The Effect of Satellite Observing System Changes on MERRA Water and Energy Fluxes

Like all reanalysis efforts, the Modern Era Retrospective-Analysis for Research and Applications (MERRA) must contend with an inhomogeneous observing network. Here the effects of the two most obvious observing system epoch changes, the Advanced Microwave Sounding Unit-A (AMSU-A) series in late 1998 and, to a lesser extent, the earlier advent of the Special Sensor Microwave Imager (SSM/I) in late 1987 are examined. These sensor changes affect model moisture and enthalpy increments and thus water and energy fluxes, since the latter result from model physics processes that respond sensitively to state variable forcing. Inclusion of the analysis increments in the MERRA dataset is a unique feature among reanalyses that facilitates understanding the relationships between analysis forcing and flux response. In stepwise fashion in time, the vertically integrated global-mean moisture increments change sign from drying to moistening and heating increments drop nearly 15 W m(-2) over the 30 plus years of the assimilated products. Regression of flux quantities on an El Nino-Southern Oscillation sea surface temperature (SST) index analysis reveals that this mode of climate variability dominates interannual signals and its leading expression is minimally affected by satellite observing system changes. Conversely, precipitation patterns and other fluxes influenced by SST changes associated with Pacific decadal variability (PDV) are significantly distorted. Observing system changes also induce a nonstationary component to the annual cycle signals. Principal component regression is found useful for identifying artifacts produced by changes of satellite sensors and defining appropriate adjustments. After the adjustments are applied, the spurious flux trend components are greatly diminished. Time series of the adjusted precipitation and the Global Precipitation Climatology Project (GPCP) data compare favorably on a global basis. The adjustments also provide a much better depiction of precipitation spatial trends associated with PDV-like forcing. The utility as well as associated drawbacks of this statistical adjustment and the prospects for future improvements of the methodology are discussed.