Sun shining through a dried leaf

Assimilation of IASI L2 temperature and humidity profiles in regional and global NWP

Sun shining through a dried leaf
Sun shining through a dried leaf

These two studies looked at assimilating forecast-independent temperature and humidity retrievals from hyperspectral sounders in numerical weather prediction (NWP) models.

Last Updated

04 December 2020

Published on

22 June 2020

Two precursor studies have been conducted with ECMWF and Météo-France to assimilate forecast-independent temperature and humidity retrievals (i.e. Level 2 products (L2)) from satellite hyperspectral sounders in numerical weather prediction (NWP) models. These studies build on the regional service EARS-IASI L2, which provides atmospheric sounding in clear and cloudy pixels to users within less than 30 minutes from sensing. The two studies also contribute to MTG-IRS preparatory activities.

Positive impact on forecasts have been obtained in both data assimilation (DA) systems with the L2 products, comparable to the impact of assimilating radiances.


Objectives

The studies aimed at evaluating the impact and assessing the practical aspects of assimilating L2 temperature and humidity profiles in an operational NWP context, relative to the assimilation of radiances.

While there are no immediate operational plans to assimilate L2 products in NWP centres, where the assimilation of satellite hyperspectral radiances is the paradigm, there is a renewed scientific interest for geophysical parameters in numerical models (Coniglio et al. 2019, Hu et al. 2019). They may represent an interesting alternative to maximise the information ingested e.g. in regional NWP contexts with strong timeliness constraints with respect to the computational requirements associated with the assimilation of radiances.

  • At Météo-France: The retrieved temperature and humidity profiles have been assimilated in the regional model AROME (1.3 km resolution). The L2 products were retrieved from IASI and collocated microwave measurements, from the AMSU and MHS companions on-board Metop. As a consequence of a recent modification of the model top from 1 to 10 hPa in AROME, a number of IASI channels with contributions above 10 hPa have been removed from the assimilation system. This effectively reduced the overall amount of information conveyed into the model. A secondary objective of this study was to review the radiance assimilation process at Météo-France, where this remains the operational baseline, with lessons from another assimilation methodology.
  • At ECMWF: The IASI products have been assimilated in the integrated forecasting system, IFS. The temperature and humidity profiles were retrieved in pure infrared mode, i.e. without help of microwave data, to serve as proxy-products for IRS - as MTG does not include microwave instruments.

Overview

Study structure

Both studies followed the following steps:

  1. Evaluation of the IASI L2 products.
  2. Definition of the assimilation experiments (horizontal and vertical data thinning, observation error determination, quality control strategy…).
  3. Assimilation runs during relevant seasons and spatial coverage.
  4. Forecast impact evaluation on objective references and individual case studies.

The IASI L2 data showed good quality as evaluated against independent observations routinely assimilated and monitored in the models, e.g. radiosoundings and airborne (AIREP) data, as well as against the two models themselves.

The preliminary assimilation experiments showed that the IASI L2 products are suitable for assimilation in NWP models.

Summary

Météo-France study:

  • The IASI L2 profiles have been treated as pseudo-sondes, i.e. without taking into account the vertical sensitivity in the observation operator and without the vertical error correlations. The experiments focused on the best quality cloud-free retrievals.
  • The assimilations were performed in a full system, excluding both L1 and L2 products from IASI/AMSU/MHS (baseline) and introducing L1 (reference) and L2 (experiments) products.
  • Positive impacts of assimilating L2 humidity and temperature profiles were obtained, as evaluated against independent reference measurements (ground-based and in situ), of similar amplitude as L1 radiances (see Figure 1 and 2). Some degradations, especially near surface and in the Upper Troposphere and Lower Stratosphere (UTLS), were also reported.
Temperature forecasts vs Aircraft (left) and Radiosonde (right) measurements, bias (top) and standard deviation (bottom) in Spring 2018. Red: Baseline, Black: Control w. radiance, Blue: L2 experiments
Figure 1: Temperature forecasts vs Aircraft (left) and Radiosonde (right) measurements, bias (top) and standard deviation (bottom) in Spring 2018. Red: Baseline, Black: Control w. radiance, Blue: L2 experiments
Humidity forecasts vs ground Radar (%RH, left) and Radiosonde (specific humidity, right) measurements, bias (top) and standard deviation (bottom) in Spring 2018. Red: Baseline, Black: Control w. radiance, Blue: L2 experiments
Figure 2:Humidity forecasts vs ground Radar (%RH, left) and Radiosonde (specific humidity, right) measurements, bias (top) and standard deviation (bottom) in Spring 2018. Red: Baseline, Black: Control w. radiance, Blue: L2 experiments

ECMWF study:

  • The observation error correlation matrix was diagnosed for the IASI L2 temperature and humidity profiles, aiming first at the best retrievals in clear-sky over ocean. The observation operator did not account for the vertical sensitivity and resolution of the satellite products.
  • The assimilation experiments were conducted in a full and in a depleted system.
  • Assimilating temperature profiles in these conditions negatively impacted the forecasts.
  • Assimilating cloud-free humidity profiles improved the humidity and wind forecasts in both the depleted and full systems, comparably to the positive impact of assimilating radiances, while having neutral impact on the temperature forecasts (see Figure 3).
  • Assimilating cloudy humidity retrievals with inflated observation errors consolidated the positive impacts on forecasts (see Figure 4).
 First guess departure (blue: radiances, grey: cloud-free L2 experiment) width=
Figure 3: First guess departure (blue: radiances, grey: cloud-free L2 experiment) standard deviation normalised by the baseline experiment for radiosonde humidity and wind measurements, in the full observing system (left panel) and depleted system (centre and right panels).
 First guess departure (clear-sky IASI L2: grey, with cloudy IASI L2: black)
Figure 4: First guess departure (clear-sky IASI L2: grey, with cloudy IASI L2: black) standard deviation normalised by the depleted observing system baseline experiment for radiosonde humidity (left panel) and temperature (right panel) observations.

Conclusions

The fact that positive results have been obtained despite the simplified assimilation configuration is very encouraging. Using averaging kernels of retrievals as observation operators would be beneficial both in the product evaluation and assimilation processes. In this perspective too, the utilisation of scene-dependent retrieval error would help exploit more retrievals with appropriate weighting in the vertical. This is the scope of a new study with ECMWF, where EUMETSAT is also looking into ways of mitigating the data volume issue this may represent.

The characterisation of horizontal error correlation would be needed to increase the density of IASI profiles assimilated in the system.