C hernykh et al. (2001, hereafter CAE) report trends in cloud-base height, cloud-top height, the number of cloud layers, and the frequency of clouds based on analysis of global radiosonde data during 1964-98. We are concerned that the changing vertical resolution of radiosonde observations over time, combined with the sensitivity of the method used to deduce cloud layers (Chernykh and Eskridge 1996, hereafter CE) to vertical resolution, undermine the credibility of the reported trends. This comment provides evidence both that the CE method is very sensitive to vertical resolution, and that the vertical resolution of soundings has increased over the past four decades. From this evidence we argue that the reported trends are, at least in part, artifacts of changing observing practices. We also raise other questions regarding the influence of data sampling on the results.The CE method for determining cloud layers from radiosonde temperature and humidity profiles involves identifying vertical layers in which the second derivative of temperature with respect to height is zero or positive, and the second derivative of relative humidity with respect to height is zero or negative. The endpoints of layers meeting these criteria are interpreted as cloud bases and tops. CAE apply this DOI: 10.1 I75/BAMS-84-2-237 AMERICAN METEOROLOGICAL SOCIETY method to data from 967 radiosonde stations and report a globally averaged increase in cloud-top height of about 154 m decade -1 , a decrease in cloud-base height of about 44 m decade -1 , a 1.7% decade -1 increase in the frequency of clouds, and an unquantified increase in the frequency of multilayer clouds.Given the implications of these findings for interpreting other changes in climate, it is important to determine whether they are robust. Changes in instruments and observing practices are manifest in time series of radiosonde temperature (Gaffen 1994; Parker and Cox 1995) and humidity (Elliott and Gaffen 1991) records and, if not accounted for, can severely undermine attempts to estimate temperature trends (Gaffen et al. 2000). Since the CE method infers cloud layers from temperature and humidity profiles, it is reasonable to question whether temporal data inhomogeneities also impact radiosondederived cloud trends.
SENSITIVITY OF CUBIC SPLINE INTERPO-LATION TO DATA RESOLUTION.Radio sonde data are reported at discrete pressure levels, resulting in a piecewise linear vertical profile. However, because piecewise linear functions have discontinuous first derivatives, the CE method involves fitting a cubic spline to sounding data to allow evaluation of the second derivatives of temperature and humidity with the respect to height. As an aside, we note that spline fits are contradictory to the reporting requirements for radiosonde data, which stipulate that a data level be reported when the soundings departs significantly from a linear fit between reported levels (Hooper 1986; WMO 1996; OFCM 1997). Although it seems likely that atmospheric profiles have continuous second derivatives...