Fig 1.
Schematic representation of the developed analytical forecasting tool.
Fig 2.
Mosquito relative abundance over time.
A Observed and forecasted number of collected specimens per trap night over time in Los Angeles County, CA. The solid line represents the number of Ae. aegypti collected by the mosquito surveillance system per week, the dashed line represents the median of the second week of the 2-week ahead forecast, and the shaded areas are the 90% IQR of the second week of the 2-week ahead forecast. B As A, but for Maricopa County, AZ. C As A, but for Key West, FL. D As A, but for Miami-Dade County, FL, respectively.
Fig 3.
Forecasts accuracy and coverage.
A Scatter plot of the observed vs. median forecasted Ae. aegypti relative abundance for each study site and week. All forecasting horizons (1- to 4-week ahead forecasts) are included in this plot. Each color represents a study site. The dashed line is the identity line, and the dotted lines represent the variability within one order of magnitude. B Violin plot representing the distribution of the sMAPE for each study site. The boxes report the 25%, 50%, and 75% quantiles of sMAPE. C Forecast coverage as a function of the forecast interquantile range. Each colored line represents a study site. The dashed line is the identity line.
Fig 4.
Mosquito abundance forecast results for each forecast horizon week.
A Scatter plot of the observed vs. median forecasted Ae. aegypti relative abundance for each study site and week. Each color represents a forecasting horizon in weeks ahead of the reference date. The dashed line is the identity line, and the dotted lines represent the variability within one order of magnitude. B Violin plot representing the distribution of the sMAPE for each forecasting horizon in weeks. The boxes report the 25%, 50%, and 75% quantiles of sMAPE. C Forecast coverage as a function of the forecast interquantile range. Each colored line represents a forecasting horizon. The dashed line is the identity line.
Table 1.
Performance of our forecasting tool with respect to a naïve model.
The first data column shows the season average WIS score of our model divided by the same score of the naïve model. The second data column shows the season sMAPE score of our model divided by the same score of the naïve model. For both these columns, a ratio smaller than 1 means that our model had better performance than the naïve model. The third data column shows the absolute 95% coverage error of our forecasting tool subtracted by the same score of the naïve model. For this column, a negative value means that our model performed better than the naïve in terms of coverage. Cells with a green background represent those where our forecasting tool performs better the naïve model; for red cells the opposite is true.
Fig 5.
Mosquito abundance forecast results by mosquito activity levels.
A Scatter plot of the observed vs. median forecasted Ae. aegypti relative abundance for each study site and week. Each color represents a mosquito activity level (low, medium and high). The dashed line is the identity line, and the dotted lines represent the variability within one order of magnitude. B Violin plot representing the distribution of the sMAPE for each activity level. The boxes report the 25%, 50%, and 75% quantiles of sMAPE. C Forecast coverage as a function of the forecast interquantile range. Each colored line represents an activity level. The dashed line is the identity line.