Passive microwave observations from satellite radiometers operating in the spectral range from 1-200 GHz are widely used to make remote-sensing measurements of the Earth’s atmosphere and surface characteristics. Observations in this frequency range are sensitive to atmospheric humidity and temperature, as well as to emission and reflection from the surface. Microwave instruments, which are primarily used to estimate atmospheric temperature and humidity profiles (e.g. AMSU-A, MHS, AMSR-2), can also have a significant contribution from the surface, depending on atmospheric conditions. Currently the calibration/validation (cal/val) of these instruments tends to be carried out only over ocean due to more trustworthy estimates of the surface contribution, but in the future, this should be extended to the land so that cal/val can be performed over the full dynamic range of the instruments. To do this, it is necessary to validate the estimated land-surface contribution to the TOA radiances, and to calculate the associated uncertainties in radiance space.

The calculation of the land-surface contribution to the TOA radiances relies on simplified radiative-transfer equations, and estimates of the surface ‘skin’ temperature and emissivity. It is assumed that the land surface represents a homogeneous body with an emission equal to the skin temperature multiplied by an emissivity. An additional contribution to the TOA brightness temperature is calculated as the atmospheric emission reflected off the surface, which can be assumed to be either a specular or Lambertian reflection (Lambertian for snow-covered surfaces, specular for many other surfaces). In reality, the surface emission is more complex, due to multi-layers with heterogeneous dielectric properties (varying both vertically and horizontally), particularly for snow cover, and the reflection is likely to be not entirely specular or Lambertian. Furthermore, microwave emissions come from layers deeper than the surface (depending on frequency and dielectric properties) and so the use of a skin temperature estimate may not be appropriate for some conditions, particularly over deserts where the penetration depth is higher (see e.g. Norouzi et al; 2012).

As with the ocean surface, physically-based models have been developed to allow the estimation of land surface emissivity (e.g. Wang and Choudhury, 1981; Njoku and Li, 1999; Weng et al., 2001) for different surface types and different frequencies. Methods to estimate the surface type from satellite observations have also been developed (e.g. Grody, 1988). However, in order to accurately calculate the emissivity using physically-based models, a large number of input parameters are required that are difficult to estimate accurately over the spatial scales needed for satellite measurements. Progress in this area is still ongoing, but as a result, it has become necessary to rely on retrievals from satellite observations, following the methods developed by Karbou et al (2006; 2010). At the Met Office, for example, the microwave skin temperature and emissivity values are retrieved simultaneously in a 1D-Var system from the window channels of temperature and humidity sounders. At ECMWF and Meteo-France, the emissivity is also calculated from window channel observations, but with the skin temperature taken from the NWP model values.

Uncertainties in the land-surface contribution to the TOA radiances are a combination of uncertainties in: the emissivity values used, skin-temperature estimates, and the simplified radiative-transfer equations. The individual uncertainties of each of these contributions should be accurately estimated. As well as validating the individual components, the overall contribution can also be validated using experimental campaigns with ground-truth data, as well as comparisons to the TOA brightness temperatures from satellite instruments. It is likely that a combination of approaches will be needed to close the gap on uncertainty.

Estimates of uncertainties in retrieved land-surface emissivity have been calculated by Prigent et al (1997, 2000) and Karbou et al. (2005a), from the standard deviations of values retrieved from the satellite observations of different instruments. The authors provided gridded maps of uncertainties, which were shown to be around 2% on average. However, these uncertainties are indicative rather than robust, and are likely to be underestimates since they do not account for uncertainties due to: the calibration of the satellite instruments used in the retrievals, the temperature-humidity profiles used to calculate the channel transmittances, cloud screening, and surface temperature data. Ruston et al (2004) also carried out emissivity retrievals from SSM/I satellite observations over the USA and estimated the uncertainties in retrieved emissivity by randomly perturbing the input parameters. The authors concluded that errors were around 2% for frequencies less than 85 GHz. Their methods did not include possible errors in the atmospheric component due to the water-vapour continuum, however.

A number of experimental campaigns have been carried out to evaluate land surface emissivities over different surface types. For example, Harlow (2011) demonstrated how airborne microwave measurements can be used to validate the emissivity of snow-covered ice, relating to snow depth and snow pack characteristics, and quasi-Lambertian reflectance behaviour. Comparisons of different emissivity models have also been carried out. Ferraro et al. (2013) attempted an inter-comparison of several EO land emissivity data sets over the USA. The authors found differences of around 10 K in radiance space (emissivity x skin temperature) for frequencies up to 37 GHz and greater differences up to around 20 K for higher frequencies. These differences appeared to be generally systematic rather than random, with similar seasonal trends captured by the different datasets. Tian et al (2014) estimated uncertainties in retrieved emissivity values by comparing retrievals from different satellite sensors. They estimated similar uncertainties to Ferraro et al (2013), with systematic differences around 3 – 12 K over desert and 3 – 20 K over rainforest (with largest differences at the higher frequencies above 80 GHz).  Random errors were estimated to be around 2 – 6 K.

As well as estimating uncertainties in the emissivity values retrieved, it is important to also consider uncertainties due to assumptions made in the simplified radiative transfer equations. For example, Karbou and Prigent (2005b) estimated the uncertainties in emissivity due to the specular assumption by performing emissivity retrievals from brightness temperatures simulated, using both the specular and Lambertian assumptions. They concluded that the errors in retrieved emissivities due to the specular assumption were less than 1% for most surfaces.

While to date there have been considerable efforts to validate the calculation of the surface contribution to TOA microwave radiances at frequencies between 1 – 200 GHz, none of the uncertainty estimates have been traceable or complete. This is in part due to the complexity of the problem and it is likely to take a combination of a number of approaches before the gap can be fully closed. However, in part 3 below we suggest two areas of development which could contribute to the estimation of uncertainties in the land surface radiative transfer to reference standards.