SURVEILLANCE OF WEST NILE VIRUS IN MARYLAND: INTEGRATING SMART VECTOR IDENTIFICATION WITH ENVIRONMENTAL AND EPIDEMIOLOGICAL INSIGHTS

Surveillance framework

Weekly collections from the last week of June (EpiWeek 26) through the last week of September (EpiWk 39) were completed at 12 sites in Anne Arundel County, Maryland in 2023 and 2024 (Figs. S1 and S2). Sites were selected by the Maryland Department of Agriculture, targeting areas proximal to water facilities where mosquito development may flourish or residential areas where vector-to-human interactions may occur. In 2023, trapping was not completed during EpiWk 27 due to labor availability. In 2024, two collection sites were replaced to increase geographical representation across the county and broaden the assessment of potential WNV transmission sites. Centers for Disease Control and Prevention (CDC) Gravid Traps (John W. Hock Company, Gainesville, FL) were utilized to standardize sampling and target the key WNV vector species, Cx. pipiens Hay infusions were prepared by mixing 3 gallons (11.4 liters) of warm water, 100 g of cut grass, and a 1/4 tsp of dry instant yeast (Lesaffre Corp., Milwaukee, Wisconsin) to a bucket. Separately, 1 tbsp of dry chicken manure pellets (Manures.com, West Covina, CA) were added to 1 gal (3.8 liters) of warm water. Both mixtures were allowed to incubate for 4 days prior to weekly collections. On the morning of collections, 2 cups (480 ml) of hay infusion were added to the chicken manure solution in 1 gal jugs. In the field, oviposition trap tubs were filled with 1 gal of hay infusion mixture to replicate stagnant water conditions preferred by ovipositing, gravid Cx. pipiens females (Lampman and Novak 1996). Traps were set at each collection site in the morning powered by a 6V lead-acid battery (John W. Hock Company, Gainesville, Florida) and collected after 24 h. Trap collection bags were placed in a cooler lined with ice packs and transferred to a −20°C freezer when collections were complete to maintain a cold chain necessary for minimizing ribonucleic acid (RNA) virus degradation.

The IDX workflow

All captured mosquitoes were identified and sorted by Vectech’s entomologists on each collection date. Culex pipiens females were processed immediately and pooled by collection site with up to 40 mosquitoes per pool. The IDX imaging of all remaining specimens (including damaged, rare, or unknown specimens to newly trained entomologists) followed. Mosquitoes were imaged to verify species and sex identifications (Figs. S3-S7) using the deployed algorithm described below, and to log such data in the IDX’s databases. All the IDX imaged and identified female mosquitoes were similarly pooled. Pools were submitted to the Maryland Department of Health Laboratories Administration, Division of Molecular Biology - Molecular Diagnostics/Bioterrorism (Baltimore, MD) on dry ice for reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR) pathogen screening which included assays for WNV, St. Louis encephalitis virus (SLEV), eastern equine encephalitis virus (EEEV), dengue virus (DENV), and the malaria parasite, Plasmodium falciparium.(Welch). Cold chain integrity was maintained throughout the process.

Vector identification model

The IDX is a smart digital microscope designed for batch imaging and identification of mosquitoes and ticks (Achee 2022, Brey et al. 2022) (Fig. S6). It is a cloud-connected lab device that streamlines vector surveillance data acquisition and curation, allowing a minimally trained user to generate high quality, standardized species-level data on captured specimens (Fig. S6). The IDX functions with a mature, deep convolutional neural network (CNN) trained on a large dataset of mosquito images to classify imaged mosquitoes based on visually distinguishable features (Goodwin et al. 2021, Brey et al. 2022). Algorithm versions v2.0.0, v2.0.1, v4.0.1, and v4.0.2 were used for these surveillance projects.

Weather data integration

Daily temperature and precipitation data were extracted from Anne Arundel County weather station archives (NCEI-NOAA 2024) to assess environmental factors driving mosquito population dynamics. Only one weather station in the county reported temperature data. Daily precipitation data was collected from four weather stations most proximal to collection sites to accurately represent the entire study area (Figs. S1 and S2). This was limited to four weather stations due to missing or inconsistent data in the archives across other county stations. Weather data from EpiWk 22-25, the 4 wk preceding the collection period, were included to assess mosquito abundance correlated to previous wk’s weather. Temperature and precipitation were reported as weekly trends to align with weekly mosquito surveillance reports. To capture thermal variability, the daily maximum and minimum temperatures (°C) of each wk were averaged across 7 days ( $\mathit{AvgMax},\mathit{AvgMin})$ then depicted as a spread ( $\mathit{TSpread})$by calculating the difference between average weekly maximum and minimum temperatures. $$\mathit{AvgMax}=\frac{\mathit{Tmax}1+\mathit{Tmax}2+\mathit{Tmax}3+\dots .+\mathit{Tmax}7}{7}$$ $$\mathit{AvgMin}=\frac{\mathit{Tmin}1+\mathit{Tmin}2+\mathit{Tmin}3+\dots .+\mathit{Tmin}7}{7}$$ $$\mathit{TSpread}=\mathit{AvgMax}-\mathit{AvgMin}$$

Weekly cumulative precipitation (mm) was averaged ( $\mathit{AvgCumulativeP})$across 7 days and error bars displaying the standard error of the mean represent rainfall patterns. $$\mathit{AvgCumulativeP}=\frac{P1+P2+P3+\dots .\hspace{0.17em}P7}{7}$$

Data scope and analysis

Species diversity, abundance, and WNV prevalence were analyzed and reported weekly to guide MDA’s mosquito control campaigns in Anne Arundel County. Species diversity and abundance are represented here as the total number of mosquitoes per species collected over EpiWeeks 26–39 in 2023 and 2024 across all traps. The abundance was calculated using the geometric William’s mean of collections across all traps, applying log transformation to reduce variation, normalize distributions, and minimize the influence of extreme values driven by environmental, technical, or site-specific conditions (Haddow 1960). $$\mathit{William\text{'}s}\hspace{0.17em}\mathit{Mean}\hspace{0.17em}=\hspace{0.17em}\mathit{exp}(\frac{1}{n}{\displaystyle \sum }_{i=1}^{n}ln(x_i+1))-1_{}$$

The minimum infection rate (MIR) of Cx. pipiens was calculated by dividing the total number of positive pools ( $\Sigma \mathit{positivepools}$) by the total number of mosquitoes screened ( $\Sigma \mathit{mosquitoesscreened}$) based on standard WNV surveillance methods (CDC 2025a). The MIR is used to estimate infection rates by assuming that each positive pool contains only one infected individual, providing a reliable lower bound of the true infection rate (Gu et al. 2008). $$\mathit{MIR}\hspace{0.17em}=\frac{\Sigma \mathit{positivepools}}{\Sigma \mathit{mosquitoesscreened}}$$

The vector index (VI) of Cx. pipiens was calculated by multiplying the average number of mosquitoes per trap night ( ${\mu }_{\mathit{mosquitoespertrapnight}}$) by the estimated proportion of mosquitoes infected ( $\mathit{MIR}$) (CDC 2025a). The VI estimates the average number of WNV-infected mosquitoes per trap night by combining mosquito abundance with infection prevalence, providing a measure of potential transmission risk (Gujral et al. 2007). $$VI\hspace{0.17em}=\hspace{0.17em}{\mu }_{\mathit{mosquitoespertrapnight}}\times \mathit{MIR}$$

We conducted statistical and graphical analyses for this study using Tableau (v2024.1) and Python (v3.13). To further assess the relationship between cumulative precipitation and average temperature with William’s mean mosquito abundance each year, we performed a correlation study by calculating Spearman’s correlation coefficient and P-value. To account for time-lagged effects, lags in abundance of 0-5 wk were assessed based on the mosquito life cycle and previous findings correlating mosquito abundance to environmental factors over these periods of time (Buckner et al. 2011, Gardner et al. 2012, Karki et al. 2016) (Table S1).

Species diversity and abundance

A total of 4,523 female mosquitoes were collected and identified over the course of the season in 2023 and 17,532 in 2024. Of these, Cx. pipiens represented 3,882 and 16,244 of all mosquitoes collected in 2023 and 2024, respectively (Figs. 1 and 2, left). Each year, ten other species were collected (Figs. 1 and 2, right). The greatest abundance of non-Culex mosquitoes was Aedes albopictus (Skuse) (48% and 56% of other species in 2023 and 2024, respectively) followed by Ae. japonicus japonicus (Theobald) (approximately 27% in both years) (Figs. 1 and 2).

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Environmental drivers of mosquito dynamics

Weather conditions were incorporated into our analysis to identify mosquito population trends and environmental factors driving vector proliferation. In 2023, temperatures gradually rose throughout the season, peaked during EpiWk 36, and dropped off at the tail end of the season (Fig. 3). Conversely, temperatures peaked early in the season in 2024, remained high, then gradually decreased across the season (Fig. 4). Rainfall displayed somewhat similar patterns between years, with little precipitation in the wk leading up to the surveillance period, peak rainfall during midseason, a dry mid-to-late summer, and rainfall picking back up in late September (Figs. 3 and 4). Two wk early in the season during 2023 (EpiWk 25–26) showcased substantial cumulative precipitation (Fig. 3). Preliminary observations seemed to reveal warmer weather and/or increased precipitation occasionally coincided with subsequent spikes in Cx. pipiens abundance (Figs. 3 and 4). Abundance peaked during EpiWk 28 in 2023 (Fig. 3) and EpiWk 27 in 2024 (Fig. 4). Less extreme peaks were observed at EpiWk 30 and 35 in 2023 (Fig. 3), and EpiWk 35 and 36 in 2024 (Fig. 4). The lowest abundance was observed in EpiWk 39 in 2023 (Fig. 3) and EpiWeek 26 in 2024 (Fig. 4), preceded by some of the lowest temperatures and cumulative rainfall of both years. Notably, results for EpiWk 26 in 2024 may have been skewed due to multiple trap malfunctions, allowing for only 5 successfully functioning traps out of 12. To further assess these relationships, we examined correlations between weather and mosquito abundance in both years (Table S1). A negative correlation (r = −0.510, P = 0.009) was found between temperature and abundance with a 5-wk lag in 2023. Moderately strong positive correlations reached r = 0.473 for abundance vs. precipitation with a 3-wk lag, and r = 0.577 for abundance vs. temperature with a 0-wk lag, both in 2023. However, these results were not statistically significant.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

West Nile virus mosquito infection rate

Molecular arboviral screening revealed trends in WNV infected pools over the mosquito seasons. A total of 211 pools were tested for pathogens in 2023 and 824 pools in 2024. West Nile virus was detected in 35 pools of Cx. pipiens in 2023 and 183 pools in 2024, and 1 pool of Anopheles punctipennis (Say) in 2024. No other pathogens were detected. The VI peaked during EpiWk 29 in 2023 and EpiWk 30–31 in 2024 (Fig. 5). Minimum infection rates followed suit in 2024, peaking in EpiWk 30–31. The peak MIR occurred later in 2023 (EpiWk 33, Fig. 5).

View Figure

• Download as Powerpoint
• Download Figure

The IDX as a comprehensive surveillance tool

Integrating the IDX into WNV surveillance improved vector identification and streamlined data tracking. Traditional methods, such as morphological inspection or using dichotomous keys, are time-consuming and error prone, with variability between labs and individuals (Jourdain et al. 2018, Goodwin et al. 2021). The IDX addressed these challenges by providing rapid batch species identification, allowing entomologists to maintain sample quality and reduce the time between collection and arboviral testing—an important metric for WNV forecasting (DeFelice et al. 2019). It also served as a training tool for interns, providing high-quality reference images and helping seasonal technicians and early career entomologists identify damaged, rare, or unknown specimens. Supporting this, a study by Placer County Mosquito and Vector Control District confirmed that the IDX outperformed seasonal technicians with limited experience (7–22 months) in both speed and accuracy (Hubble et al., personal communication, pending publication). This study supports the notion that the IDX may help decrease time to identification when used by early career entomologists or seasonal technicians with limited experience. In this project for Anne Arundel County, the IDX functioned primarily as a confirmatory identification tool and digital logging platform, improving data collection, reporting, and reducing labor costs. Further, the IDX is a scalable tool with a workflow applicable to many arboviral and vector surveillance efforts. Although WNV was the only arbovirus detected, the IDX’s rapid vector identification capacity is relevant for pathogens like the emerging Oropouche virus, dengue virus, and malaria parasite, especially in light of recent locally acquired malaria cases in Maryland (Duwell et al. 2023, Guagliardo et al. 2024, Wheaton et al. 2025). With rising outbreak risks and invasive vectors, early detection through the IDX is essential for timely control.

Subsequent to exploring mosquito diversity and abundance, effective reporting and communication of the data become essential. The IDX reports dashboard helps MCOs and public health agencies to visualize and share geographical, chronological, and taxonomical vector information. By incorporating these reports into health communications or databases, agencies may gain a “bigger picture” view of WNV surveillance, allowing for enhanced and refined mitigation efforts.

Integration of environmental data

Through weekly surveillance efforts during the summers of 2023 and 2024 in Anne Arundel County, we tracked Cx. pipiens abundance at 12 sites, gaining critical insights into the population dynamics of one of the primary vectors for WNV. Continuous data collection enabled identification of seasonal mosquito abundance peaks and facilitated the delineation of hotspots and immediate vector control interventions in surrounding areas. Interestingly, strong correlations between weather variables and mosquito abundance were not found, with the exception of a negative correlation between temperature and abundance at a 5-wk lag in 2023 (r = −0.510, P = 0.009), showing a possible delayed effect of temperature on mosquito abundance. Other significant relationships between weather variables and abundance (Table S1) were observed, yet the lack of strong associations suggests site-specific weather events may produce differing mosquito outcomes. For instance, higher temperatures may accelerate breeding cycles and shorten larval development periods and light rainfall can create breeding habitats while heavy rain may wash them out (Geery and Holub 1989, Koenraadt and Harrington 2008, Ruiz et al. 2010). Regardless, with many collection sites close to water treatment plants, aquatic habitat availability may not be limited to major precipitation events alone. Supporting this, our correlation study showed weak links between environmental factors and mean abundance, suggesting that mosquitoes may not immediately respond to microclimatic events. Instead, more constant and larger aquatic grounds might play a greater role in abundance, and our use of limited weather data and geometric means across traps may have further diluted site-specific relationships. Notably, two trap sites were changed between the 2023 and 2024 collection seasons. To assess this as a possible confounding factor to our study, we analyzed site level contributions to overall abundance and diversity (Figs. S9 and 10). In 2024, new sites contributed only 11.3% to overall abundance (versus 9% from replaced sites in 2023), suggesting a minimal impact as 2024 abundance was nearly 4-times that of 2023. Most sites common to both years doubled in abundance from 2023 to 2024, pointing towards external drivers rather than site differences. In both years, Cx. pipiens was the most dominant species at all sites, comprising over 80% of the collections in new sites in 2024 (Fig. S10) and one replaced site in 2023, and 56% of the second replaced site in 2023 (Fig. S9). The remaining species—Ae. albopictus, Ae. japonicus, and Cx. Erraticus (Dyar and Knab)—were similarly distributed, indicating comparable mosquito ecology and diversity across replaced sites. Furthermore, sporadic interference from animals, people, or weather compromised trap integrity and reduced catches in 2023 and 2024. While our data represent a comprehensive analysis of Anne Arundel County vector abundance, such limitations should be considered when interpreting results. Still, in forthcoming mosquito seasons, further analyzing weekly weather patterns, potentially at a more granular, hyperlocal scale through weather stations at each trap, may help track and predict concentrated mosquito surges in the following week, ultimately to guide preemptive control measures.

Refining our assessment of the vector species composition in the region will help us understand disease dynamics over time. Some mosquitoes, their associated pathogens, and vectorial capacity are expected to proliferate with temperature increase, while others are predicted to dwindle (Shocket et al. 2020, Brüssow and Figuerola 2025). This variability, coupled with Anne Arundel County’s situation within a mid-Atlantic temperate region, warrants the continued evaluation of the intersection of weather and WNV cases over time. Future work could adopt climate-based models (Gong et al. 2010) to predict vector abundance and disease risk. Continued tracking of Culex abundance and long-term surveillance data will support modeling climate change impacts on mosquito populations and WNV transmission.

Public health implications

Our mosquito abundance, MIR, and VI data from EpiWk 26–39 (July-September) in 2023-2024 offers insights for public health decision-making. In 2023, abundance remained low and stable with a slight peak around EpiWk 28, while MIR rose gradually, peaking near EpiWk 33. Their weak correlation (r = 0.37, P = 0.16) suggests other influences on infection dynamics, such as pathogen introduction timing, mosquito age structure, or environmental factors (McMillan et al. 2023). In 2024, abundance peaked earlier (EpiWk 27), with MIR closely following (r = 0.49, P = 0.053), indicating a more synchronized transmission cycle. This mirrors McMillan et al. (2018), who found WNV amplification was strongest when Cx. quinquefasciatus overlapped with avian breeding under favorable environmental conditions. Humphreys et al. (2021) similarly showed that WNV risk depends on complexities between climate, host communities, and demographics, not vector abundance alone. Together, these findings under the need to pair abundance and infection data with ecological and demographic context for timely, targeted public health interventions.

Although similar climate and localities suggested comparable trends, 2024 saw a nearly 4-fold rise in mosquito abundance and arboviral pool submissions, and a 5-fold increase in WNV-positive pools compared to 2023. Species diversity remained largely consistent. However, limited surveillance (one trap-night/week at only 12 sites across 588 square miles) cannot be generalized to the broader region. Abundance variability may reflect factors such as trap performance, differences in avian populations, aquatic habitat availability, technician experience, or refined and improved surveillance protocols, though these remain speculative. Tracking with our findings of increased mosquito and WNV activity, Maryland reported 13 human WNV cases in 2023 and 23 in 2024 (CDC 2025b), though no county-level data were available. Overall, these findings emphasize the value of combining abundance and pathogen data to better assess human risk.

Surveillance data helped MDA identify hotspots and deploy targeted control efforts. In response to WNV detection (at least two positive pools), MDA implemented ultra-low volume permethrin spraying within 1.2 km of collection sites each year, with larvicides (granules or briquets) deployed in 2024. However, impacts of abatement efforts were not immediately reflected in surveillance results—likely due to timing, environmental conditions, and critically, the spatial and temporal disconnect between trapping and treatment. As shown by McMillan et al. (2023), mosquito abundance synchronized within 10-20 km, but arbovirus detections rarely synced beyond 5 km. This discrepancy highlights the need to continue pathogen testing while rethinking traps and control zone placement for better responsiveness to dynamic transmission.

Overall, this surveillance project tracked seasonal and spatial mosquito trends, confirming Cx. pipiens as the key WNV vector. Local Cx. pipiens picked up the virus through avian feeding early in Maryland’s mosquito season, depicted by peak vector index in July. Transmission to humans and subsequent confirmed cases shortly followed in August (MDH 2024). Our results also convey that the IDX supported timely identification and robust data curation, helping MDA optimize abatement strategies, reduce WNV risk, and promote public health outcomes in Anne Arundel County. Ultimately, this work underscores the potential of smart surveillance platforms like the IDX to enhance vector control precision and outbreak preparedness to protect community health through faster, data-driven decisions.