Item

Abstract

This study employs a two-way unbalanced analysis of variance (ANOVA) to investigate the factors responsible for the variance of subseasonal to seasonal (S2S) scale monthly mean 10 m wind speed forecasts in the Copernicus Climate Change Service (C3S) multi-model ensemble (MME). The MME consists of the following six ocean-atmosphere coupled models: SEAS5, GCFS 2.0, Météo-France's System 6 (MF-6), CFSv2, GloSea5 GC2-LI, and SPS3. The study regions are the seven homogenous climate zones of India. The time-period spans from 1994-2016 and ANOVA is applied individually to 1, 2, 3, 4, and 5 months lead time forecasts of all twelve months of the year i.e. January through December. Results show that across all lead times, the highest fraction of the MME variance is due to the inter-model biases. The variance due to the residual error term is the second highest contributor to MME variance, followed by the near-similar contributions from variances due to the time-varying behaviour of the models and the differences due to initial conditions. This behaviour is consistent across all the lead times. Overall, contributions to MME variance from the errors are low during the monsoon months. The MME 10 m wind speed forecasts all round the year show high potential predictability, which drops with increasing lead time. This decrease in potential predictability with lead time can be explained by the increase in the fraction of MME variance due to residual error.