Hello,
Could someone more familiar with the GRAVSOFT package programs help me on the questions related to COVFIT:
· What values are appropriate to be chosen for model parameters – variance of gravity anomaly at zero altitude, scaling factor for degree variances and Bjerhammar sphere? I know there should be reading between the lines, but is there some tips to choose an appropriate values?
· What should be the difference between the covariance function from observed values and covariance function from the model (COVFIT)? Is it possible fitted function to be quite different from the ‘observed’ one – mean that the fitted function goes faster to zero from the first one?
· If we are using grid export as residual data in EMPCOV, we should use the same grid specifications for mean latitude and longitude spacing in COVFIT, yes? I can’t find information for grid spacing in EMPCOV, but I have read somewhere that such kind of information is existing in the .log file?
· Program DEGV.exe is used for calculating error degree variances for the global model, but the values produced are quite different from the one I’m calculating and wondering why. I use spherical harmonic coefficients from EGM08 and make summation of square of the errors of the coefficients – sqrt(sum(sigmaC^2 + sigmaS^2)). The file used in COVFIT should contain dimensionless error degree variances, not in dimension of the data used, yes?
Thanks in advance for the help!
for most algorithms that require starting values there is a theorem ( e.g. Gershgoren's circle theorem) that say you must be "close" to the value you seek because of the problem of local optima when you seek a global solution. Generally there are two ways to approach the problem. The one i try first is to use known values for the solution to a similar problem. Otherwise you can do something like a grid search. This is exactly like the problem of locating a downed aircraft. Use all the info you have available (don't forget graphing the problem). Always check multiple start sets so you min the chance of finding a local optimum. may the force be with you. D. Booth PS Oops forgot: the last reason is that you want the solution to be indept of the variable dimension. If one is in cm and one in km then any algorithm will make one have more weight that the other, so standardize your variables before you start. Further the more different two functions are the less likely they both describe the same thing. For specific GRAVSOFT info see: https://www.google.com/search?client=firefox-b-1-d&sxsrf=ALeKk008WFHcGE6ocnTSyCg5uDGDdaMFjQ%3A1605132579520&ei=I2GsX_ydH8GmtQbsp6OQBg&q=Gravsoft+software+manual&oq=Gravsoft+software+manual&gs_lcp=CgZwc3ktYWIQAzoECAAQRzoFCAAQzQI6BQghEKsCUILIAVjFugJg3b8CaABwAngAgAHDAogBgh2SAQgwLjIuMTIuMpgBAKABAaoBB2d3cy13aXrIAQjAAQE&sclient=psy-ab&ved=0ahUKEwi8u-DpwPvsAhVBU80KHezTCGIQ4dUDCAw&uact=5
Good luck and Best wishes, D. Booth
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero