QIN JIANG

Dottoressa di ricerca

ciclo: XXXV



Titolo della tesi: Statistical/Climatic models to predict and project extreme precipitation events dominated by large-scale atmospheric circulation over the central-eastern China

The observable and happening global warming has posed non-negligible effects on regional extreme precipitation changes and increased the uncertainties when meteorologists predict and project such extremes. More importantly, floods, landslides, and waterlogging caused by extreme precipitation have catastrophic societal impacts and led to steep economic damages across the world, in particular over central-eastern China (CEC), which often suffers excessive precipitation due to the Meiyu-front and typhoon activities. There is mounting evidence that the anomaly atmospheric circulation systems and water vapor transport have a dominant role in triggering and maintaining the processes of regional extreme precipitation. In this Ph.D. thesis, the self-organizing map and event synchronization approaches were applied to cluster the large-scale atmospheric circulation and quantify the synchronization degree between circulation patterns (CPs) with extreme precipitation over CEC. Within the understanding of which CPs were associated with extreme precipitation events, and corresponding water vapor transport fields, a deep learning model that hybrid multilayer perceptron and convolutional neural networks (MLP-CNN) was proposed to achieve the binary predictions of extreme precipitation. The accuracy of MLP-CNN was compared with the independent predictions from MLP, CNN, and two machine learning models (random forest and support vector machine). Since the coarse spatial resolution of global circulation models and its biases in extremes estimations, to reduce the uncertainties in future precipitation projection, a new precipitation downscaling framework combination of ensemble-learning and nonhomogeneous hidden Markov model (Ensemble-NHMM) was developed. The main contents and relevant conclusions are enumerated as follows. 1) The dominant CPs over CEC were robustly identified based on the self-organizing map. From the period 1960–1989 to 1990–2015, the categories of identified CPs generally remain almost unchanged. Among these, the circulation patterns characterized by obvious positive anomalies of 500 hPa geopotential height over the Eastern Eurasia continent and negative values over the surrounding oceans are highly synchronized with extreme precipitation events. An obvious water vapor channel originating from the northern Indian Ocean can be observed for the representative CPs, response to the southwesterly airflow. Contribution analysis results indicate that the CP changes produce an increase in extreme precipitation frequency from 1960–1989 to 1990–2015. The 2–4 yr oscillation in the annual frequency of representative CPs are closely related to the phase of El Niño and Southern Oscillation (ENSO); while the 20–25 yr and 42–50 yr periodic oscillations are responses to the monthly mean sea surface temperature anomalies in the North Pacific and North Atlantic. 2) The regional extreme precipitation predictability based on the large-scale atmospheric circulation is investigated. Using the anomalous fields of two atmospheric variables, that is, geopotential height at 500 hPa and vertically integrated water vapor transport (IVT) as the predictors, the MLP-CNN model exhibit an overall accuracy of 86% in the classification task of extreme or non-extreme precipitation days. The MLP-CNN can correctly predict 91% (81%) of extreme precipitation days on the training (testing) sets and shows a fairly performance in different seasons. Evaluation metrics illustrate that the skill of MLP-CNN is higher than the independent predictions from MLP and CNN models, and two machine learning approaches (i.e., random forest and support vector machine). As the prediction time shifted to 1 to 15 days ahead, the performance of MLP-CNN tends to decrease, but attach importance to the 1-2 days’ advanced circulation anomalies could benefit for predicting the occurrence of extreme precipitation. 3) The accuracy of 10 global circulation model (GCM) control runs, in extreme precipitation and large-scale atmospheric circulation simulations were evaluated over the central-eastern China (CEC). Four indices were selected to measure the characteristics of extreme precipitation. Compared to the gauge-based daily precipitation analysis dataset over the Chinese mainland (CGDPA), all GCMs are difficult to accurately estimate the extreme precipitation amounts, with an underestimation exceeds 85%. Relative to the identified CPs of ERA5, GCMs can reflect most categories accurately, and MPI-ESM1-2-HR could be considered excellent according to its correctly capture the pattern labels. 4) To improve the precipitation estimates from global circulation models, a new precipitation downscaling framework was proposed. This framework comprises two techniques, that is, ensemble-learning model for multi-level precipitation probability predictions and NHMM downscaling. The ensemble learning model constructed by extreme gradient boosting (XGBoost) and random forest (RF), aims to predict the occurrence probabilities for the different levels of daily precipitation aggregated by multi-sites; and the precipitation downscaling is done using the predicted occurrence probabilities of daily precipitation events as predictors to NHMM. Statistical metrics show that the Ensemble-NHMM downscaled results are match best to the observations in precipitation variabilities and extreme precipitation simulations. For model validation, the downscaled precipitation from a GCM model (MPI-ESM1-2-HR) observed the precipitation climatology well. 5) Three climate scenarios with different Shared Socioeconomic Pathways were selected to explore the future precipitation projection. The downscaled high-resolution projections indicate that the CEC will receive more precipitation in the future, by ~30% through the 2075–2100 period. Compared to the nearly 26 years (1990–2015), significantly increases in the frequency and amount of extreme precipitation by 21.9–48.1% and 12.3–38.3% respectively by the late century when under the worst emission scenario. In particular, the south CEC region is projected to suffer more extreme precipitation than the north. Physical mechanism analysis shows that climate warming would increase the probability of stronger water vapor transport over CEC. More wet weather states due to the enhanced water vapor transport and the strengthen pressure gradient are the possible mechanisms for the increased precipitation. Keywords: Extreme precipitation, circulation pattern, climate change, precipitation downscaling, projections

Produzione scientifica

11573/1688825 - 2023 - A stacked ensemble learning and non‐homogeneous hidden Markov model for daily precipitation downscaling and projection
Jiang, Qin; Cioffi, Francesco; Conticello, Federico Rosario; Giannini, Mario; Telesca, Vito; Wang, Jun - 01a Articolo in rivista
rivista: HYDROLOGICAL PROCESSES (John Wiley & Sons Limited:1 Oldlands Way, Bognor Regis, P022 9SA United Kingdom:011 44 1243 779777, EMAIL: cs-journals@wiley.co.uk, INTERNET: http://www.wiley.co.uk, Fax: 011 44 1243 843232) pp. - - issn: 0885-6087 - wos: (0) - scopus: 2-s2.0-85171384559 (0)

11573/1671545 - 2022 - Analysis of changes in large-scale circulation patterns driving extreme precipitation events over the central-eastern China
Jiang, Qin; Cioffi, Francesco; Giannini, Mario; Wang, Jun; Li, Weiyue - 01a Articolo in rivista
rivista: INTERNATIONAL JOURNAL OF CLIMATOLOGY (Chichester : Royal Meteorological Society : John Wiley & Sons, Ltd.) pp. 519-537 - issn: 1097-0088 - wos: WOS:000825912500001 (0) - scopus: 2-s2.0-85134030395 (2)

11573/1678435 - 2021 - Evaluation of the {ERA}5 reanalysis precipitation dataset over Chinese Mainland
Jiang, Qin; Li, Weiyue; Fan, Zedong; He, Xiaogang; Sun, Weiwei; Chen, Sheng; Wen, Jiahong; Gao, Jun; Wang, Jun - 01a Articolo in rivista
rivista: JOURNAL OF HYDROLOGY (Tokyo ; Oxford ; New York ; Lausanne ; Shannon ; Amsterdam : North-Holland : Elsevier) pp. 125660- - issn: 0022-1694 - wos: WOS:000641592600009 (106) - scopus: 2-s2.0-85095593984 (113)

© Università degli Studi di Roma "La Sapienza" - Piazzale Aldo Moro 5, 00185 Roma