Probabilistic Convective Initiation Nowcasting with Reduced Satellite - NWP Predictors over Iran
Autor: | Arastoo Biazar, Seyed Kazem Alavipanah, Maryam Gharaylou, Reza Khandan |
---|---|
Rok vydání: | 2018 |
Předmět: |
Atmospheric Science
010504 meteorology & atmospheric sciences Nowcasting 0208 environmental biotechnology 02 engineering and technology Numerical weather prediction 01 natural sciences 020801 environmental engineering Random forest Correlation Statistics Satellite False alarm Temporal difference learning 0105 earth and related environmental sciences Mathematics Parametric statistics |
Zdroj: | Asia-Pacific Journal of Atmospheric Sciences. 54:431-443 |
ISSN: | 1976-7951 1976-7633 |
DOI: | 10.1007/s13143-018-0009-z |
Popis: | Geostationary satellites are able to nowcast Convective Initiation (CI) for the next 0–6 h. Compared to using satellite predictors only, the incorporation of satellite and Numerical Weather Prediction (NWP) predictors can provide the possibility to reduce false alarm rates in 0–1:30 Convective Initiation Nowcasting (COIN). However, the correlation among these predictors not only can cause error in COIN, but also increases the runtime. In this study for the first time, all effective predictors in Satellite Convection Analysis and Tracking version 2 (SATCASTv2) and NWP were applied over Iran from 22nd March 2015 to 9th January 2016. In applying SATCASTv2 over Iran, it was necessary to make some modifications to the algorithm, such as removing case specific thresholds of satellite predictors and rearranging COIN predictors. Then, SATCASTv2 was tested and evaluated with both the full and reduced set of predictors. The results suggested that using fixed thresholds for temporal difference predictors could miss COIN in some cases. To investigate the possibility of improving computational efficiency, a dimension reduction was conducted by Factor Analysis (FA) and the number of predictors was reduced from 22 to 11. The NWP-satellite, reduced NWP-satellite, and satellite predictors were used as input in Random Forest (RF), as a parametric machine learning method, for COIN evaluation. The Combination of NWP model and satellite predictors had lower false alarm rates in contrast with satellite predictors. This is in agreement with previous studies. The results from statistical metrics showed that the reduced NWP-satellite predictors had comparable performance to the NWP-satellite predictors over study area, but decreased the run time by almost 50%. The results indicated that Convective Inhibition (CIN) was the most significant predictor when the reduced set of predictors was used. |
Databáze: | OpenAIRE |
Externí odkaz: |