A Probabilistic Analog-Year Approach for Drought Prediction in California
Tuesday, April 21, 2015
Shahrbanou Madadgar, University of California Irvine, Irvine, CA, United States and Amir AghaKouchak, University of California Irvine, Civil and Environmental Engineering, Irvine, CA, United States
Abstract:
Precipitation forecasting is a major challenge in California, which is often affected by multi-year droughts. Climate oscillation indicators have shown limited skills for precipitation forecasting in the southwestern US, mainly because the degree of association between precipitation and those indices is not strong. This study outlines a framework for improving drought prediction using an analog year concept and based on a conditional probabilistic approach. To implement precipitation forecasting during the wet season in California (Oct-Apr), we used the three major indicators of climate variability over the southern US; namely Southern Oscillation Index (SOI), Pacific Decadal Oscillation (PDO), and Multivariate ENSO Index (MEI). These climate indices are embedded in a conditional probabilistic model to estimate the predictive uncertainty of precipitation with different forecast initiation and lead times (monthly to intra-annual scales). The proposed probabilistic model is built on copula functions that can be used to describe the dependence between multiple variables. In this approach, the near-past meteorology is also integrated to improve forecast results. Standardized Precipitation Index (SPI) is used as an indicator of the near-past drought conditions which carries the meteorological memory of the weather system. This study uses SPI to pick the percentile of probability distribution associated with deterministic forecast. The bias, correlation coefficient, and Kling-Gupta Efficiency (KGE) along with a few other measures are used to verify the accuracy of deterministic forecasts. Results are also tested for the terciles of precipitation distribution (Above Normal, Normal, Below Normal). On average, precipitation forecasts could capture above 75% of the observations in each tercile for all the initiation and lead times during the wet seasons. In summary, the proposed model showed high forecast skill in both probabilistic and deterministic viewpoints.
