Study on Integrated Typhoon Long-Distance and Short-Distance Rainfall Forecast Model Based on Enso Scenarios

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

MOST103-2111-M464-001

Translated Name/計畫中文名

聖嬰南方振盪情境下遠距與近距颱風降水預測模式之研究

Project Coordinator/計畫主持人

Chih-Chiang Wei

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Marine Environmental Informatics

Website

https://www.grb.gov.tw/search/planDetail?id=8323663

Year

2014

Start date/計畫起

01-08-2014

Expected Completion/計畫迄

01-07-2015

Bugetid/研究經費

546千元

ResearchField/研究領域

大氣科學
防災工程

台灣位於西北太平洋颱風路徑上，同時受到氣候異常因素影響（如聖嬰現象），常導致異常降水事件的發生。本計畫的目的在發展一套聖嬰南方振盪（El Niño/Southern Oscillation, ENSO）情境下颱風即時降水預測模式，此模式將提供熱帶氣旋離台灣遠距（Long distance）時逐日預估颱風在颱風預測路徑下，未來可能影響台灣某區域之總降雨量；而當颱風近距離（Short distance）影響時，即逐小時預測降水量。本計畫將以兩年為一期完成研究工作。第一年將建立近距即時降雨預測模式（Short-distance typhoon rainfall forecast model, STRF），根據 100年國科會計畫（Wei, 2012）建模過程中輸入資料採用地面測站氣候資料以及颱風氣象資料；而在 102年國科會計畫（Wei and Roan, 2012; Wei, 2013）時則採用衛星微波頻道之亮度溫度，然而衛星頻道非逐時刻通過台灣上空，因此在預報上有其缺點。本計畫將改進上述相關研究，探討雷達回波結合地面氣候資料以及颱風氣象資料之可行性，並研擬可增進降水量預測能力之最佳輸入資料。本計畫將採用新近預測技術（如小波支援向量機、調適性網路模糊推論系統）建立 STRF 預測模式。本計畫在第二年建立遠距總降雨量預測模式（Long-distance total rainfall forecast model, LTRF）。本計畫將設定幾種不同氣候情境，包括：（1）聖嬰現象發生與否，即聖嬰年（El Niño）、反聖嬰年（La Niña）、正常年；（2）不分類。本計畫在遠距離預測時，將根據 Joint Typhoon Warning Center (JTWC)之熱帶氣旋預測路徑下建立 LTRF模式。本計畫將利用衛星微波資料以及海平面汽壓（Sea level pressure, SLP）和海表面溫度（Sea surface temperature, SST）反演相關氣象參數，包括散射指數（Scattering index, SI）、修正極化溫度（Polarization-corrected temperature, PCT）、聖嬰指數（SOI）、Niño 3.4以及 SST指數等。本計畫將採用新近分類技術之資料探勘法（如貝氏網路和決策樹）建立 LTRF預測模式。本計畫完成上述 LTRF和 STRF預測模式時，將進行整合工作。本計畫為應用於實務面上，因此近、遠距離之預測起始時刻定義為：（1）遠距離時，將以 JTWC 發布熱帶氣旋發生並且預測未來行經路徑可能影響台灣的時間點，即開始利用 LTRF模式進行逐日預測颱風通過期間某地區之總降水量；（2）近距離時，將以中央氣象局（Central Weather Bureau, CWB）所發布之海上陸上颱風警報單之時間，亦即當陸上颱風警報發布時，即開始利用 STRF模式預測小時降水量（1至 6小時）。本計畫在整合 LTRF和 STRF模式時，將再評估在近距預測時氣候變異對台灣地區所造成之影響。一般而言，聖嬰現象在台灣地區並不顯著，因此本計畫將初步再分析兩種不同氣候情境，包括：（1）聖嬰年、非聖嬰年；（2）不分類，完成分析後並予討論。另外，整合後的模式將模擬數場颱風事件，以驗證模式之可行性。本計畫將以石門水庫集水區或其他區域為示範例。 Taiwan lies in the main track of western North Pacific typhoons. The typhoon season is a climatic characteristics in Taiwan. The ENSO (El Niño/Southern Oscillation) is the major cause of climate variability on seasonal to interannual time scales. It can be characterized by an interannual cooling (La Niña) and warming (El Niño) of the eastern equatorial Pacific. Though it originates in the tropical Pacific, it has an impact on weather and climate globally. On average, 3.5 typhoons pass near or over Taiwan annually. These invaded typhoons lead to considerable economic losses and casualties caused by extreme precipitations which might be affected by ENSO effects. As a result, developing a practice model is necessary for identifying the total rainfalls prior to typhoon invasions and real-time forecast of hourly rainfalls in Taiwan. The purpose of this study is to forecast typhoon rainfalls over Taiwan based on climate variability. A two-year project is proposed here. For the first year, the short-distance typhoon rainfall forecast model (STRF) will be modeled. According to Wei (2012), the model inputs comprise the weather data measured from the automatic meteorological gauges, and the climatologic characteristics of typhoons issued by Central Weather Bureau (CWB). However, rainfall is measured by ground rain gauges, a simple and convenient method, but their spatial representations are exceedingly poor. Wei and Roan (2012) and Wei (2013) used the microwave data sources originated from the Special Sensor Microwave/Imager (SSM/I) as model inputs. However, the disadvantage is SSM/I sensors cannot provide hourly microwave data which limits the ability of rainfall predictions. To improve this problem, this study will employ the reflectivity measurements from radar sensor instructions, combined with the surface weather data and the typhoon climatological data. The study will uses the artificial intelligence forecast techniques for modeling STFR, e.g., wavelet-kernel support vector machine for regressions (WSVR) and adaptive network-based fuzzy inference system (ANFIS). In the second year, the long-distance total rainfall forecast model (LTRF) will be modeled. The different climate scenarios will be designed, namely {El Niño year, La Niña year, Normal year} and {All analyzed years}. Based on the TC tracks predicted by Joint Typhoon Warning Center (JTWC), LTRF will be modeled when TC is far from Taiwan. This study will receive various senor raw data, namely, brightness temperature (BT), sea level pressure (SLP), and sea surface temperature (SST). Then, retrieve the various parameters, namely scattering index (SI), polarization-corrected temperature (PCT), El Niño/Southern Oscillation (ENSO) based index (SOI), Niño 3.4, and SST index served as model inputs. The LTRF model will be constructed by data mining algorithms, e.g., the Bayesian networks (BN), classification and regression tree (CART), and C4.5. Finally, the proposed LTRF and STRF models will be integrated. For practicability, the initial forecasting time for both models is set. For long-distance situation, the initial forecasting time of LTRF is set at the TC affecting Taiwan based on the JTWC forecast tracks. For the short-distance situation, the initial forecasting time of STRF is set when typhoon land warning over Taiwan issued by CWB. Because the El Niño/Southern Oscillation (ENSO) phenomenon is insignificant, this study will preliminarily evaluate both climate scenarios for the short-distance situation, namely {El Niño year, non-El Niño year} and {all analyzed years}. Additionally, several typhoons which made landfall over Taiwan will be simulated and examined by using the integrated LTRF and STRF model. The presented methodology will be applied for rainfall forecasts at the watershed of Shihmen Reservoir in the upper stream of Tahan River.

Keyword(s)

颱風
降水
聖嬰南方振盪
預測
模式
Typhoon
precipitation
climate variability
ENSO
prediction
model

DSpace CRIS

Study on Integrated Typhoon Long-Distance and Short-Distance Rainfall Forecast Model Based on Enso Scenarios

Details

Description