Downscaling climate variability associated with quasi-periodic climate signals: A new statistical approach using MSSA

SummaryA statistical method is introduced to downscale hydroclimatic variables while incorporating the variability associated with quasi-periodic global climate signals. The method extracts statistical information of distributed variables from historic time series available at high resolution and uses Multichannel Singular Spectrum Analysis (MSSA) to reconstruct, on a cell-by-cell basis, specific frequency signatures associated with both the variable at a coarse scale and the global climate signals. Historical information is divided in two sets: a reconstruction set to identify the dominant modes of variability of the series for each cell and a validation set to compare the downscaling relative to the observed patterns. After validation, the coarse projections from Global Climate Models (GCMs) are disaggregated to higher spatial resolutions by using an iterative gap-filling MSSA algorithm to downscale the projected values of the variable, using the distributed series statistics and the MSSA analysis. The method is data adaptive and useful for downscaling short-term forecasts as well as long-term climate projections. The method is applied to the downscaling of temperature and precipitation from observed records and GCM projections over a region located in the US Southwest, taking into account the seasonal variability associated with ENSO.

Research highlights
► A data-adaptive downscaling statistical method based on MSSA is introduced.
► The method is applied to ENSO-related hydroclimatic variables in the US Southwest.
► The method is useful for short-term forecasts and long-term GCM climate projections.