LAUSR

12 Adding a Forecast Sensitivity-based Observation Impact component to version 2 of the 13 Modern Era Retrospective-analysis for Research and Applications, the present study pro14 vides an assessment of the impact of nearly 40 years of observations on short-range (2415 hour) forecasts over the Amazon basin. Under self-verification, forecast errors are found 16 to slightly increase from the early data-sparse days to the more recent years, when data 17 dramatically increases. Throughout the reanalysis, satellite radiances dominate in vol18 ume, but only before 1999 they dominate the impacts. Beyond 1999, over 50% of fore19 cast error reduction is associated with conventional observations (radiosondes). Atmo20 spheric Motion Vectors (AMVs) are also found to be large contributors to error reduc21 tion, but their contribution reduces in dry periods. In opposition to AMVs, satellite ra22 diances tend to contribute more in the dry season. Results provide motivation for ad23 ditional conventional observations and the use of all-sky treatment of radiances. 24 Plain Language Summary 25 Observations of atmospheric variables are of fundamental importance to allow for 26 reliable weather predictions and to enable scientists to improve their modeling of the at27 mosphere. Conventional observing systems measure temperature, winds, humidity, and 28 pressure directly. These amount to a small fraction of the global observing system, which 29 is dominated by indirect satellite observations. Objective evaluation for how different 30 components of the observing systems contribute to improving weather predictions have 31 become essential to help scientists understand how best to build future observing sys32 tems. The present study provides an evaluation of nearly 40 years of observations used 33 in the context of a procedure called reanalysis, which essentially blends observations and 34 model predictions in a carefully designed manner. Our particular work examines the im35 pact of observations over the Amazon basin. In this region, conventional observations 36 are found to still contribute most to reducing forecast errors, especially in the later years 37 of the reanalysis, while satellite-derived winds are found to contribute most in the wet 38 season. The work suggests that improving the treatment of other satellite observations 39 allowing their use over cloudy and precipitating regions might change their ranking in 40 comparison to conventional observations. 41