A study of atmospheric gradient for the analysis of weather events and prediction of rainfall in the tropical region

• Main Author: Naha Biswas, Anik
• Other Authors: Lee Yee Hui
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/178353
• Institution: Nanyang Technological University

Water vapor is an essential parameter in the troposphere which is involved in several atmospheric processes including cloud formation, convection and precipitation. The water vapor variability in the lower troposphere is a major indicator for the occurrence of rainfall and precipitable water vapor (PWV) is the measure of the total moisture content in the troposphere. PWV is therefore a key parameter for rainfall forecasting. At present, Global Navigational Satellite System (GNSS) can be used to measure the PWV values and by virtue of its improved spatiotemporal resolution and operational capability in all weather conditions, it has advantages over radiosonde and water vapor radiometer (WVR). Several literatures have conducted the study to explore the relation between GPS-derived PWV and rainfall. PWV has been thoroughly studied for some moderate to severe rainfall events and a GPS-derived PWV-based algorithm has been proposed for predicting the rainfall for tropical regions. However, PWV alone is insufficient to forecast the precipitation accurately as the false alarm rate is significantly high. Also, PWV values in the atmosphere generally start to rise 2 to 3 hours before the rainfall in the tropical region and PWV commonly attains its maximum value just 15-30 minutes prior to a rain event. Therefore, it has limited suitability as a parameter for long-term weather prediction. Moreover, convective precipitation cannot be predicted effectively with precipitable water vapor. There are several other tropospheric parameters that need to be considered to enhance the accuracy of rainfall prediction algorithms. Until now the atmospheric gradient has mostly been studied for precise point positioning rather than weather forecasting. The tropospheric gradient is another crucial parameter that has a significant contribution to analyzing the development of a weather system. Recent work has demonstrated that the study of gradient provides detailed atmospheric information as it gradually changes its direction with the development of a convective system. Therefore, the tropospheric gradient is an essential parameter that needs to be considered for rainfall prediction and analysis of weather events for a larger region. The primary aim of our study is to examine the behavioral pattern of different atmospheric parameters (horizontal gradient, post-fit residual) along with GPS-derived PWV values. We have investigated the behavioral pattern of the atmospheric resultant gradient before the occurrence of rainfall events in Singapore. We have observed that the gradient direction changes abruptly (either changes its sign or differs by more than 10 degrees) within 5-6 hours prior to the rainfall event. Also, we have found that atmospheric gradient orientation is mostly aligned with wind directions. Moreover, we have analyzed the seasonal variation of the gradient in Singapore as the wind changes its direction in different monsoon seasons. The rise in the values of gradient magnitude and gradient slope beyond a certain threshold before rainfall makes them potential features for rainfall prediction. By virtue of lessening the false alarm with a higher true detection rate, the atmospheric residual has been introduced with the tropospheric gradient in the rainfall forecasting algorithm. Our results are combined to derive a predictive algorithm that aids the forecast of the rain event within a 6-hour window with more than 85% true detection and a satisfactory false alarm rate of 35%. The proposed algorithm provides a relatively long-term prediction for the next 6 hours with a significant improvement in false alarm and true detection rate as well. Furthermore, we have perused the behavior of atmospheric gradient to monitor the weather events for a larger region instead of an individual station. The tropical region in southeast Asia has been mostly considered for conducting our study to explore the relation of the atmospheric gradient with rainfall for an area. The atmospheric gradient in the surrounding region converges at the location of precipitation which is also an indication of atmospheric convergence leading to a vertical updraft of warm air. The gradient magnitude exhibits negligible shift in values between rainy and non-rainy hours while forecasting precipitation over a region. The weather system can be forecast by the change in the gradient flux for a larger area almost 12 hours preceding the rain event affecting the particular region. The temporal and spatial variation of gradient convergence and flux have been studied rigorously for predicting rain events over an area in the tropical region. Both the gradient parameters show an increasing trend prior to a significant weather event over a larger region in the tropical climatic zone. The north and east components of atmospheric gradient have also been investigated in relation to rainfall occurring over a larger region. The total gradient follows a similar trend as that of its wet component which shows a converging nature at the region of precipitation while the hydrostatic part exhibits a difference in magnitude between two different weather scenarios. Based on the statistical analysis of gradient orientation preceding a rain event, we have designed an algorithm using gradient convergence and flux for forecasting the rainfall over a region. In this study, we have incorporated precipitable water vapor alongside the atmospheric gradient in order to augment the accuracy of the forecasting algorithm with a 6-hour prediction window as PWV experiences an increase in its value before precipitation over an area. The single-layer dual-parameter algorithm with gradient convergence and PWV provides a high true detection rate of 90.4% alongside a substantially low false alarm rate of 21.9%. We can achieve almost similar accuracy with gradient flux as well in combination with PWV for predicting rain over a region in next quarter of a day. Furthermore, a two-layer algorithm has also been developed with both PWV and atmospheric gradient which precisely predicts the region to be afflicted by rainfall in the next 6 hours. Deep learning model has also been implemented to forecast the rain events for the following 6 hours with various atmospheric parameters which attains a substantial prediction accuracy of around 80% for tropical regions.