Table 1.
Details of individual lesser kestrels tracked during the study period.
Fig 1.
Acceleration signature of different lesser kestrel behaviors.
Raw acceleration measured in the surge (X), sway (Y) and heave (Z) axes, ODBA, pitch, and fundamental frequency of the heave axis (rows) in flapping flight, soaring-gliding flight, hovering flight, perching, and incubating/brooding behaviors (columns). Blue dots indicate ODBA and mean pitch at 1-second interval in their respective panels. Red dots indicate fundamental frequency of the heave axis at 1-second interval.
Table 2.
Mean and standard deviation of raw acceleration measured in the surge (X), sway (Y) and heave (Z) axes, ODBA, pitch and fundamental frequency of heave axis per lesser kestrel behavior.
Sample size = 3,024,000 intervals from six individual lesser kestrel (35 complete days of tracking).
Table 3.
Confusion matrix for the final classification model of behaviors.
We built this matrix using the 30% of the tagged acceleration data selected at random to test the model after training it with the remaining 70%. Soaring-gliding and incubating/brooding are indicated as Gliding and Incubating, respectively. Observations correctly classified per behavior are shown in bold.
Fig 2.
Decision tree for the final behavior classification model.
Table 4.
Estimates (β), standard error (S.E.) and statistical significance of predictors included in the GLMM fitted to daily ODBA (energy expenditure per day) of the lesser kestrel.
Statistically significant variables are shown in bold. Sample size = 35 complete days of tracking.
Table 5.
Estimates (β), standard error (S.E.) and statistical significance of predictors included in the GLMM fitted to daily time expenditure in different behaviors by the lesser kestrel.
Statistically significant variables are shown in bold: * p < 0.5, ** p <0.01, *** p<0.001. Sample size = 35 complete days of tracking.
Table 6.
Evaluation of the relative importance of each predictor in the GAMMs fitted to response variables analyzed at the foraging trip level.
ΔAIC indicates the difference between the best model and the same model adding (negative values) or removing (positive values) the target predictor (depending on the predictors included in the best model). The higher the ΔAIC, the higher the relative importance of the predictor in the model. The predictors are coded as follows: Phenological Period = “PP”, sex = “S”, and hour-of-day = “H”. Hour-of-day was smoothed with a spline. Sample size = 444 foraging trips.
Table 7.
Percentage of time and energy (ODBA) expenditure in different behaviors (mean value ± standard deviation) during the entire foraging trip, commuting flights and foraging event of the lesser kestrel.
Sample size = 444 foraging trips, 888 commuting flights and 444 foraging events.
Fig 3.
Partial effect of hour-of-day in the model fitted to lesser kestrel foraging trip duration.
A penalized smoothing spline of 2.15 degrees of freedom was adjusted to hour-of-day. Grey shading represents the standard error of the mean effect. Sample size = 444 foraging trips.
Fig 4.
Partial effect of hour-of-day in the model fitted to lesser kestrel foraging trip maximum distance from the colony.
A penalized smoothing spline of 4.70 degrees of freedom was adjusted to hour-of-day. Grey shading represents the standard error of the mean effect. Sample size = 444 foraging trips.
Fig 5.
Partial effect of departure time (hour-of-day) on the percentage of time allocated to each behavior along foraging trips.
Flapping flight (upper left panel), soaring-gliding flight (upper right panel), hovering flight (bottom left panel) and perching (bottom right panel). Penalized smoothing splines of 4.78, 7.00, 3.82 and 5.55 degrees of freedom were adjusted to hour-of-day for flapping flight, soaring-gliding flight, hovering flight, and perching, respectively. Grey shading represents the standard error of the mean effect. Sample size = 444 foraging trips.
Fig 6.
Partial effect of departure time (hour-of-day) on duration of commuting flights (left panel) and duration of foraging events (right panel).
Penalized smoothing splines of 4.89 and 2.84 degrees of freedom were adjusted to hour-of-day for commuting flight and foraging event duration, respectively. Grey shading represents the standard error of the mean effect. Sample size = 888 commuting flights and 444 foraging events.
Fig 7.
Frequency histogram of flapping (left panel) and hovering ratio (right panel) during commuting flights and foraging events, respectively, of foraging trips.
The red dashed lines indicate the median value of ratios. Sample size = 888 commuting flights and 444 foraging events.
Fig 8.
Effect of solar radiation on flapping ratio of lesser kestrel commuting flights predicted by the GLMM.
Circles represent the observed flapping ratio of commuting flights and the solid line represents the model prediction. Sample size = 888 commuting flights.
Table 8.
Evaluation of the relative importance of each predictor in the GAMMs fitted to response variables analyzed at foraging trip segment level.
ΔAIC indicates the difference between the best model and the same model adding (negative values) or removing (positive values) the target predictor (depending on the predictors included in the best model). The higher the ΔAIC, the higher the relative importance of the predictor in the model. The predictors are coded as follows: Phenological Period = “PP”, sex = “S”, hour-of-day = “H”, commuting flight type = “CF”, air temperature = “T” and wind speed = “W”.—indicates predictor not considered in the model. Predictors hour-of-day, air temperature, and wind speed were smoothed with splines. Sample size = 888 commuting flights and 444 foraging events.
Table 9.
Estimates (β), standard error (S.E.) and statistical significance of predictors included in the GLMM fitted to flapping ratio of lesser kestrel commuting flights.
Statistically significant predictors are shown in bold. Sample size = 888 commuting flights.
Fig 9.
Partial effect of wind speed (left panel) and air temperature (right panel) on the model fitted to hovering ratio of lesser kestrel foraging events.
Penalized smoothing splines of 1.54 and 4.15 degrees of freedom were adjusted to wind speed and air temperature, respectively. Grey shading represents the standard error of the mean effect. Sample size = 444 foraging events.
Fig 10.
Partial effect of hour-of-day in the models fitted to the number of hovering bouts (left panel), the number of perching bouts (middle panel) and the number of hovering-perching bouts (right panel) per foraging event.
Penalized smoothing splines of 2.35, 3.56 and 3.47 degrees of freedom were adjusted to hour-of-day for the number of hovering bouts, perching bouts, and hovering-perching bouts per foraging event, respectively. Grey shading represents the standard error of the mean effect. Sample size = 444 foraging events.
Table 10.
Evaluation of the relative importance of each predictor in the GAMMs fitted to the number of hovering bouts, perching bouts and hovering-perching bouts per foraging event.
ΔAIC indicates the difference between the best model and the same model adding (negative values) or removing (positive values) the target predictor (depending on the predictors included in the best model). The higher the ΔAIC, the higher the importance of the predictor in the model. The predictors are coded as follows: Phenological Period = “PP”, individual sex = “S”, and hour-of-day = “H”. Hour-of-day was smoothed with a spline. Sample size = 444 foraging events.
Fig 11.
Example of two foraging flights with different strategies with behaviors tagged using the final classification model.
GPS fixes recorded at 1-second frequency. The colors of the icons represent different behaviors: flapping (blue), soaring-gliding (green), hovering (orange), and perching (red). The black star indicates the breeding colony and the pink star indicates an overnight roost. Black arrows indicate movement direction. Boxes include a zoomed view of the foraging trip segment indicated with the same color in the main panel. Early morning hunting behavior: A) foraging event (hovering ratio = 0.02); (B) commuting flights (mean flapping ratio = 0.71). Noon hunting behavior: (C) commuting flights (mean flapping ratio = 0.29); and (D) foraging event (hovering ratio = 0.77).