We analyzed a total of 2196 morbidity reports received in Israel between November 2018 and October 2021 (Fig. 1A). The majority of these were for adult bats 1,783 (81.2%), of which 1432 (80.3%) were from urban areas (settlements populated by 30,000 or more people) and the remaining 351 (19.7%) were from rural areas, including small villages, nature reserves, and army bases. Out of all these adult cases, the animal’s sex was identified in only 295 cases (16.6%), with 171 (58%) being females and 124 (42%) being males. 413 (18.8% of all cases) were pups, up to 4 months old, from throughout the country, without sex identification.
We found a dramatic and significant increase in adult bat morbidity during winter (see Fig. 1B) (p < 0.0001, F = 90.9, DF = 153, one-way ANOVA). The average number of cases per week during the winter was ~ 2.3 fold higher than that during summer (a mean of 17.0 ± 7.9 SD per week with a maximum of 39 in winter, vs. 7.4 ± 4.6 SD with a maximum of 23 cases in summer). The winter effect was significant for both rural and urban areas (p < 0.0001, F = 96, p = 0.0.01, F = 10.9, respectively, DF = 153 for both, one-way ANOVA). We also found a significant difference between males and females during the research period, with a higher number of reported cases of female morbidity (p = 0.007, p < 0.0001 for sex and week, respectively, DF = 307, GLM with Poisson distribution, explained parameter: number of cases, explaining parameters: week number, sex (fixed)).
Pup morbidity exhibited different temporal patterns to those of adults (Fig. 1C, blue line). Pup morbidity peaked twice a year, in April and September, one month after the peak of parturition in March and August12,13. Pup morbidity was significantly negatively correlated with adult morbidity (GLM for pup morbidity as the explained parameter and adult morbidity, as the fixed explaining parameter; p = 0.0003, F = 13.8, DF = 153).
We observed a significant increase in morbidity in winter of all morbidity types. Of all adult cases, 59.7% were categorized and 40.3% were undefined. Of the diagnosed cases, 78.2% were caused by acute trauma, 18.4% had a putative infectious origin, and 3.4% were attributed to chronic, non-infectious conditions of the animal’s feet. 70% out of the cases with infectious etiology were diagnosed as abscesses of bacterial origin (see “Methods” section for morbidity categories, and Supplementary Fig. 2). The type of morbidity proportions remained constant year-round (p = 0.84, 0.66, 0.76, 0.56, for acute trauma, infection, feet condition, and undefined, respectively, DF = 344, GLM model for the explanatory parameter: rank of cases for morbidity, explanatory parameters: month, see “Methods” section).
Since, compared to the tropics, winter (November–March) in Israel is characterized by relatively harsh weather and a decrease in food (fruit availability), we next examined which of these factors better explained the increase in adult morbidity. For this analysis, only adult cases from the central coastline of Israel were selected (see Fig. 1A, 718 cases, 40.3% of all adult cases). We focused on this region, monitoring the movement and foraging data of 36 bats roosting in a colony at Tel-Aviv University during the same period (2018–2021). First, we examined the pair-wise correlations between all explanatory measured parameters. The time of year, expressed as the week number, was found to be highly correlated with several parameters, (see Fig. 2A, and “Methods” section). We thus tested models with all parameter combinations (see below), and the models’ residuals were also auto-correlated in time (Durbin Watson test, see “Methods” section). We applied a First Order Auto Regression (AR1) model14, that predicted the weekly morbidity by the temperature and the previous weekly morbidity. It demonstrated similar performance to the temperature-week model, which included the week along winter as a parameter representing time (R2 = 0.36, 0.35; RMSE = 2.8, 2.9, respectively). Therefore, we included in our analysis the week-number as an explanatory parameter with interactions with the other parameters.
Statistical models explaining morbidity. (A) The pairwise correlation matrix between the analyzed variables (only for the central coastline data). Color and text depict the Pearson correlation coefficient (AdultsMorbidity (AM): weekly adult cases; PupMorbidity (PM): weekly pup cases; WeekNumber (WM): week-number from November 1st in each year; MinTemp (MT): average of the daily minimum temperature over a week (°C); DailyRain (DR): the average daily precipitation over a week (mm); WindSpeed (WS): the average of the daily maximal wind-speed (m/s); TotalDistance (TD): the total distance the bats flew each day (km); CommuteVelocity (CV): the average flight velocity during the commute (m/min); TimeOutside (TO): the average time the bats spent outside their colony (hours); Top5Visits (T5): the average number of visits by each bat to one the five preferred tree species; TotalSpeciesAvailable (TS): the average total number of different tree species the bats visited each week). (B) Adult Morbidity in March: an extremely cold period in March 2022 (left panel) was correlated with a significant increase in morbidity (right panel) relative to the equivalent period in the previous three years. Asterisks indicate a significant difference, see “Results” section. (C) Adult morbidity along the central coastline over time. (D) Adults morbidity in six different regions in Israel. In panels (A,D): The average minimum daily temperature is depicted by a color code ranging from blue (~ 5 °C) to yellow (~ 27 °C). Blue lines and shaded areas indicate the prediction and the 95% confidence intervals of the temperature-winter model for the entire period. Red lines and shaded areas indicate the prediction of the best-fitted model, including the monitored visits to the favorite fruit trees (see text).
Next, we analyzed the effect of weather factors alone on adult morbidity for the entire period (156 weeks) and separately for wintertime only (66 weeks). We examined all possible models with the following weather factors: minimum temperature, daily precipitation, and maximum wind speed; with the week number representing time (with and without interactions between parameters), and ranked them according to their AICs (a total of 23 models for each period, see Sup. Tables 1–3). Temperature and time were significant parameters in all of the top three models. Moreover, the temperature-week model (i.e., a GLM with the minimum temperature, week-number, and their interaction as fixed explanatory parameters) was the only model appearing in the top three models for both periods (p < 0.0001, DF = 151, R2 = 0.35, GLM for the entire period; p < 0.0001, DF = 62, R2 = 0.3, GLM for the winter). This model also consisted in the lowest number of coefficients (four), and the AIC difference was insignificant in comparison to the top three models (< 1.45 difference from the best in each period). The significant effects of temperature and time in this model even during the wintertime only (a total of 468 cases in November–March), indicate that the impact of temperature and time was not a result of the winter versus summer batching phenomenon. Consequently, we chose the temperature-week model as the best-fitted model for the weather factors (see Fig. 2C).
We also analyzed which of the model’s parameters better explained morbidity’ and found that temperature explained it better than the week number, see Sup. Table 2, model #11 (temperature only), and model #19 (week only). This is reinforced by the increased morbidity during weeks that were particularly colder than the average (Fig. 2C). For example, in March 2022, the weather in Israel was exceptionally cold, with an average drop of ca. 2.5 °C compared to the average minimum temperature in March 2019–2021 (P = 0.01, F = 8.64, DF = 14, One Way ANOVA). Accordingly, bat morbidity increased significantly (almost doubled) from 4.17 ± 0.6 to 8.25 ± 1 cases per week, Fig. 2C (p = 0.005, F = 11.22, DF = 14, One Way ANOVA).
To determine whether foraging behavior and food availability are significantly related to morbidity, in addition to the weather conditions, we added the data collected by our GPS tags to the temperature-week model (the explanatory factors were all for week averages and comprised total flight distance, commute velocity, time spent outside, number of visits to the top five tree species, total number of tree species visited by the bats, minimum temperature, and week number, see “Methods” section). Again, we examined all possible models (total of 224 models, see “Methods” section) explaining morbidity and chose the best model with the minimum AIC. If several models produced the same AIC, we chose the one with the least number of parameters (see Supplementary Table 4). The model that included temperature, number of visits to the top five tree species, and interaction with the week number (i.e., time) was revealed as the best explanatory model (i.e., the temperature-week-trees model). This is a relatively simple model (with 6 coefficients, including the interactions, all of which are significant), which best explained morbidity (p < 0.0001, DF = 62, R2 = 0.36 GLM, see Supplementary Table 1). It insignificantly differed from two more complex models (featuring16 and 18 parameters; the delta AIC less than 1.7, see Supplementary Table 4). Compared to the temperature-week model, the new model prominently improved performance: the R2 increased from 0.3 to 0.36 and the root mean squared errors decreased from 3.4 to 3.1 (see Fig. 2C and Supplementary Table 1).
The temperature-week-trees model and the temperature-week model for the central coastline were also used to explain adult morbidity countrywide. To this end, the cases were geographically divided according to their closest distance to one of six meteorological stations, and the categorical locations were added to the models as fixed explanatory parameters. Both models explained the morbidity well, see Fig. 2D (the extrapolated temperature-week-trees model: p < 0.0001, DF = 354, R2 = 0.51, RMSE = 1.97, the extrapolated temperature-week model: p < 0.0001, DF = 356, R2 = 0.51, RMSE = 2.03, GLM). The marginal effect of the visits to the preferred trees along the central coastline on the countrywide morbidity was still significant (p= 0.04, F = 4.05, DF = 354), but with no impact on the R-squared criterion and only a 3% improvement to the RMSE.
Source: Ecology - nature.com
