Influence of clouds on the solar radiation budget

This thesis investigates the accuracy of a current radiative transfer scheme and a single. column model for representing the influence of clouds on the solar radiation budget. An extensive set of pyranometer measurements and satellite observations collected throughout the Netherlands is used for this purpose.
First, an overview of the relevant parts of the theory of atmospheric radiative transfer is given and a literature study of the instrumental uncertainties of the dataset is presented. Special attention is paid to the thermal offset of pyranometers, which has recently been identified as the origin of a significant systematic underestimate of measured solar irradiances. A mechanism is proposed whereby the thermal offset can lead to additional calibration eITorS not accounted for in previous estimates of the pyranometer uncertainty. In combination with the direct effects of the thermal offset, a large range of systematic pyranometer errors has to be expected, which significantly exceeds the range specified by the World Meteorological Organization for these type of instruments.

A correct modeling of clear-sky atmospheric radiative transfer is an essential prerequisite for a correct treatment of cloudy atmospheres in radiative transfer models. To evaluate its accuracy, six years of global surface irradiance records obtained throughout the Netherlands during cloud-free situations are compared to model output of five different radiative transfer schemes, providing 6,132 individual cases. Two models calculate a significantly lower atmospheric absorption by gases in comparison to the other models. This is probably the consequence of the older spectroscopic datasets used for the parameterization of gas absorption. The three models based on more recent spectroscopic data show good agreement in radiative fluxes at the top of atmosphere and the surface. However, large uncertainties are introduced by the limited knowledge of aerosol optical properties for the current dataset. Compared to pyranometer measurements, the recent models overestimate the atmospheric transmissivity by 4.5% on average throughout the full range of solar zenith angles. This causes a mean overestimate of the global irradiance of 27 W m-2. Systematic error sources in the measurements and the model assumptions are large enough to explain the full magnitude of the discrepancy.
Therefore, the accuracy of the data is too limited to identify any shortcomings in the models or in the representation of aerosols used.

The main parameter determining the reduction of atmospheric transmissivity in cloudy conditions is the cloud optical depth. Retrieval results of two algorithms to infer the cloud optical depth from transmitted broad-band and reflected narrow-band solar radiation are compared in Ch. 5. The algorithms are applied to a six-year dataset of about 40,000 pyranometer measurements collected by 32 meteorological stations in the Netherlands and to 1.500 simultaneous and cofocated observations of the 0.63 10-6m radiance inferred by the AVHRR instrument on board the NOAA-14 satellite. Both a strong negative bias of 22 - 46% and a large random scatter are found for satellite retrieved versus pyranometer-retrieved cloud optical depths. While the magnitude of the scatter is reduced if a selection of cases is made to restrict horizontal cloud in-homogeneity, no such dependence is present for the bias. Quantifying the impact of measurement uncertainties, it is possible to attribute the full bias to systematic errors in satellite calibration and pyranometer measurements. This finding highlights the importance of high instrumental accuracy for trustworthy retrieval results. Furthermore, the assumptions inherent in a comparison of a spatially extended instantaneous measurement to a temporally averaged time series of a point measurement are shown to contribute substantially to the observed scatter and to introduce a small bias.

Based on the methods used for the previous comparison, an algorithm is presented in Ch. 6 to derive the downwelling solar surface irradiance from satellite estimates of the 0.63 pm reflectance, which explicitly accounts for variations in cloud optical depth and integrated water vapor. For validation, the dataset of 40,000 pyranometer measurements and 1,500 NOAA-14 AVHRR satellite scenes is used. A mean overestimate of the satellite-retrieved irradiance by 7 % is found, which is consistent with numerous other studies reporting positive biases of atmospheric transmissivities calculated by radiative transfer schemes in comparison to measurements. If a water cloud is assumed in the retrieval, a strong dependence of retrieval bias on solar zenith angle is found. The information available for the present study is insufficient to find a full explanation for this behavior. Using 40-minute averages of pyranometer measurements and a region of about 150 km? for the satellite analysis, a good performance of the retrieval is found for monthly averages and averages of all stations for individual satellite overpasses, resulting in RMSEs of 12 W m-2 and 33 W m-2, respectively. A comparison with individual measurements shows a much larger scatter of 85W m-2. Evidence is presented which suggests that a significant fraction of the scatter originates from the variability of the irradiance field caused by cloud inhomogeneities.
The study presented in Ch. 7 evaluates the representation of humidity, cloud amount and shortwave atmospheric transmissivity in a single column model derived from the ECHAM4 GCM in comparison to radiosonde profiles, pyranometer measurements and synoptic reports. The model captures only part of the observed variance of relative humidity in the middle troposphere, and only about 40 % of the variance is explained by the model forecasts. A mean underestimate of total cloud amount by 14 % is found in comparison to synoptic reports. In the presence of clouds, the atmospheric transmis sivity is strongly underestimated, indicating a general overestimate of the cloud optical depth. The model skill for predicting both the cloud amount and the atmospheric transmissivitv on a case-by-case basis is found to be rather low. Similar types of shortcomings are typical for the current generation of climate models and highlight that the representation of clouds and cloud-radiation interactions need to be significantly improved for an accurate simulation of the climate system.