Emissivity, Sensor Noise, Water Content and Viewing Angle

We have seen in previous chapters that the major obstacle for LST estimation is the lack of information about LSE. If we are trying to make the measurement far from the surface, for instance using a sensor onboard a far-off satellite, other obstacles are revealed; if we use thermal infrared observations to make the LST estimation, we come across the problem of cloud masking, since in the TIR region, the surface can only be seen under clear sky situations. A possible solution is to use the NWC SAF (http://www.nwcsaf.org/) software, which provides several satellite products in Near Real Time (NRT), among them the cloud mask, an operational product that has been validated against independent datasets.

So, assuming the cloud masking is solved, what are the sources of uncertainty in LST estimation?

It is obvious that errors in the input variables (emissivity and atmospheric information) will propagate into errors in LST. We have previously seen that channel emissivity can either be estimated a priori, or using a method where LST and emissivity are retrieved simultaneously. In any case, uncertainties in emissivity computation will propagate to LST estimation. Generally, under normal conditions, an uncertainty of 1% in the LSE will result in about 0.5 K error in LST. The effect of atmosphere on the total measured infrared radiances at the TOA must also be accounted for. The accurate correction of this effect requires accurate information on the vertical profiles of temperature and humidity, which are highly variable in both space and time. Moreover, the atmospheric optical path depends on the viewing angle.

Satellite sensors are complex systems that capture energy from a given source and convert it into an electric signal. One limitation of these systems is sensor radiometric noise, which may have several contributors, such as the temperatures of the sensor and the environment. Values are provided for each infrared channel by the institutions that operate each satellite. As an example the expected radiometric noise of SEVIRI channels IR108 and IR120 onboard MSG-2 are 0.11 K and 0.16K, respectively.

Figure 15 shows the error bars (in degrees Celsius) of LST as computed by the LSA SAF for the SEVIRI sensor, for 23 March 2008 at 07:15 UTC. In this figure bars arranged in diamond-shapes, with the diamonds' apexes as compass points, show the relative contribution of uncertainties in emissivity (north), water vapor content (east), sensor noise (west), and uncertainty in the LST associated with specific retrieval conditions (south).

In this example (Freitas et al., 2010), LST is estimated with a reasonable degree of accuracy (errors < 1.6 °C) for the cloud-free parts of northern Europe, Middle East and Saudi Arabia. Since these regions present a relatively low atmospheric water vapor content, the largest contribution to LST uncertainties is sensor noise. For the dry region of north Africa, close to the equator, the major contributors for LST errors are emissivity uncertainties. The west region of South America near the edge of the disk is viewed at a higher angle than the aforementioned areas. This, together with high tropospheric humidity, causes the LST retrievals to be useless.

Fig. 15: Error bars (in degrees Celsius) of LST as computed by the LSA SAF for the SEVIRI sensor in 23 March 2008 at 07:15 UTC. The diamonds show the relative distribution of uncertainties in emissivity (north), water vapor content (east), (west) and the uncertainty in the LST associated with specific retrieval conditions (south).

It is essential that satellite products are distributed together with realistic estimations of the respective error bars on a pixel-by-pixel basis to allow users to make the decision about the usability of the product, which could depend on the intended application.