Drought frequency projection using regional climate scenarios reconstructed by seasonal artificial neural network model

作者：Lee J H; Moon S J; Kang B S^*

来源：Journal of Water and Climate Change, 2014, 5(4): 578-592.

DOI：10.2166/wcc.2014.130

• artificial neural network
• climate change
• drought projection
• SPI
• statistical downscaling

The climate change impacts on drought in the Korean peninsula were projected using Global Climate Model (GCM) output reconstructed regionally by an artificial neural network (ANN) model. The reconstructed model outputs were subsequently used as an input to project drought severity evaluated by Standard Precipitation Index (SPI). The original GCM output corresponds to the CGCM3.1/T63 under the 20C3M reference scenario and the IPCC A1B, A2 and B1 projection scenarios. Because in general GCM shows limitation in capturing typhoon generation occurred at sub-grid scale, the training and validation of the ANN model utilized a precipitation data set with typhoon-generated rainfall eliminated for enhancing the ANN%26apos;s computational performance. The non-stationarity characteristics of SPI was examined using the Mann-Kendall test. The projection was implemented for the near future period (2011-2040), mid-term (2041-2070) and long-term (2071-2100) future periods. The results indicated mitigated drought severity under all scenarios in terms of frequency, magnitude and drought spells even for the mildest B1 scenario. The SDF (severity-duration-frequency) curves illustrate the common patterns of alleviated drought severity for most future scenarios and elongated drought duration. The reconstructed GCM projection recovers the underestimated precipitation and provided more realistic drought projection even though there would be still uncertainties of spatial and temporal variability.

• 出版日期2014