Medical and Veterinary Entomology (1990) 4, 181-193 Monitoring tsetse fly populations. 11. The effect of climate on trap catches of Glossinapallidipes BRIAN WILLIAMS, ROBERT BRIGHTWELL and ROBERT DRANSFIELD International Centre for Insect Physiology and Ecology, Nairobi, Kenya ABSTRACT. In Part I it was shown that the sampling distribution of trap catches of tsetse flies, Glossina pallidipes Austen, at Nguruman, Kenya, using unbaited biconical traps follows a Poisson distribution. In this paper we examine the effect of humidity and temperature on day-to-day and seasonal variations in the trap catches. It is shown that the seasonal varia- tion is significantly correlated with maximum daily temperature, the catches increasing with temperature when the maximum temperature is below 34°C and decreasing with temperature when it is above 34°C. The correlation between trap catches and relative humidity is not as good as the correlation with the maximum temperature, and the two together do not improve the fit to the trap catches. The day-to-day variation is significantly greater than the intrinsic variation due to the stochastic nature of the sampling process and for some traps it is correlated with temperature and humidity. An autoregressive model gives a half-life for the decay of depar- tures from the mean of about 1 day and it is suggested that this indicates the movement of flies in response to animal movement or to climatic factors other than temperature or humidity. After removing the temperature de- pendent part of the seasonal variation and the autoregressive component of the data, the male and female catches are still significantly correlated. Key words. Glossina pallidipes, tsetse, traps, climate. Introduction In Part I we considered the sampling distribution of tsetse flies caught in three unbaited biconical traps at Nguruman, Kenya, and details of the study area are given in that paper. Trap catches of tsetse flies are influenced by many factors includ- ing the number of flies which are present in the sampling area, the extent of dispersal and migra- tion, the activity of the flies and the efficiency of the traps. These factors depend in turn on host Correspondence: Dr B. Williams, ICIPE, P.O. Box 30772, Nairobi, Kenya. movement, changes in the vegetation and on the local climate. In this paper we consider the effects of several climatic variables on trap catches of male and female Glossina pallidipes Austen. Materials and Methods In addition to the number of male and female G.pallidipes caught in the traps the rainfall, rela- tive humidity at 08.00 and at 15.00 hours, and daily minimum and maximum temperatures were recorded on each day during 1986 at a sta- tion about 2 km from the traps. The temperature 181
Transcript
Medical and Veterinary Entomology (1990) 4, 181-193
Monitoring tsetse fly populations. 11. The effect of climate on trap catches of Glossinapallidipes
BRIAN WILLIAMS, ROBERT BRIGHTWELL and ROBERT DRANSFIELD International Centre for Insect Physiology and Ecology, Nairobi, Kenya
ABSTRACT. In Part I it was shown that the sampling distribution of trap catches of tsetse flies, Glossina pallidipes Austen, at Nguruman, Kenya, using unbaited biconical traps follows a Poisson distribution. In this paper we examine the effect of humidity and temperature on day-to-day and seasonal variations in the trap catches. It is shown that the seasonal varia- tion is significantly correlated with maximum daily temperature, the catches increasing with temperature when the maximum temperature is below 34°C and decreasing with temperature when it is above 34°C. The correlation between trap catches and relative humidity is not as good as the correlation with the maximum temperature, and the two together do not improve the fit to the trap catches. The day-to-day variation is significantly greater than the intrinsic variation due to the stochastic nature of the sampling process and for some traps it is correlated with temperature and humidity. An autoregressive model gives a half-life for the decay of depar- tures from the mean of about 1 day and it is suggested that this indicates the movement of flies in response to animal movement or to climatic factors other than temperature or humidity. After removing the temperature de- pendent part of the seasonal variation and the autoregressive component of the data, the male and female catches are still significantly correlated.
In Part I we considered the sampling distribution of tsetse flies caught in three unbaited biconical traps at Nguruman, Kenya, and details of the study area are given in that paper. Trap catches of tsetse flies are influenced by many factors includ- ing the number of flies which are present in the sampling area, the extent of dispersal and migra- tion, the activity of the flies and the efficiency of the traps. These factors depend in turn on host
Correspondence: Dr B. Williams, ICIPE, P.O. Box 30772, Nairobi, Kenya.
movement, changes in the vegetation and on the local climate. In this paper we consider the effects of several climatic variables on trap catches of male and female Glossina pallidipes Austen.
Materials and Methods
In addition to the number of male and female G.pallidipes caught in the traps the rainfall, rela- tive humidity at 08.00 and at 15.00 hours, and daily minimum and maximum temperatures were recorded on each day during 1986 at a sta- tion about 2 km from the traps. The temperature
181
182 B. Williams, R . Brightwell and R . Dransjield
and humidity were recorded in a Stevenson that a square-root transform will stabilize the screen. Details of the traps and trapping variance of the number of flies caught in each methods are given in Part I . trap. Fig. 1 shows the catches discussed in Part I
but transformed to a new variable v , given by
Results v = v(n+0.5) (1) where n is the number of flies caught. In what follows we shall refer to v as the 'transformed catch'. It is apparent from Fig. 1 that trap 3 catches significantly more flies than either trap 1
The trap catches of G.pollidipes are given in Part I where it is shown that the sampling variability of the catches follows a Poisson distribution and
or trap 2. All three traps show a peak in their catches in April and December and the catch in trap 3 also shows a peak in June and July. There is a particularly noticeable dip in the catch in trap 3 at the end of May which appears as a deviation from an otherwise relatively smooth trend.
Table 1 gives the correlations between the catches in the va ious traps. The number of males in each trap correlates well with the number of females in the same trap and the weakest correlation is between traps 1 and 3 which are also the farthest apart (see Fig. 1 in Part I).
The daily minimum and maximum tempera- ture\. relative humidity at 08.00 hours and at 15.00 hours, and the rainfall, for 1986 are shown in Fig 2. The vapour pressure deficit was calcu- lated using the maximum temperature and the relative humidity at 15.00 hours as described in the Appendix. Table 2 gives the correlation coefficients between each pair of climatic vari- ables as well as the regression coefficients from a
B. Williams, R. Brightwell and R. Dransfield
linear fit of each one against each of the others. The standard errors in the correlation coefficients, r , are, for large values of n, approxi- mately equal to (l-?)/V(n-3) (Bulmer, 1976). The two measurements of relative hu- midity are strongly correlated, the two tempera- tures are less strongly correlated. The maximum temperature correlates well with the relative humidity at 15.00 hours, as one would expect, while the minimum temperature is not sig- nificantly correlated with either of the mea- surements of humidity. Since the minimum temperature occurs a t dawn, one would not ex- pect it to provide a good predictor of trap catches.
Table 3 gives the correlation coefficients be- tween the transformed catches in each trap and the temperatures and humidities. The catch in trap 1 does not correlate well with any of the climatic variables, the catch in trap 2 correlates best with the maximum temperature although the female catch also correlates well with the
TABLE 1. Correlation coefficients (with standard deviations) between the transformed catches in the three traps.
TABLE 2. Regression and correlation coefficients (with.standard devia- tions) between the climatic variables used in this study. The variables at the top of each column are regressed against the variables in the left-most column, a and b give the regression coefficients and c the correlation coefficients. T,,, is the daily minimum temperature, T,,, is the daily maximum temperature, RH08.00 is the relative humidity at 08.00 hours and RH15.00 the relative humidity at 15.00 hours. The coefficients a and b are not given in two cases for which there was no significant correlation.
T m a x I c a 3.924.2 b 0.447L0.049 C 0.43950.043
RH08.00 a 0 42.3f1.7 b 0 -0.108f0.010 C 0.007 f 0.053 - 0.5032 0.040
RH15.00 a 0 41.2f10 37.5 f 4 .O b 0 -0.176f0.010 0.894k0.043 C 0.066f 0.053 -0,678 2 0.029 0.743 f 0.024
Monitoring tsetse fly populations. II 185
humidity at 15.00 hours. For trap 3 the best correlation is with the maximum temperature although the minimum temperature correlates well with the male catch. The maximum tempera- ture gives the most consistent correlation with the climatic data and none of the fits discussed below are improved significantly by incorporating the other climatic variables or the vapour pressure deficit.
Fitting catches to temperature
In order to determine whether or not there are any significant delays in the system, each catch was convoluted with the maximum temperature using a fast Fourier transform (FFT) algorithm (Brigham, 1974) and the results scaled so that each point in the convolution corresponds to a wrrela- tion coefficient with the appropriate lag. To take full advantage of the speed of the FFI only 256 points were used starting from day 50. Precisely
TABLE 3. Correlation coefficients between the catches in the three traps (males and females treated separately) and the daily minifnum temperature ( Tmin), the daily maximum temperature (Tmax), the re- lative humidity at 08.00 (RH08.00) and at 15.00 hours (RH15.00). The standard deviations all lie between 0.04 and 0.06.
the same results could have been obtained by cal- culating the correlation between the two series di- rectly, shifting one set of data by one time point, recalculating the correlation, shifting the same set of data by one more time point, and so on. How- ever, the fast Fourier transform enables one to ob- tain the correlations for all possible lags directly.
TABLE 4. Parameters (with standard deviations) in the fitting of the transformed trap catches, Y. The regression of trap catch against maximum temperature for temperatures below 34°C is v=a+b T,,, and c is the correlation coefficient. d gives the factor by which the number of fliescaught decrease when the temperature drops from 34°C to 30°C. For temperatures above 34°C the regression equation is u=e+fr , , , and g is the correlation coefficient. h gives the factor by which the number of flies caught decreases when the temperature rises from 34°C to 38°C. The maximum temperatures were smoothed with a Gaussian of FWHM=2 days and lagged by 8 days for trap 1, 7 days for trap 2 and 5 days for trap 3.
16 1 9 2 8 29 3 8 39 -
a b
d
e
g h
C
f
-20f 12 0.74f0.15
0.35120.067 5.5
18.7f7.4 -0.389f0.082 -0.332 f0.067
2.0
-21216 0.82 f 0.20
0.298f0.070 3.7
33.1k9.4 -0.77f0.010
-0.479f0.058 3.2
-9211 0.40k0.13
0.225f0.073 2.4
21.427.1 -0.486i0.079 -0.416+0.062
2.8
-4f14 0.30f0.17
0.137 f0.075 1.5
33.9f8.4 -0.819+0.093 - 0.544f0.053
4.7
-10f24 0.6220.3 1
0.153f0.075 1.7
3 1 f l 1 -0.64iO. 13 - 0.350f0.066
1.9
35f33 -0.65k0.42
-0.1 18 f0.076 -0.7
49 i 1 3 - 1.11 f0.14
-0.504f0.056 2.7
TABLE 5 . The residuals from the fits in 1 for each day, d , were fitted using an autoregressive model 6,, = as,,_, . The residuals from these fits were then fitted to the high frequencycom- ponents of the maximum temperature so that 8 = STt and c gives the correlation coefficient. The variance of the residuals from these fits were then calculated and these are given by d .
I d 1 9 28 29 3d 3 9
a 0.61 k0.04 0.62f0.04 0.53f0.05 0.51f0.05 0.50t0.05 0.57f0.04 b 0.11f0.08 0.23f0.10 0.09+0.08 0.15f0.10 0.50f0.14 0.53f0.18 c 0.07f0.05 0.12+0.05 0.06+0.05 0.08+0.05 0.19f0.05 0.15f0.05 d 1.07 1.89 1.47 I .81 3.19 4.28
186 B. Williams, R. Brightwell and R . Dransfield
The results are shown in Fig. 3 in which the maximum inverse correlations between the catches and the maximum daily temperature occur at lags ranging from 5 days (trap 3 ) to 8 days (trap 1). Each of the other climatic variables, namely minimum temperature and the two relative humidities, were examined in the same way and where the correlations were significant they indi-
cated the same lag as obtained using the maximum temperature.
In order to separate seasonal effects from day- to-day variations, the temperature data were smoothed with a Gaussian of full width at half- maximum (FWHM) equal to 2 days. In all cases smoothing the maximum temperature increased the correlation with the transformed catches
c 2ol
L
‘1 ’ 1
~ , ~ , ~ , ~ , ~ , ~ , ~ , ~ , ~ , ~ , l , ~ , ~ , J F M A M J J A S O N D
FIG. 2(a) Months
FIG. ’2. Daily climatic data for 1986. T,,, and T,,, are the daily maximum and minimum temperatures, RH08.00 and RH15.00 are the relative humidities measured at 08.00 and 15.00 hours, respectively. The vapour pressure deficit is measured in Pascals (100 Pa=l mb) and the rainfall is measured in mm.
FIG. Z(b)
I I I I I I I I I 0 0 0 LD
L 4
L 0 0
0 In
h
v 0 N
h
x
Monitoring tsetsefly populations. I1 187
I I I I I I I I ~ I I I I J I I I I ~ I I I I ( I I I
0 0 0 co
0 0 0 co
0 0 0 d
c N
188
slightly indicating that the correlations and the 5-8 day lags are related to seasonal rather than day-to- day changes.
In Fig. 3 the transformed. lagged catches are plotted against the smoothed maximum tempera- ture, using all the data points for 1986. The catches appear to increase with temperature up to about 34°C and decrease with temperature above 34°C. Separate fits were therefore made to the data above and below 34°C as shown in Fig. 4. The
R. Williams, R. Brightwell and R. Dransfield
FIG 3(a)
0.2
c S a,
parameters for these regressions are given in Table 4.
The increases in trap catches with temperature up to 34°C and the decrease above 34°C are statis- tically significant in all but two cases: the female catches in traps 2 and 3 below 34°C. The fit for males in trap 3 below 34°C is significant at the 5% level and all the rest at the 1% level.
In the case of trap 3 males above 34T, the cor- relation coefficient is 10 standard deviations away
1 0
FIG. 3 Correlation between standardized trap catches and maximum temperature as a function of the lag in days. (a) Trap 1 males. (b) trap 1 females, (c) trap 2 males, (d) trap 2 females, (e) trap 3 males, ( f ) trap 3 females. Only days SO to 50+2.56=306 were used in the analysis. The lines were drawns by eye.
1 ' 1 ' 1 ' 1 ' ' 1 ' 1 ' 1 ' 1 1 I I I I I I ' I r - -
.c. C a, 0
a,
.-
8
t \
E
FIG 3(h)
0 10 tagidays
Monitoring tsetsepy populations. II 189
(? ?
In t P
h
v -0
t 9 m
0 Lo P
In
s Lo ? m
(? 9
T 9
h
v m
190
from zero. In order to make comparisons with other experiments the factors by which the trap catch decreases when the temperature falls by 4°C from 34°C or increases by 4°C from 34°C are also given in Table 4. These factors range from 1 to 6 and it is interesting to note that below 34°C the males appear to be consistently more sensitive to temperature changes than females, while above 34°C the females appear to be consistently more sensitive to temperature changes than males.
B. Williams, R. Brightwell and R. Dransfield
l ' l ' l ' l '
Having fitted the trap catches to the tempera- ture data there is still a significant degree of correlation in the data and so an autoregressive model was fitted to the residuals. If i jd is the re- sidual on day d , the model is then
6d=a8&1 ( 2 ) and the coefficients a are given in Table 5.
The regression coefficients lie between 0.5 and 0.6 and are all highly significant. Using addi-
tional terms with greater lags does not improve the fit. Since the autoregressive coefficients lie between 0.5 and 0.6, the model indicates that de- viations from the value obtained by fitting catches to the maximum temperature, decay ex- ponentially with a half-life of about 1 day.
The residuals from the autogressive model, 6, were then fitted to the high frequency part of the temperature variation, 7". This was calculated by subtracting the daily maximum temperatures from the smoothed temperatures and using this in the regression so that
6 = h T (3) and the coefficients b are given in Table 5. The fits for the trap 3 are significant at the 1% level and for the males in trap 1 at the 5% level and in all cases the residuals are positively correlated with the maximum temperature indicating that on a day-to-day basis the catch increases with temperature in all traps.
In order to determine how much of the var- iance still needs to be explained, row d in Table 5 gives the variance of the residuals from the last fit carried out and this is seen to range from 1 to 4. Now we have shown in Part I that the variance which arises from the stochastic nature of the sampling process is 0.25 when all of the signifi- cant variations have been accounted for. In other words, between 75% and 94% of the re- maining variance has still to be accounted for. The correlation coefficients between each pair of these residuals are given Table 6 and the residu- als are still highly correlated for all three traps.
TABLE 6 Correlation coefficients between the re- siduals from the fits given in Table 5 . The standard de- viations are all between 0.02 and 0.04.
~
Id 1 9 2 8 2 9 36
I Y 0.79 211 0.59 0.57 20 0.52 0.54 0.77 3 6 0 57 0 . s 0.53 0.43 3 9 0.44 0.54 0.45 0.45 0.79
Discussion
The catches of G.pallidipes in trap 3 are, on av- erage, about 4 times higher than the catches in traps 1 and 2. This IS probably due to the greater prevalence of game animals in the vicinity of trap
3 as compared to traps 1 and 2 as described in Part I. Furthermore, the correlation between the catches is strongest for traps 1 and 2 and weakest for traps 1 and 3 indicating some depend- dence of the correlation on the distance between the traps.
The correlation between the trap catches and the maximum temperature is negative when the maximum temperature is above 34°C and this is probably due to the movement of flies from the denser, more extensive vegetation to the north and south of the three traps which are in com- paratively open vegetation. The study area is close to the upper temperature limit for tsetse flies and when it is very hot and dry the flies will tend to retreat into thicker cooler vegetation. As the temperature drops the flies would tend to spread out into more open areas giving rise to the high temperature variation in the catches. This argument is reinforced by the lag between temperature and catch since the lag may reflect the time taken to move from neighbouring areas of high population density to the three traps discussed above. Rogers (1977) shows that for several species of tsetse fly the mean daily dis- placement is between 100 and 400 m. Taking 250 m as an estimate of the distance moved by tsetse flies in 1 day, a lag of 5-8 days corresponds to a distance of 1-2 km which is approximately the distance to the closest regions of dense vegeta- tion within which tsetse densities are signifi- cantly higher than in the region of the traps dis- cussed here. There is separate evidence, to be published elsewhere, that suggest that popula- tion changes are significantly affected by fly movement from the top of the escarpment, to the west of the study area, and from the area to the north of the traps discussed here. Further ex- periments to determine the extent and nature of movement between the various ecological zones are being undertaken.
At low temperatures the correlation between maximum temperature and the catch of G.paZ- lidipes is positive but this may reflect changes in the efficiency of the traps. Brightwell et al. (1987) show that when the mean temperature is below 30°C the catch of baited biconical traps in- creases by a factor of 4 for an increase in the mean temperature of 4°C but that at mean tem- peratures above 30°C the catch does not vary sig- nificantly with temperature. These changes are of the same order of magnitude as the changes observed in this study at low temperatures.
Monitoring tsetse fly populations. I1 193
After removing the seasonal, temperature de- pendent part of the variation in catch, the day- to-day variation in fly numbers is still greatly in excess of the residual variation due to the stochastic nature of the sampling process. This day-to-day variation is just correlated signific- antly with the maximum temperature and is probably also related to changes in the efficiency of the traps. The weakness of the correlation probably arises because the trap catches are strongly influenced by the temperature 5-8 days earlier.
The catches are still significantly correlated from one day to the next and an autoregressive model with a lag of 1 day removes most of the correlation; including terms with greater lags does not improve the fit. This implies that devia- tions from the mean disappear within 1 or 2 days. Changes on this time scale may well be as- sociated with movement of the flies in response to changes in climatic factors or, perhaps, move- ment of wild hosts.
After removing the correlation a significant amount of variation remains to be explained. Since the day-to-day correlation has been re- moved this remaining variation is likely to be due to changes in the activity pattern, possibly determined by climatic variables which we have not been able to measure such as wind speed and direction or perhaps irradiance.
The data discussed in this paper brings out clearly the large day-to-day variations in trap catches which have not previously been investi- gated in detail. The trap catches are influenced by many factors and determining the role and importance of each is not easy. Further experi- ments are needed if we are to fully understand the changes which are observed. In particular, detailed studies of fly movement between differ- ent ecological zones which are found at Nguru- man are planned and these studies may help us to understand the data presented here. More de- tailed meteorological data is also needed includ- ing variables such as solar radiation and wind
speed and direction and such data should be re- corded throughout the active period of the flies which for G.pallidipes begins at dawn, increases through the day, and drops to zero at dusk.
Acknowledgments
We would like to thank the International Centre for Insect Physiology and Ecology for the provi- sion of research facilities and Dr David Rogers for a number of important comments and suggestions, and our field assistants for all their work.
Appendix
The saturation vapour pressure P, in Pascals is given by In P,=A-BIT-C In Twhere A=57.96 Pa, B=6731.0 PaK, C=4.796 Pa, and T is the temperature in degrees Kelvin. If P is the actual vapour pressure and R is the relative humidity, R=lOOPIPs, so that the saturation pressure de- ficit, D, is D=Ps-P=P,(lOO-R)/lOO (Wood- ward & Sheehy, 1983).
References
Brighman, E.O. (1974) The Fast Fourier Transform. Prentice Hall, Englewood Cliffs.
Brightwell, R., Dransfield, R., Kyorku, C., Golder, T.K., Tarimo, S.A. & Mungai, D. (1987) A new trap for Glossina pallidipes. Tropical Pest Manage- ment, 33,151-159.
Bulmer, M.G. (1976) Principles of Statistics. Oliver & Boyd, London.
Rogers, D. (1977) Study of a natural population of Glossina fuscipes fuscipes Newstead and a model of fly movement. Journal of Animal Ecology, 46, 281-307.
Woodward, F.I. & Sheehy, J.E. (1983) Principles and Measurements in Environmental Biology. Butter- worths, London.
