The PhD thesis of Ramses Sivira (advisors: Dr Cécile Mallet and Dr Hélène Brogniez, UVSQ/LATMOS) has the following title: "Development of a retrieval method of relative humidity profiles and their conditional distribution of errors for Megha-Tropiques observations"
We designed a methodology that allows us to develop a purely statistic water vapor profile restitution algorithm from Megha-Tropiques SAPHIR and MADRAS instruments with synthetic observations, and specially to quantify the restitution of conditional uncertainties. Three statistical models were optimized using this learning database to estimate seven layers tropospheric water vapor profiles and their conditional error probability density function (pdf). The optimized models lead us to conclude a model-independency restitution accuracy and this accuracy is directly related to physical constraints. Also, maximal precision was achieved in mid-tropospheric layers (maximal bias: 2.2% and maximal correlation coefficient: 0.87) while extreme layers show degraded precision values (at surface and the top of the troposphere, maximal bias: 6.92 associated to a fort dispersion with correlation coefficient: 0.58), this behavior could be explained by instrumental information lack. From conditional error probability functions, knowing observed brightness temperatures, humidity confidence intervals were estimated by each layer. The two hypotheses were tested and we obtained better results from the Gaussian Hypothesis. This methodology was tested using real data and results are consistent with the learning database with better accuracy (bias: -5.77%) at mid-tropospheric layers, degrading it to extreme layers.