Each year large human and economical losses are due to fog episodes, which, by the large reduction of visibility, affect air, marine, and land transportation . Fog forecasts remain quite inaccurate due to the complexity, non-linearities and fine scale of the physical processes taking part in the fog life cycle. Fog results from a combination of radiative, turbulent and microphysical processes as well as interactions with surface heterogeneities which will drive the relative importance of local and large-scale circulations. Recently, three-dimensional models have replaced one-dimensional models to forecast fog in most national weather services. Currently, convective-scale numerical weather prediction (NWP) models run with a horizontal resolution of approximately 1 km with frequent data assimilation cycles. While the importance of vertical resolution , aerosol activation or water deposition have recently been highlighted to improve fog forecasts, fog is also known to be highly sensitive to initial conditions . Therefore, accurate initial temperature, humidity and wind profiles are crucial to successfully forecast fog. However, the atmospheric boundary layer (ABL) has also been identified as a part of the atmosphere which is undersampled by observations. Even though satellite data enable global coverage all over the world, they provide limited information on the ABL due to the attenuation by clouds and degraded vertical resolution in the ABL. Additionally, uncertainties in surface properties (such as skin temperature and emissivity) limit the assimilation of surface-sensitive channels over land . Recently, an observing system simulation experiment (OSSE) by has demonstrated that temperature and moisture at the surface have a larger impact on fog forecast than surface wind observations, concluding that temperature and humidity profilers could potentially play a major role in the improvement of fog forecast initialization. Ground-based microwave radiometers (MWRs) are robust instruments providing continuous observations of temperature and humidity profiles as well as integrated liquid and water contents during all-sky weather conditions. Even if their vertical resolution degrades with altitude , most of their information content resides in the ABL and their high temporal resolution (few minutes) makes them suitable to monitor the evolution of fog. Despite the potential impact of MWRs in NWP models, assimilation experiments of their data have been limited to a few attempts. The first preliminary study of has demonstrated a positive impact of the assimilation of a single MWR unit into the 10 km horizontal resolution MM5 (https://www2.mmm.ucar.edu/mm5/, last access: 17 November 2020) mesoscale model in the context of a winter fog event. The impact of a simulated network of 140 MWRs through an OSSE was also investigated by and on a winter storm case. This study confirmed a positive impact on temperature and humidity analyses as well as up to 12 h forecasts on moisture flux. More recently, a real network of 13 MWRs was assimilated by into the 2.5 km horizontal resolution convective-scale model AROME in the context of heavy-precipitation events in the western Mediterranean. The impact of this network was found to be neutral on temperature and humidity fields but positive on quantitative precipitation forecasts up to 18 h. In addition, have demonstrated the positive impact that could be expected on NWP temperature profile analyses by the direct assimilation of MWR brightness temperatures into the AROME model with a one-dimensional variational framework (1D-Var). All these studies showed an encouraging positive impact of the assimilation of MWR observations into NWP, though they are limited to single deep-convection case studies on low-resolution limited-area models or restricted to temperature analyses only. The purpose of this article is to evaluate the expected benefit of MWRs on kilometre-scale NWP analyses during fog events on an extended dataset over a 6-month fog experiment. This expands the studies by to humidity and liquid water path retrievals and evaluates the impact of new tools developed to optimize the assimilation of MWRs during COST Actions TOPROF and PROBE . A fog-dedicated field experiment was carried out in the north-east of France during the winter 2016–2017 during which a 14-channel MWR has been operated. The impact of MWR brightness temperatures on temperature, humidity and liquid water content profiles forecast by AROME has been evaluated during the 6-month period against in situ data collected during intensive observation periods (IOPs) and continuous measurements deployed on a 120 m instrumented tower. This paper begins with an overview of the dataset and the AROME model and a description of the 1D-Var settings in Sect. . A sensitivity study of the 1D-Var retrievals to the background-error-covariance matrix and bias correction to select the optimal configuration is presented in Sect. . Section  presents a case study of the first IOP showing large AROME errors during a thin radiative fog event that are corrected when using the 1D-Var retrieval. Section  generalizes the results obtained in Sect.  through a statistical evaluation of 1D-Var retrieval errors and expected impact on the AROME analyses. Section  presents the deployment of a regional-scale MWR network for fog forecast improvement as continuity of this study, while finally Sect.  summarizes the main conclusions.

This section focuses on a thin radiative fog case observed on 28 October 2016. Figure  shows the cloud base height retrieved from a CL31 ceilometer (top panel), the visibility measurements on the instrumented tower at 10 and 120 m altitude (blue and green lines, respectively, middle panel), and the 1 h AROME forecasts of liquid water content (LWC) for the same day (bottom panel). During the whole period, fog is only observed at 10 m altitude during 40 min at midnight and then during 4 h from 05:00 to 09:00 UTC. A stratus cloud is then observed from 10:00 UTC until midnight with a cloud base height between 300 and 500 m. The AROME backgrounds simulate a continuous thick fog event from 00:00 to 13:00 UTC, which is then lifted until 15:00 UTC into a stratus cloud at 500 m altitude. The stratus cloud is then dissipated to appear again after 20:00 UTC. In this example, two main deficiencies in the AROME 1 h forecasts are observed: a temporally longer and vertically thicker fog event and the erroneous dissipation of the stratus cloud between 15:00 and 20:00 UTC.

Figure  compares the time series of temperature profiles (top panels) and specific-humidity (bottom panels) forecast by AROME (left panels) and retrieved with the 1D-Var scheme using the optimal configuration. We can note the large temperature increment by up to 5 K from 00:00 to 12:00 UTC essentially in the first 250 m after 1D-Var is applied; this is the period when the model simulates a thick fog event not confirmed by the observations. This is followed by a temperature cooling within 2 K during the stratus cloud (16:00 to 24:00 UTC). The specific humidity is only modified during the fog event (05:00 to 09:00 UTC) with an increase of 1 gkg-1 in the first 1500 m.

In order to quantify the accuracy of the 1D-Var increments in this specific fog case, Fig.  evaluates the corresponding diurnal evolution of temperature, specific humidity and relative humidity at 50 and 120 m altitude. A large underestimation of the temperature by 4 to 6 K is observed in the AROME forecasts by night until 13:00 UTC. AROME forecasts are also found to be too warm by 2 K after 18:00 UTC. The assimilation of MWR brightness temperatures in a 1D context greatly improves the model background (temperature) during the night-time fog event with temperature errors smaller than 2 K after assimilation. The 1D-Var retrievals almost perfectly fit the in situ observations after 13:00 UTC for temperature (both at 50 and 120 m altitude).

In terms of specific humidity, AROME tends to underestimate the specific humidity at night-time probably due to an overestimation of the saturation. Indeed, as the fog layer was thicker in AROME than in the observations, we believe the model converts too much water vapour into liquid erroneously, which makes it underestimate specific humidity. On the contrary, the specific humidity is overestimated in the afternoon. After 1D assimilation of MWR measurements, specific humidity is nearly identical to the AROME forecasts except during the longest fog event (between 04:00 and 09:00 UTC) where the 1D analysis is closer to the tower measurements than the background. This is likely due to the use of the cross-correlated fog B matrix under these conditions as opposite to the use of a block diagonal B matrix when fog is not observed. Most of the model increment is thus produced by the B matrix cross-covariances. Background errors are reduced from 0.5 to 0.1 gkg-1. Although closer to the in situ observations, 1D-Var retrievals slightly overestimate specific humidity between 04:00 and 09:00 UTC. This is most likely due to over-estimated positive cross-correlations between temperature and humidity in the B matrix. In terms of relative humidity, the temperature warming by night leads to the effect that the fog layer is not saturated any more in agreement with the tower in situ measurements. However, this field is degraded after 13:00 UTC. In fact, the 1D-Var scheme correctly reduces the temperature but is not able to decrease the specific humidity. The relative humidity is thus wrongly increased by the 1D-Var analysis.

In view of the future inclusion of hydrometeors in the data assimilation control variables, the information brought by MWRs to the liquid water path (LWP) could also be very valuable. Figure  shows the time series of LWP forecast by AROME, retrieved through the 1D-Var and retrieved from a quadratic regression applied on BT measurements. It can be seen that the AROME model clearly overestimates the fog LWP with a maximum reaching 90 gm-2 at 07:00 UTC. This value, however, decreased down to 25 gm-2 after the 1D assimilation of MWR brightness temperatures. During the period when the model fails to simulate the stratus cloud, the LWP is significantly increased in the 1D-Var analysis with values between 30 and 80 gm-2 even if the background profile has no cloud layer between 14:00 and 20:00 UTC. These LWP modifications brought by the 1D-Var are consistent with the in situ observations on the instrumented tower as well as ceilometer observations.

While the previous section focuses on an extreme fog case, this section aims at more general conclusions on the expected impact of MWR BTs assimilation on AROME analysis. To this end, a statistical evaluation of the expected model increments (analysis minus background differences) after assimilating MWR measurements has been conducted using the tower measurements during the 6-month period. The 1D-Var retrievals have been performed using the optimal configuration described in Sect. . A total of 351 h of fog (rain events have been removed) could be observed with the MWR. In order to evaluate the performance of the AROME background profiles (1 h forecast) to accurately forecast fog events, statistics based on the hit ratio (HR), false alarm rate (FAR), frequency bias index (FBI) and critical success index (CSI) were computed. If GD (good detection) is the number of fog profiles well detected, ND (not detected) is the number of undetected fog profiles and FA is the number of false alarms, then these scores are defined by HR=GDGD+ND,FAR=FAGD+FA,FBI=GD+FAGD+ND,1CSI=GDGD+ND+FA.

To detect fog profiles in the model space, a new visibility diagnosis specifically developed for the AROME model has been used (Ingrid Dombrowski-Etchevers, personal communication, July 2020). In this new diagnosis, the visibility is directly deduced from the liquid water content at ground. It was computed through a statistical regression between hourly maximum of liquid water content forecast by AROME and observed minimum of visibility on 100 ground stations during 5 months. A hit ratio of 73 % and a false alarm rate of 58 % was found. A FBI of 1.77 means that the AROME background profiles tend to forecast too many fog events. CSI equal to 0.35 means that only 35 % of fog events (observed and/or predicted) are correctly forecast by the model. These statistics emphasize that quite large errors are observed in the AROME 1 h forecasts of fog with an excessive number of false alarms. In order to evaluate the potential benefit of MWR observations to adjust the AROME background profiles, the statistical study of model increments is split between the good detections, missed fog profiles and false alarms. Firstly, the frequency distributions of differences of 1D-Var analysis and background with tower measurements at 50 m are displayed in Fig.  (both for temperature and specific humidity). For temperature and for all subsets, the distributions of 1D-Var analysis errors are more centred and more symmetric compared to the background-error distributions. Thus, the largest background errors (above 2 K in absolute values) are successfully corrected by the 1D-Var analysis. Background error distributions also present a larger tail towards negative values with a secondary peak centred around -4 K in the case of false alarms and to a smaller extent in the case of good fog detections. The largest temperature improvement is observed in the case of false alarms with only 35 % of the background errors being within -0.5 to 0.5 K, which is against 69 % for the analysis. RMSEs with respect to tower measurements are also significantly improved with values between 1.3 and 1.9 K in the background against 0.6 K in the analysis. The frequency distribution of specific humidity errors for 1D-Var analysis and background are close, with similar bias and RMSE for good detections and false alarms. A slight degradation is observed for missed fog detections with a RMSE of 0.33 gkg-1 in the analysis against 0.25 gkg-1 in the background. Overall, the impact on humidity is less evident than on temperature at 50 m altitude.

To get a vertical perspective, Fig.  shows the profiles of the frequency distribution of analysis minus background differences. As more than 90 % of the water vapour increments are within 1 gkg-1 up to 1500 m altitude, only the impact on temperature is discussed. For each vertical bin, the frequency of the temperature increments within a given range of values is shown. The frequency distribution of 1D-Var increments has been separated between cases of correct fog detection, missed fog and false alarms. For all of the dataset, most of the temperature analysis increments are observed below 750 m and span the range -5 to 5 K. The largest increments are observed between 100 and 300 m altitude for which around 20 % of the analysis minus background differences are larger than 2 K in absolute values. We can note significant differences in the shape of the increment distributions depending on the forecast score. While the distribution of good detections is quite symmetric, it is not the case for missed fog profiles and false alarm distributions. In the case of missed fog events, the distribution is negatively skewed close to the ground, whereas it is positively skewed above 100 m altitude. This asymmetry means that the largest analysis increments in magnitude tend to decrease the temperature close to the ground and increase the temperature above 100 m. Consequently, we can expect 1D-Var analyses to increase the atmospheric stability in the first 150 m, which is key for fog formation. In the case of false alarms, the distribution is positively skewed for all vertical levels. This asymmetry means that the largest analysis increments, though less frequent in the distribution, occur when the AROME forecasts tend to significantly overestimate the temperature cooling. By limiting the temperature cooling, the 1D-Var analyses might limit the erroneous saturation leading to false alarms in the background.

The additional value of MWR data for NWP forecasts and process studies is in the LWP product. In fact, MWR is one of the most reliable sources for this variable , which is key for better understanding the microphysics of fog life cycle and limiting the forecast spin-up (i.e. the unbalance of thermodynamic profiles with microphysical variables during the analysis). In fact, as hydrometeors are currently not included in the control variables of most operational variational data assimilation schemes; these fields are kept unchanged during the analysis. Thus, the analysed hydrometeor fields correspond to the previous background. Consequently, in the following statistics, the background values of LWP correspond in fact to the LWP in the operational AROME analysis. These fields are then modified according to the updated temperature and humidity analyses in the first time steps of the forecast through the model physics. The statistical study performed here is also useful to evaluate the expected impact on the AROME analyses if MWR observations were assimilated and the LWP included in the control variables. To this end, Fig.  investigates the frequency distribution of LWP increments split by forecast skill (good detections, undetected fog, false alarms). Firstly, we can note that the LWP increments are higher than 50 gm-2 in absolute values for approximately 50 % of good detections and missed fog profiles and 30 % of false alarms. During false alarms, 95 % of the background LWP values are below 20 gm-2 (not shown), which is close to the MWR sensitivity which might explain smaller 1D-Var increments during false alarms. The mean increment is the highest in the case of missed fog events (57 gm-2) and the smallest in the case of false alarms (15 gm-2). It is important to note that during false alarms, the LWP increment might be positive due to the presence of cloud layers, though we would expect the 1D-Var analysis to decrease the LWP within the fog layer. If we restrict the statistics to false alarms without cloud aloft, the mean increment is reduced to -2 gm-2. As expected, large positive increments occur more often in fog cases undetected by AROME with 47 % of the distribution showing increments above 50 gm-2 against 35 % in good detections and 22 % in false alarms (8 % for false alarms without cloud layers aloft). To further investigate the LWP increments and retrieved values, more in situ data are necessary, e.g. from the cloud droplet probe mounted on the tethered balloon or cloud radar measurements. However, the lack of cloud radar measurements to differentiate the LWP within the fog layer and cloud aloft makes this evaluation complex. Too few cases during which MWR observations were co-located with an entire sounding of the fog layer with the tethered balloon have been sampled to make an independent evaluation of this product. This is why we use the LWP derived from the MWR alone through a quadratic regression as a reference. The expected accuracy of this product is 15 to 20 gm-2 according to . To this end, Fig.  shows the scatterplot between the LWP retrieved with the MWR alone (through multichannel regressions provided by the manufacturer) and the 1D-Var analyses or background profiles (left panel). We can note the large improvement in correlation between the LWP forecast by the background (0.72) versus the 1D-Var analysis (0.98) with respect to the MWR multichannel retrieval. This is of course expected as the 1D-Var minimization tends to get closer to the MWR brightness temperatures which are also used in the multichannel retrieval. However, this evaluation is a good sanity check, showing the good behaviour of the 1D-Var algorithm and its capability to extract the information from the observation even with very large errors in the first-guess background profiles. The mean error of the AROME LWP is -49 gm-2 and is reduced to -2 gm-2 after 1D assimilation. The root-mean-square error is significantly reduced from 102 to 27 gm-2.

The same evaluation has been carried out on the IWV (Figs.  and ). Since MWRs are more sensitive to column integral than vertical distribution, a more significant impact is expected on IWV than specific humidity profiles. The IWV increments span from -4 to 4 kgm-2, which correspond to a change in the background IWV of up to 30 %. The distribution of IWV increments is positively skewed for correct fog detection, meaning that the largest increments in magnitude are observed when the background underestimates the integrated water vapour content. On the contrary, it is negatively skewed for missed fog profiles, meaning that the largest increments occur when the model overestimates the integrated water vapour content. It is more symmetric in the case of false alarms. The correlation coefficient with respect to the MWR multichannel retrieval (Fig. ) is slightly increased from 0.97 to 1. The RMSE is improved from 1.30 to 0.71 kgm-2. The impact of MWR observations is thus positive on IWV, though the good quality of AROME humidity forecast leaves little room for improvement. This could be explained by the assimilation of observations sensitive to the total column water vapour like Global Navigation Satellite System (GNSS) zenith total delay. Further investigation on multiple sites would be needed to confirm this hypothesis.

The next natural step of this study would be to calculate updated scores of fog detections with the new 1D-Var analyses compared to the background profiles. However, forecast scores are only based on the LWC at ground, whereas the 1D-Var works on the liquid water path without information on the cloud vertical structure. During false alarms, conclusions on the impact on forecast scores are complexified by the presence of cloud layers above fog in a majority of false alarms, which can cause an increase in LWC at ground. As for the hit ratio, it is increased from 73 % in the background to 81 % in the analysis. The rate of missed fog events is also decreased from 27 % in the background to 19 % in the 1D-Var analysis. However, as this evaluation is only based on the LWC change at the ground, it is necessary to evaluate the impact of the new temperature and humidity fields on the LWC after a few time steps of forecasts, but this is beyond the scope of this paper. This investigation into the forecast impact will be studied in the future within the framework of the SOFOG3D experiment (Sect. ).

This study has proved MWRs to be potential good candidates to be assimilated into current mesoscale models with a special focus on fog forecast improvement. However, our conclusions are currently limited by the small dataset (only one winter at one site) and the lack of impact studies on fog forecast. Although, a positive impact is expected on the analysis of the ABL temperature profile and the LWP and, to a smaller extent, to the IWV; the next step will be to quantify the impact of a more accurate initial state on fog forecast capability. Among the huge number of observations currently assimilated into operational models, the assimilation of only one MWR unit would probably not be efficient to effectively constrain the boundary layer in the model analysis and to keep the valuable information brought by this local observation over the forecast range. In order to go further into this evaluation, the deployment of a dense network of MWRs is necessary to perform a data assimilation study into the operational AROME 3D-Var assimilation system. Thanks to the strong European collaboration built in the framework of the COST Action TOPROF (https://www.cost.eu/actions/ES1303/#tabs|Name:overview, last access: 17 November 2020), pursued by the COST Action PROBE (PROfiling the atmospheric Boundary layer at European scale; ), an unprecedented regional-scale network of eight MWR units has been deployed in the south-west of France during the period October 2019 to April 2020. This work will serve the data assimilation experiment, fog process studies and model evaluation of the international SOFOG3D (SOuth FOGs 3D experiment for fog processes study) experiment led by Météo-France. Figure  shows the domain of the dedicated 500 m horizontal resolution AROME version in test for evaluation during SOFOG3D and the location of the eight MWR units deployed for the experiment. MWR locations have been chosen for an homogeneous spread over the AROME domain at sites known for their high frequency of fog occurrence. An increased density of MWRs is found at the super-site with two co-located MWRs and a third humidity profiler deployed approximately 7 km away from the super-site to document the impact of surface heterogeneities on fog characteristics. The methodology introduced in this paper will be extended to the eight MWRs deployed during SOFOG3D. This large dataset will help with quantifying the spatio-temporal variability of fog parameters (thermodynamics and microphysics) between the different sites; better understand the main processes playing a role in fog formation, dissipation, and development; and run real data assimilation experiments using the operational 3D-Var assimilation scheme of the AROME model to quantify the expected fog forecast improvement thanks to ground-based MWRs.

In this study, the expected benefit of ground-based MWRs on NWP analyses during fog conditions has been investigated with a 1D-Var technique. Temperature, humidity and LWP have been retrieved through the optimal combination of short-term forecasts and MWRs brightness temperatures. In this study, a new retrieval algorithm, combining the NWPSAF 1D-Var and the fast radiative transfer model RTTOV-gb, has been evaluated on a 6-month period spanning 351 h of fog conditions. The first part of this work aimed at deriving an optimal background-error-covariance matrix for fog conditions with the use of the newly developed AROME EDA. Similarly to , background-error standard deviations were found to be approximately 40 % larger within the first 250 m for temperature compared to a commonly used climatological B matrix. For specific humidity, similar standard deviations were observed. Most of the differences between a climatological B matrix and a fog B matrix were observed in the cross-correlations between temperature and specific humidity, with a strong positive coupling within the fog layer and uncoupling between the fog layer and atmospheric layers above. The impact of the B matrix and bias correction has been investigated through a statistical evaluation of the retrieval accuracy with respect to the in situ measurements on the instrumented tower at 50 and 120 m altitude. The optimal configuration has been defined through the definition of the error reduction brought by the analysis over the background for each variable (temperature and specific humidity) and each altitude. The best scenario mimics the use of a “flow-dependent” B matrix by using a cross-correlated fog B matrix when fog is detected by visibility measurements and an uncorrelated climatological B matrix during the other conditions. The retrievals of specific humidity at 120 m altitude are the most impacted: contrary to the significant degradation of the background by around 20 % with a suboptimal B matrix, an improvement of 21 % of the background is obtained with an optimal B matrix. This demonstrates the crucial role of the B matrix cross-correlations when assimilating observations with low information content on the vertical structure. Consequently, the ongoing development of a 3D-EnVar scheme for the AROME model is a necessary step to optimally assimilate MWR observations into the AROME model. The use of a static bias correction based on the monitoring of observation minus background innovations was also evaluated. Biases of less than 0.5 K were observed for K-band and opaque V-band channels and up to -4.7 K for the most transparent V-band channels. The found bias is similar to previous studies; its correction applied to BT measurements improves humidity retrievals above 2000 m but degrades temperature retrievals in the first 200 m. This degradation is most likely due to well-known larger model errors in the boundary layer during stable conditions, which are incorrectly included in the bias correction. Restricting the computation of the bias correction to clear-sky unstable conditions was found to remove most of the degradation. Overall, with the best configuration (flow-dependent fog B matrix and no bias correction for most opaque channels), temperature and humidity profiles could be retrieved with RMSE below 1.6 K and 1 gkg-1 up to 6 km in the troposphere.

A thin radiative fog sampled during the first IOP of the experiment was then described. For this specific case, the AROME model was found to simulate a temporally longer and vertically thicker fog event and is not able to maintain the stratus cloud in the afternoon. After 1D assimilation of MWR observations, a large warming up to 5 K is observed within the first 500 m during the fog event associated with an increase in specific humidity and a decrease of LWP by 40 to 70 gm-2 consistent with in situ measurements showing the large impact brought by MWR observations to modify the initial state of the model in fog conditions.

Finally, a statistical evaluation of the expected model increments after assimilating MWR measurements has been conducted using tower measurements. Large forecast errors were observed in the AROME backgrounds with a tendency to overestimate the presence of fog. During missed fog profiles, 1D-Var increments pull towards lower temperatures close to the ground and higher temperatures above 100 m altitude, i.e. higher atmospheric stability. The largest analysis increments and background errors are observed during false alarms when the AROME forecasts tend to significantly overestimate the temperature cooling. Overall, RMSE values from 1.3 to 1.9 K are observed in the background against 0.6 K in the analysis. For specific humidity, analysis increments are small and below 1 gkg-1 within the fog layer. On the contrary, a large impact has been found on the LWP with increments up to 200 gm-2 in extreme missed fog events. A larger impact was found on the IWV than the humidity profile with a RMSE with respect to tower measurements that decreased from 1.3 to 0.7 kgm-2 during observed fog profiles. However, it was noted that the AROME backgrounds are more accurate for the IWV compared to temperature and LWP, which leaves less chances for improvement.

Using for the first time the RTTOV-gb fast radiative transfer model, this study investigated the impact of assimilating MWR observations in the AROME model during fog conditions. This evaluation, previously limited to temperature profiles only, was extended to humidity and LWP. Promising results are shown, with significant positive impact on temperature and LWP and small but slightly positive impact on humidity. In order to confirm the results obtained in a 1D-Var framework, the next step is now to assimilate a real network of ground-based MWRs through a 3D-Var or 3D-EnVar data assimilation scheme. Following the recommendations of and thanks to the strong European collaboration built within the TOPROF and PROBE COST Actions, eight MWRs have been deployed in the south-west of France from October 2019 to April 2020 in the context of the international fog campaign SOFOG3D. The locations of the MWR units have been chosen to optimize their impact in the model specifically for fog forecast evaluation. A 1D-Var plus 3D-EnVar approach will be used to assimilate profiles retrieved through the 1D-Var algorithm presented here, taking the most out of the lessons learnt in this work.