Radiometer retrievals of LWP Nicolas GAUSSIAT
Statistical/physical retrieval SOME BASICS: Radiometers measure Tb at 22.2 GHz end 28.8 GHz with Physical retrieval : kl, kv, d are estimated using GCM model vertical information, then with Statistical retrieval : a22, a28 and b are assumed to be constant and directly deduced from observations of Tb and LWP
Model data /radiosondes 2 years of model data 4 years of radiosondes LWP VWP
Physical retrieval d22, d28 (Kl22, kv22, d22), (Kl28, kv28, d28) are estimated using GCM model (ECMWF) for each profiles, and a22, a28 and b are deduced. a22 Kl22, Kl28 Kv22, Kv28 a28 d22, d28 b
Statistical retrieval Use of a bi-linear regression to work out a22, a28 and b directly We assume that LWPi depends linearly on the optical depths 22i and 28i Ideally, we want to find a22, a28, b for all the N measurements : a22 22i + a28 28i + b =LWPi But the best we can do is to minimize the sum : R2 = (a22 22i + a28 28i + b - LWPi)2 In matrices notations A=[…(22i 28i 1) ;…(22N 28N 1) ] X=[a22 ; a28 ; b] Y=[… LWPi ;… LWPN] X = (AtA)-1 AtY a22 (kg/m2xm) a28 b RMS error g/m2 Using (m) LWP<500 g/m2 -1.9962 6.8043 -0.1530 17 g/m2 0.0017 cm Using Tb instead of -0.009 0.029 -0.197 20 g/m2 Simpson et Al. -0.008 0.025 -0.126 18 g/m2 0.0018 cm
Results can look good but … Lidar identify liquid-cloud-free regions (LWP should be 0) statistical physical clear sky
when it comes to the details… minutes Fluctuations must be due to a calibration problem…
Cunning technique… In clear sky conditions LWP=0 so we have : then (1) A second constrain is found by minimizing the cost fct J, then (2) So, from (1) and (2), we work out C22 , C28 in the clear sky regions and interpolate values in between
…cunning results(1) Chilbolton offset (lens ?) … solved
…cunning results(2) Palaiseau calibration problem… not a problem.
Further improvement Lidar measure cloud-base and using model temperature a better estimate of Kv is obtained.