West Nile virus (WNV; family Flaviviridae) has been an annual public health concern in the continental United States since its introduction in 1999. It is particularly difficult to determine when and where human infections will occur because WNV is highly focal and amplifies in bird-mosquito cycles with incidental spillover to humans. WNV is transmitted by female mosquitoes in the genus Culex, most commonly Cx. tarsalis, Cx. pipiens, and Cx. quinquefasciatus. The mosquitoes’ capacity to transmit WNV depends on the environment, particularly temperature, resulting in seasonal cycles with most human cases occurring between July and September. WNV prevention is predominantly through vector management, consisting of a combination of larval and adult control. Integrated vector management programs in California conduct surveillance to monitor mosquito abundance and mosquito infection prevalence as a means for targeting control strategies, but these estimates are not adjusted for biases that could be induced by short-term variation in weather. This dissertation investigates the complicated relationship between environmental factors, entomological surveillance observations, and human WNV disease.Chapter 1 focused on the relationship between entomological surveillance indicators and risk for human WNV disease. In particular, we evaluated the ability of the vector index (VI), the product of mosquito abundance and infection prevalence, to predict periods of above-average WNV incidence. We used receiver operating characteristic (ROC) curves to identify the VI threshold that maximized sensitivity and specificity of these predictions and found that these thresholds were highly dependent on the dominant vector species in an area and the trap type used for targeting surveillance. We also used statistical models with observed entomological surveillance and human disease data aggregated to different spatial scales to examine the effect of spatial scale on the ability of the VI to predict human WNV disease incidence. These results found cities to be the best balance between being small enough to be operationally relevant but large enough to have adequate predictive accuracy during high-risk periods. Chapter 2 considered short-term weather variability as a potential source of bias in entomological surveillance that may affect estimates of WNV disease risk. We collected mosquitoes and gathered weather data from 10 field sites equipped with devices to record mosquito counts, temperature, and wind speed every 15 minutes in the rice-growing region of northern California. We used the 15-minute mosquito data to estimate four outcomes for the primary WNV vector in the study area, Cx. tarsalis: the total overnight count, the onset time of evening host-seeking activity, the median time of nightly host-seeking activity, and the hour of peak host-seeking activity. We related each of these outcomes to the wind speed and temperature recorded at a range of times in the afternoon leading to the host-seeking night through statistical models and found both wind speed and temperature at 20:00, or just prior to the onset of host-seeking, to be the best predictors of all four outcomes. These predictable factors will help guide the timing of vector control applications to maximize their effects on the local vector population, thus reducing the overall risk of WNV transmission. Chapter 3 applied spatio-temporal predictive models that accounted for ecological factors and were capable of highlighting gaps in surveillance coverage for estimating risk of WNV transmission. We used a generalized additive model (GAM) to model the nonlinear weekly seasonal trends of Cx. tarsalis abundance in a range of land use types using trap count data for the years 2008-2020 collected by 20 Central Valley mosquito and vector control districts in California. Overall, the model captured strong seasonal patterns in abundance, modified by local land use and ephemeral spatio-temporal anomalies. Models maintained a similar predictive accuracy for out-of-sample data compared to that for the training data set. To determine the ability of the model to extrapolate from known surveillance locations, we then used the GAM to predict weekly Cx. tarsalis abundance at unmeasured locations across a 2.5-km grid of the Central Valley. We found that these predictions were most accurate when the nearest observed trap was within 2 km and one week prior. Taken together, these three chapters provide a basis for improving WNV risk estimation through entomological surveillance. These chapters will inform efforts to prevent human WNV disease through a more complete understanding of the link between WNV vector dynamics and the risk of human disease, improving the ability to predict risk and target mosquito control.