Using oceanic wind data, effects of nonstationary events on the spectral computation of wind speed and wind direction were estimated by analysing stationary records before and after superimposing spike and shift nonstationarities. The differences of the results show that the inclusion of nonstationarities in environmental sampling can produce large errors in the data and their analysis. It is concluded that specialized observational techniques are required for nonstationary time series for any efficient environmental sampling system. A hypothetical sampling system is proposed which provides for filtering the stationary environment as described in Part I and continuous checking for the occurrence of nonstationarities. Significant nonstationarities are reported separately and not included in stationary processing. As an example of a nonstationary detector, a simple automated method to detect and report wind shift nonstationarities was developed. It is concluded that not only does the proposed sampling system improve the synoptic observation value, but it also satisfies the special observation requirement for weather transmissions. It is further stated how such a system allows an increase of the minimum spacing between observation stations, thus increasing the cost-effectiveness of the observational network.