Statistical downscaling to predict drought events using high resolution satelite based geopotential data

9th Annual Basic Science International Conference 2019

自然科学（总论）

Kuswanto, H.^1 ; Yuliatin, I.L.^1 ; Khoiri, H.A.^1

Departement of Statistics, Faculty of Mathematics, Computing and Data Science, Institut Teknologi Sepuluh Nopember (ITS), Indonesia^1

关键词: Degree of uncertainty;    Geo-potential heights;    High resolution data;    High-dimensional dataset;    Meteorological station;    Principle component analysis;    Standardized precipitation index;    Statistical downscaling;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/546/5/052040/pdf
DOI  :  10.1088/1757-899X/546/5/052040

学科分类：自然科学（综合）

来源: IOP

【 摘 要 】

Drought prediction is a very challenging work due to high degree of uncertainty in the climate system. Geopotential height has been investigated as one of the dominant variables that can be used to predict drought events. This paper discussed the use of high resolution satelite based (reanalysis) data as the predictor of drought events, resulting on a high dimensional dataset. To deal with this, dimension reduction has been carried out by using Principle Component Analysis (PCA), prior to the development of the downscaling models which incorporate the past SPI (Standardized Precipitation Index) combined with the geopotential height at some specific atmosperic levels i.e. 500hPa, 850hPa, 900hPa, 975 hPa and 1000hPa. The SPI, as the drought risk measure is derived from the reduced dimension of precipitation data observed from the corresponding meteorological stations, while the geopotential height is reduced from gridded high resolution data. The downscaling process found the best model to predict the drought risk with various degree of R-squares. The outsample validation showed that predicting drought using SPI3 (three month period SPI) with geopotential at the 900hPa level as the predictor outperforms the others with R-square reaching 77%.