Sensitivity analysis of cirrus height, temperature, and pressure retrievals using remotely sensed thermal infrared observations

Physical properties of transmissive cirrus clouds can be retrieved from remotely sensed measurements via a two-step process. First, cloud emissivity and temperature are estimated using brightness temperature measurements near 3.7 and 10.8 um, accounting for a temperature-dependent relationship between ice crystal size and spectral emissivity. The method requires estimation of equivalent cloud-free brightness temperatures for cloudy image pixels, to account for the radiance transmitted through thin cirrus. Second, cloud height and pressure are estimated by comparing the retrieved temperatures with coincident profile measurements.

The accuracy and precision of these retrievals depend on 1) sensor calibration errors and noise, 2) inaccurate estimation of clear-scene brightness temperatures, and 3) errors in measured profiles of temperature and pressure. The impact of all three error sources on cloud top retrievals was assessed using a scene simulation study. Images containing cirrus clouds of varying temperatures and optical depths were created by superimposing a simulated cloud field upon an otherwise cloud-free AVHRR image. Top-of-atmosphere radiances were determined using a simplified radiative transfer model. Use of simulated images with realistic spatial variability of surface and cloud properties allowed assessment of the impact of errors in clear-scene radiance estimates derived from interpolation of cloud-free radiances. The impact of sensor errors was assessed by adding calibration biases and random noise with a Noise Equivalent Difference in Temperature (NEDT) of 3.0 K in the 3.7 um channel and 1.0 K in the 10.8 um channel. Profile errors were investigated by perturbing standard profiles according to errors typical of retrievals from infrared sounders.

Of all sources of error, results are most greatly affected by inaccurate estimates of clear-scene radiance. Problems due to sensor noise are small because of the pixel averaging implicit in interpolation from several cloud-free observations. Errors in retrieved profiles affect retrieved cloud height and pressure, but their importance is small compared to the effects of inaccurate cloud top temperature estimates.

Sensitivity is greatest when clouds are thin (optical depths < 1.0) due to the large contribution from surface radiance. Clouds which are high and cold are particularly problematic. Temperature errors as high as 10 K occur for these most stressing cases. For thicker cirrus clouds (optical depths of about 3.0), errors of 1 -- 2 K can be obtained.