Field observation & numerical simulation of nocturnal atmospheric boundary layer: Fog prediction over the kempegowda international airport, Bengaluru Using ML

Abstract

Fog, usually considered a calm-weather phenomenon, impairs horizontal visibility near the ground and badly affects many sectors of society like transportation, power, agriculture, and communication, leading to huge socio-economic losses. We have investigated fog occurrence at the Kempegowda International Airport, Bengaluru (KIAB), which includes extensive field campaigns during 2020-24 and numerical simulations using models like the deterministic and ensemble Weather Research and Forecasting (WRF) model, single-column model, as well as machine learning (ML) models. We have used observations from field experiments and the Meteorological Aerodrome Report (METAR) to understand fog occurrence, its long-term trends, and the effects of different meteorological factors on its occurrence and development, whereas numerical simulations have been used to improve visibility parameterization, bias corrections, and to develop an effective operational forecasting model for fog prediction at the KIAB. We have analyzed 16 years (2008-24) of the METAR dataset and shown that fog at the KIAB usually occurs around sunrise from October to February and holds significant intra- and inter-seasonal variability. December faces the highest fog hours and days, followed by January and November. Although mist is mostly ignored in numerical weather prediction (NWP) model, its occurrence in total hours is 6–7 times higher than fog occurrence in the fog season and can play an important role in preconditioning of fog.To identify the favorable conditions for fog occurrence at the KIAB, we have compared many parameters between foggy and non-foggy days of the fog season during 2020-24. Near ground cooling and middle-layer warming are some of the favored conditions for fog occurrence. Contrary to the assumption that fog usually appears in high-humidity environments near the ground, fog in December occurs in dry environments where dryness can be observed from the surface to a few kilometers above the ground. Other favorable conditions include middle-level drying, anticyclonic conditions (i.e., clear-sky conditions), a relatively unstable boundary layer in the day, and a stable layer at night. Along with the fog study, we also elucidate the impact of aerosols/fog on longwave cooling of the ABL using field experiments and simulations that integrate aerosol interaction in a radiation-conduction model. Field observations indicate that under calm and clear-sky conditions, the evening transition typically results in a distinct vertical thermal structure called the Lifted Temperature Minimum (LTM). We observe that after sunset, the prevailing profile near the surface is the LTM profile. Additionally, the occurrence of LTM increases with a decrease in parameters like downward and upward longwave flux, soil sensible heat flux, wind speed, and turbulent kinetic energy measured at two meters above ground level (AGL). In such scenarios, the intensity of LTM profiles is primarily governed by aerosol-induced longwave heating rate (LHR) within the surface layer. Furthermore, the presence of clouds leads to increased downward flux, causing the disappearance of LTM, whereas shallow fog can enhance LTM intensity, as observed in both field observations and simulations.Usually, prevailing radiation models underestimate aerosol-induced longwave heating rate (LHR) by an order compared to actual field observations. We attribute this difference to aerosol induced radiation divergence. We show that the impact of aerosol-induced LHR extends hundreds of meters into the inversion layer, affecting temperature profiles and potentially influencing processes such as fog formation. As the fog layer develops, LHR strengthens at its upper boundary, however, we highlight the difficulty in detecting this cooling using remote instruments such as microwave radiometer. A comparison between WRF output, input, and observations has revealed that the WRF model shows significant biases of varying magnitude at vertical levels, which are primarily introduced via initial and boundary conditions. We attempt to reduce biases in temperature and humidity profiles from the WRF model at 7 hours of lead time using the ML model and show that total biases in temperature and humidity profiles, including all levels up to 10 km, reduce to 0.49 K and 0.63 g kg−3 from 1.06 K and 1.26 g kg−3 after applying ML-based corrections, respectively. Based on extensive observations and analysis, we have performed visibility and Runway Visual Range (RVR) parameterization using the WRF model and METAR. Further, determinis tic, ensemble, and ML-based approaches have been tested to get an operational fog forecasting model at KIAB. Analysis shows that the ML-based model, configured and tested using satellite and WRF datasets, outperformed on many metrics. Concurrently, we have performed oper ational fog forecasting to predict visibility at KIAB for three fog seasons, where accuracy in predicting foggy/non-foggy days was 69% in 2021-22, 72.4% in 2022-23, and 74.1% in 2023-24.