Analysis of global irradiance measurments from pyranometer and AVHRR

This thesis is about analysis of the accuracy of downwelling solar surface irradiance (DSSI) retrieval schemes for satellites with ground-based measurements for different atmospheric conditions. A two-step retrieval scheme is used to obtain DSSI values from satellite measurements. The first step of the retrieval scheme uses bi-directional reflectivities as input to obtain narrowband cloud optical depths.
The second step uses these cloud optical depths to obtain broadband atmospheric transmissivities (BAT). However, a conversion from narrowband to broadband cloud optical depths has to be made first. At the end of this two-step approach, the BAT is used to calculate DSSI. Inversely, the second step is used in order to obtain broadband cloud optical depths from derived BAT values measured by the pyranometer.

For the interpretation of the AVHRR channel 1 reflectivity and the pyranometer BAT, calculations are performed on mean atmospheric state. Look up tables (LUTs) store values of both the 0.63 μm bidirectional reflectance and the BAT as calculated for a wide range of satellite viewing geometries and solar zenith angles for 14 values of cloud optical depths. For a single satellite pixel, multilinear interpolation in the geometrical parameters and the vertically integrated water vapour is used to extract the corresponding pairs of reflectance/ transmissivity for a given measurement at all values of cloud optical depth included in the LUTs. Then, the relation of reflectance and transmissivity is established by bicubic spline interpolation and the transmissivity corresponding to the AVHRR channel 1 measurement is determined.

Three cases representing three different atmospheric conditions are analysed. These cases, which are referred to as "clear sky'. overcast, and 'broken clouds' are selected on their cloud cover faction. The clear sky case acts as a reference case, since it is not complicated by the variability of clouds and their optical properties introduces large uncertainties in the modelling of cloudy atmosphere. As a consequence, clear sky observations provide the most reliable reference case for evaluating the representation of the interaction of radiation with atmospheric aerosols and gases in a radiative transfer model (RTM) (Deneke, 2002).

The clear sky case shows that the satellite retrieval yields results that agree with independent pyranometer measurements. The bias is 17.1 W m-2. This is 2.5% of the mean. A significant part of the bias could be attributed to an offset in calculating the solar zenith angle. After correcting for this offset. the bias is reduced to 2.9 W m-2. which is 0.4% of the mean. Atmospheric constituents, e.g. aerosol concentration and water vapour that dominates the solar flux during clear skies, are highly variable in time and space. This results in a RMSE of 0.017 that is 2.3% of the mean BAT. This variability is not accounted for in the retrieval scheme in its current form. This is a limitation, but also marks the boundaries of the satellite retrieval.
The most important parameter determining the atmospheric transmissivity in cloudy atmosphere is the cloud optical depth. The variability in water vapour and aerosol concentration is less important in a cloudy atmosphere. The results of the overcast case show that discrimination between precipitation and non-precipitation cases is desirable. Comparison of measured and retrieved quantities show no correlation if precipitation was recorded. For the non-precipitation case, retrieval results are in agreement with the ground-based measurements. A correlation of 0.89 between retrieved and measured DSSI is found.

Comparing satellite measurements with ground-based observations is difficult in case of broken clouds. In all comparisons of this broken cloud case the problem of collocation occurs. In situ pyranometers measure continuously in time, while the AVHRR instrument measures instantaneouslv a spatial distribution. When comparing ground and satellite quantities, measured or retrieved, it is always questionable which part of the time series corresponds to which part of the spatial distribution.
The observed deviations between pyranometer measured and satellite retrieved DSSI are of fundamental nature, as the static view of a cloud field assumed in the retrievals does not reflect all aspects of real clouds. Broken clouds show significant variability in time and space (4D). Therefore, results of this case study are difficult to interpret within the boundaries of a 1D radiative transfer model, used in this thesis.