Using data from 385 events recorded at Lanzhou in central China, I compare two empirical methods for predicting and correcting for path effects on regional-phase amplitude ratios. The first method interpolates source- and distance-corrected seismic data to create geographical, path-correction surfaces. The second, or wave-guide, method employs correlations with path-averaged physical data such as topography to correct the seismic data. Path corrections can vary over an order of magnitude, and their use will sharpen differences in regional-phase ratios arising from different types of natural and man-made sources of seismic energy and, thus, will be important for monitoring a Comprehensive Test Ban Treaty.
For Pg/Lg phase ratios in the band 0.5 to 1 Hz, interpolated path corrections reduced variance best using kriging (62%) and nearly as well using a radius 250 km, moving-window-mean smoother (56%). Wave-guide methods based on mean and rms topography reduced variance by 43%. Variance reduction is given relative to the standard correction for distance alone. Wave-guide methods were further tested by attempting to find correlations with topographic data that had been rotated geographically. For conservatively chosen sets of wave-guide parameters, the true orientations yielded the best variance reduction, adding confidence to the use of this method. Wave-guide methods complement interpolation methods because they extrapolate behavior beyond active seismic areas. However, I find broad regions that are fit poorly by wave-guide predictions, especially where the path effect changes rapidly, such as along the eastern edge of the Tibetan plateau where Pg/Lg ratios are underpredicted. The two methods can be combined using Bayesian kriging, forcing predictions to fit the data in seismically active areas and allowing more powerful extrapolation beyond.