RADIOSONDE TEMPERATURE BIAS ESTIMATION USING A VARIATIONAL APPROACH Marco Milan Vienna 19/04/2012

MOTIVATION Radiosondes data often used as reference for other data Radiosondes data, especially upper air data, are not unbiased Different radiosonde type have, generally, different bias Different location for the same radiosonde leads to different bias All long-period stations have shifts leading to different bias A bias adjustment, which take into account all this problems, is needed Without bias adjustment, temperature trend are not believable

Previous works: –Radiosonde adjustment during ERA-INTERIM Current approach –Variational Bias correction –Type or predictors –Grouping Preliminary results Conclusions OUTLINE

Previous homogenization of radiosonde temperature dataset Adjustment of annual mean bias –Use of RAOBCORE (Haimberger et al. 2007, 2008) –Based on time series of individual station –Detection of shifts in background departure –Adjustment can change temperature trends Adjustment of daily/seasonal bias –Method based on solar elevation adjustment (Andrae et al. 2004) –Based on station groups –Four classes of solar elevation –Corrections calculated from the statistics of background departures over the previous 12 months ERA-Interim adjustment

Russian radiosonde, 12 UTC, 200hPa Start from 1979, when satellite data are available First guess departure, using uncorrected radiosondes Bias correction to apply Bias correction only until 2008, no applied any more After 2008 less departure but still existing, a bias correction is needed Limited departure probably due to changes on radiosondes dataset in the Russian federation

The observations are considered biased, a linear predictor model is used as observation operator in the 4DVAR equations: Introduction of a “bias term” in the variational cost function With x b and b b a priori estimations of model state and bias control parameters A weak constraint (large B b ) allows the parameter estimates to respond more quickly to the latest observation. The adjustment of the radiosondes depends on the resulting fit of the analysis to all other OBS, given the Background from the model. Variational Bias correction

Bias in observation can change during the time Seasonal and daily variations in bias exist The Bias model : Must be choose according with observations and physical origins of the bias. VarBC can be applied in the period where RAOBCORE detects a shift We assume the model unbiased, the presence of model bias attributes a wrong bias to the observations where there are not enough observation to correct the analysis Variational Bias correction

First results using only a constant bias parameter Pressure, for every class (group) j: First approach, good for US and Japanese radiosonde Solar elevation The equation can be formulate also for classes of solar elevation, grouping stations with similar solar elevation. Radiosonde temperature bias correction

First guess departure night Analysis of July 2011 Results divided per station type Large differences between different station type Average of first guess departure 1.Control run without variational bias correction 2.Same analysis applying a basic variational bias correction (only b 0 ) No significant differences are visible with and without VarBC 1 2

RMS first guess departure night The negative departures do not counteract the positive departures RMS give more weight to the bigger first guess departure The Russian stations has larger RMS in the upper levels and near 200 hPa No significant differences are visible with and without VarBC 1 2

First guess departure solar elevation > 22.5° 1 2 Different behaviour between station during the night and with high solar elevation Station in USA and Japan have larger positive departure in the upper levels No significant differences are visible with and without VarBC

RMS first guess departure solar elevation > 22.5° 1 2 Large RMS in the upper levels About 0.5K smaller for USA For Japan RMS in the upper levels different than during the night Russia has still problems around 200 hPa No significant differences are visible with and without VarBC

NIGHT SOLAR ELEVATION > 22.5° Time series for bias correction at 20hPa Generally very small bias corrections (as aspected). For higher solar elevation larger bias corrections Radiosondes with larger first-guess departure (Russia) have also the higher bias correction The bias correction are in the “right” direction but the amount is too low Bias correction at 20 hPa

First-guess departure bias correction Station 23921, Russia Vertical averaged first guess departure quite constant positive We aspire a bias correction which converge to a positive value of about 0.4K The bias correction for this station increase until a value around 0.02K and then decrease (negative fg- departure) The bias correction is in the “right” direction but the amount is too low B too large We do not exclude the occurrence of computational problems =0.389 K

CONCLUSIONS AND OUTLOOK Bias correction for temperature for radiosondes is far away from have a final solution. Different approach as in ERA-CLIM Use a “physical approach” (function of predictors) taking into account grouping of radiosondes VarBC can be applied where RAOBCORE detects the shifts First results, too low Bias correction but in the right directions Different predictors and functions for different groups have to be tested (work in doing)

First-guess departure bias correction Station 27730, Russia Vertical averaged first guess departure change from positive to negative The negative bias corrections could counteract the positive The bias corrections could be too slow =0.137K