Evaluation of the Slow-the-Spread project
Evaluation of the Slow-the-Spread projectSpread rate changes can be detected from the analysis of population boundaries for different population thresholds (e.g., 1, 10, 100 moths/trap; 1000 egg masses/ha). Two criteria are used to detect decreases in gypsy moth population spread rates caused by STS activities:
A simple model shows that inter-boundary distance should be proportional to the rate of population spread.
Data from areas with a barrier-zone:
Single-year databases can be used for estimation of inter-boundary distances. In some cases, they can be compared with other databases separated in time and used for estimation of average boundary increment in these years.
County-level data can be used only for estimation of historical spread rates averaged over long time intervals (>20 years) because otherwise their spatial precision is relatively low.
Caution should be used when comparing data from areas with different climate, topography, human demography, or vegetation. For example, population spread patterns in Michigan and Central Appalachians may be different because of different land use patterns and climate. It is possible that gypsy moth spread rate (or spread pattern) in North Carolina is different from that in VA/WV STS.
Another limiting factor is suppression and eradication activity outside of the STS area. For example, gypsy moth eradication has been conducted in the Virginia piedmont area, and this may limit it's use as a control.
From the above discussion, it is obvious that we plan to make a variety of comparisons of boundaries in STS areas with patterns in other "control" areas. Clearly these comparisons will be limited by both the lack of extensive historical data and by biological factors that confound comparisons. Nevertheless, we expect that by making a variety of comparisons with a variety of historical data, a useful "meta-analysis" of STS treatment effects will emerge.
The accuracy of evaluation criteria depends on the size of evaluation zone, the number of replications in different years, and intertrap distance. Analysis of spread rate autocorrelations indicated that the size of evaluation zone should be not less than 100 x 100 km. No autocorrelation in spread rate over time was detected, and therefore, data from different years are true replications which can increase precision for comparison of boundary changes.
Intertrap distance in the evaluation zone was 3 km in 1994, and we are examining the possibility of increasing the intertrap distance in order to reduce project costs. The error of population boundary estimates has two components: statistical error (sampling and interpolation errors) and natural variability of the boundary which are related to the variability of local factors, e.g., wind direction or forest composition.
In order to evaluate the statistical error of boundary estimates, historical AIPM trapping data were subdivided into two portions and boundaries were estimated independently using each half of the data. Standard error of average boundary position (averaged along the boundary) was 1.23 km. Natural variability of average boundaries was measured as the interaction of two factors: year x population threshold. This interaction indicated the error of 6.28 km. Rates of spread were analyzed in the same way. The results are: statistical error = 1.53 km, and natural variability error = 5.62 km.
We analyzed 5- and 10-km grids extracted from the AIPM data and found that even for a 10-km grid, statistical error (2.5 km) was less than a half of the natural variability.
Conclusion: Natural variability of population boundaries was larger than statistical error, and thus, investment in intensive sampling in order to increase statistical precision is not justified. Thus, the increase of intertrap distance to 5-km will be sufficient for population boundary analysis.
If boundaries were estimated without missing values, then both methods will give the same result. However, with missing values, results may be different. We used both methods.
We restricted our analysis to the Appalachian Mts. region (between yellow dashed lines on the map) because:
1. From 1995 data it follows that the distance between population boundaries in the coastal plain is much larger than in the Appalachian Mts., and hence the process of population spread may be different in these two regions.
2. Most of our data was obtained from the Appalachian Mts. To analyze population spread in the coastal plain we need more data.
Estimated spread rates significantly declined from 20-40 km/year to 5-14 km/year (Fig. 1, 2):
Fig. 1. Gypsy moth spread rates (average for all population thresholds).
Fig. 2. Gypsy moth spread rates measured from individual population thresholds.
Inter-boundary distance (estimated between neighboring boundaries in the
same year) significantly declined from 1988 to 1995 which may indicate "boundary
compression":
Fig. 3. Inter-boundary distance in 1980-95; regression is estimated for 1988-95.
Interboundary distance in 1980-1984 was unexpectedly small (Fig. 3) as compared with large spread rates in these years (Fig. 1-2). In 1980, most of boundaries were located outside of the mountain region where we estimated the distance between boundaries. Thus, there may be not enough data in 1984. However, in 1980 boundaries were complete and still the distance between them was small. Two hypotheses can be considered to explain this:
1. Male moths disperse relatively far from eclosion sites (by 20-30 km) and thus, counts of moths are not good for observing boundary compression.
2. 1980 was an abnormal year, and in other years (e.g., in 1981-83), population boundaries were separated by a much larger distance.
The average gypsy moth spread rate in VA/WV area in 1988-94 was 10.4 km/year. Estimated average spread rate can be used for prediction. of gypsy moth spread in space and time.
Conclusion: suppression and eradication of gypsy moth populations in the Central Appalachians may have already resulted in the reduction of population spread rates.
