• English 
  • Spanish 

Process-conditioned bias correction for seasonal forecasting: A case-study with ENSO in Peru

An important limitation of bias correction (BC) methods is that they can introduce arbitrary temporal changes which can deteriorate the interannual variability of the raw predictions. To partially alleviate these problems, Maraun et al. (2017) advocated the development of process-informed BC methods, combining the statistical modeling with the knowledge about the relevant processes for the problem under study. However, the application of this type of methods remains unexplored yet for the case of seasonal forecasts. Thus, this work assesses the suitability of a first simple attempt for processconditioned BC in the context of seasonal forecasting. To do this, we focus on the northwestern part of Peru and bias correct one- and four-month lead seasonal predictions of boreal winter precipitation from the ECMWF System4. In order to include information about the underlying large-scale circulation which may help to discriminate between precipitation affected by different processes, we introduce here an empirical quantile-quantile mapping which runs conditioned on the state of the Southern Oscillation Index (SOI), which is accurately predicted by System4 and is known to affect the local climate. Our results show that a standard implementation in which the quantile-quantile mapping is directly applied over the entire period of study (1981-2010) broadly preserves the temporal structure of the raw model precipitation and, as a consequence, does not improve its unskillful predictions -beyond correcting the mean biases.- Contrarily, the SOI-conditioned version presented here (which is separately applied for three different sets of years; defined according to the terciles of the SOI) can modify the temporal sequence of the raw model output, providing more realistic local time-series, which results in improved ROC Skill Scores and reliability over the entire study area. This suggest that conditioning the bias correction on simple but wellsimulated large-scale processes relevant to the local climate may be a suitable approach for seasonal forecasting.