figure 6), but would have been improved with exclusion of three outliers with low

6 proportion of forest but high dbz scores. These three points were from cells along the southeast corner of the lake. Proportion of water or urban had the least support (both Σw i <.35). Exploratory analysis showed no negative effect of distance from the radar in spring, so this variable was dropped before the final analysis. In fall, distance from the radar was the most important variable (Σw i = 1.), with more birds detected closer to the radar (Figure 7). Forest cover was also highly weighted (Σw i =.91, Figure 8), although this relationship included some low dbz scores for hexagons with 5-6% forest cover along the northwest corner of the lake. No other variables had Σw i >.44. These analyses all excluded the outlier in the northeast corner. 12 Spring dbz Figure 5. Hexagons with greater proportion marsh had lower dbz scores in spring. The line is a linear regression (R 2 =.2) Proportion marsh 12 Spring dbz Figure 6. Hexagons with greater proportion of sustainable and transitional forest had higher dbz scores in spring. The three points with high dbz scores at 5-3% forest were all from the southeast rim of the lake. The line is a linear regression (R 2 =.11) Proportion (sustainable + transitional) Fall dbz Band (distance from radar) Figure 7. Fall dbz scores decreased with increasing distance form the radar. The line is a linear regression (R 2 =.18).

7 Fall dbz Proportion (sustainable + transitional) Figure 8. Fall dbz scores increased with increasing proportion of sustainable and transitional forest. Low dbz values at about 5-7% forest were all from the northeast rim of the lake. The line is a linear regression (R 2 =.11). We also analyzed landcover associations from the reduced dataset that was used for the analysis of location effects (Figure 4). For spring data, we collapsed the location data into the two groups of locations that differed from each other: N+E and NE+S+W. We then modeled dbz as a function of landcover alone, as a function of location alone, and as a combination of the two (with location as a categorical effect and landcover variables as covariates). In spring, all of the models with location performed better than models with landcover alone, and no model with landcover alone was even moderately supported (all Δ AICc > 9.7). Based only on the models that included location, proportion of water was the most informative variable (Σw i =.64). Proportion of forest had approximately the same support as in the analysis of landcover alone (Σw i =.41). Proportion of marsh had less effect than in the analysis of landcover alone (Σw i =.33), possibly because most of the hexagons with high proportion of marsh were concentrated within a single location (E). As before, proportion of urban cover had the least information (Σw i =.21). For the fall data, we modeled landcover and location three ways. First, we included only landcover. Second, we combined landcover with location, as in the spring analysis. Finally, we combined landcover with distance from the radar. In all of these analyses we excluded the NE hexagons, which had much higher dbz than any other location (Figure). The best-supported models included landcover and distance from the radar; no models that included location or did not include distance from the radar received appreciable support (ΔAICc > 6.). These results suggest that, for fall, our analysis of landcover and distance from the radar were adequate. In both spring and fall, the three sample hexagons from the Pearl River basin had higher dbz scores than nearly all of the hexagons around Lake Maurepas. In spring, the lowest Pearl value (16.9) was nearly equivalent to the highest Lake Maurepas value (16.8), but no other Maurepas values were >1, compared to 149 and 28 for the other two Pearl sites. In fall, the hexagon in the northeast corner of the Maurepas sample (dbz

8 = 29) was comparable to the Pearl values (216, 29, and 323). Three other hexagons in the northeast corner approached these values (dbz>177), but the other fall samples from Lake Maurepas were considerably lower. Interpretation Our results show the broad pattern of migration around Lake Maurepas. In spring, birds are more common along the south and west sides of the lake, possibly because they stop their preceding flight before the barrier of the lake. This hypothesis is also supported by the apparent concentration of birds along the southern rim of the lake. We did not find detect an effect of distance from the radar, although this would be expected at nearly 8 km from the radar (see 4A.HTM). We probably underestimated birds south and west of the lake, indicating an even greater difference between the two sides of the lake. In fall, bird were clearly concentrated along the northeast corner of the lake. Again, this probably indicates that they ended their preceding flight before the barrier of the lake. As expected, we found more birds in the fall, presumably due to the inclusion of young of the year. The broad patterns of bird movements make it difficult to draw strong inference about landcover associations, as landcover also differs strongly around the lake. Even so, bird abundance was always explained best by models that included landcover variables. Birds were more common over sustainable and transitional forest than over open marsh, leading to the fundamental conclusion that degradation of swamp forest removes habitat used by migrating birds. At a slightly larger scale, the Maurepas system does not appear to be used as heavily as the Pearl River basin in spring. In fall, the northeast corner of the Maurepas approached the bird density of the Pearl, although the rest of the Maurepas did not. Qualitatively, we always saw stronger radar signals over the Pearl, even as far north as the central Washington Parish (e.g. Figure 1). The actual hexagons we sampled, however, were only 2-3 km from the radar, close enough that they would be expected to show more detections that the Maurepas sites. Also, birds took off slightly earlier in the east, so more may have been in the air there than over the Maurepas at the time we took our measurements. Even with these considerations, however, we are confident in the general conclusion that the bottomland hardwood forests of the Pearly River Basin are much more important than the marshes and baldcypress-tuperlogum forests around Lake Maurepas. We need to mention several additional caveats about these analyses. First, we don t know the species composition of the birds we detected. We assume that the spring birds are mostly Neotropical migrants. In fall, the sample probably includes more shortdistance migrants that will remain in south Louisiana for the winter. Second, are sample

9 dates only include nights favorable to migration, without regard to the weather the preceding day(s). Our work does not necessarily identify areas important for migrants in bad weather. Finally, we regret having just one year of radar data to include. Due to changes in the image archives we used, we were unable to apply the same analysis to images from 25. We suspect that the broad patterns of habitat use over entire seasons are generalizable across years, but we can t yet test this assumption. References Gauthreaux, S. A., and C. G. Belser. 23. Bird movements on Doppler weather surveillance radar. Birding 35: Larkin, R. P Flight speeds observed with radar, a correction: slow "birds" are insects. Behavioral Ecology and Sociobiology 29: Larkin, R. P. 24. Radar techniques for wildlife. in T. W. Society, editor. Techniques for wildlife investigations and management, 6th edition. The Wildlife Society, Bethesda, MD. Zoller, J. A. 24. Seasonal differences in bird communities of a Louisiana swamp and manipulation of the breeding density of Prothonotary Warblers. M.S. thesis. Southeastern Louisiana University.