Some general points you can consider:

1) Look how the retained model fits the seismic data.

2) Examine the continuity of the solution and see if fault pattern make sense.

3) What is the residual misfit of the retained model.

4) What are the confidence level attributes say.

5) Do the datasets all point to a similar solution, are they giving a stable answer,  and/or do they give complementary info, refining the model.

6) Does the retained solution make geologic / petrophysical sense.

Results also depend on the way the joint inversion is done: simultaneous or sequential. Simultaneous is of course better but requires more computing power. 

wbr Paul

A rigorous joint inversion will perform at least as well as the best resolved single method it includes. For example if you have very dense and high quality seismic data and you jointly invert it with sparse gravity data, the joint inversion should fit the seismic data equally well or better than if you were inverting the seismic data alone. By adding the gravity data you are adding information to the system so you should improve the resolving power. Of course it will depend on the specifics of the joint inversion method.

To evaluate the quality of the solution you will have to perform all the usual tests. Paul has suggested a few already above. In addition you could try checkedboard tests, anomaly recovery tests, jack knifing etc ... There are other questions here on ResearchGate that address these briefly.

Michele is absolutely right, more data mean more information and can potentially improve your resolution (if the information is complementary and compatible). For some more detailed discussion with a very simple illustrative example, you can look at the draft of a review paper I am working on http://www.complete-mt-solutions.com/mtnet/division/papers/EMWKSHP_ReviewVolumes/2016ChiangMai/EMIW2016_Keynote_Moorkamp_preprint.pdf   There I show in some detail how different ways of combining data can help to increase the resolution and how to identify problematic situations where joint inversion is not applicable.

Joint inversion permits to more accurately derive subsurface structures provided that all source of information are reliable. In other words, if one data set is noisy and unreliable, it will deteriorates the final results and it should be removed from the entire inversion process. But in general, due to including more source of information, joint inversion can resolve subsurface structures with more resolution than the single inversion. Moreover, you should consider misfit between measured and modeled data after single or joint inversion. The result with lower misfit usually consider as a best result, however you should be careful about the over fitting of your data (sometimes a balance between misfit and smoothing should be considered). Furthermore, sensitivity analysis can also help you to evaluate the final results. For gradient-based inversions, singular value decomposition (SVD) and for global-inversion techniques, topography of the objective function with respect to the unknown parameters are evaluated for sensitivity analysis of the inversion:

http://onlinelibrary.wiley.com/doi/10.1111/1365-2478.12281/full

Just check, for example, the integral sensitivity or resolution distribution over your models and compare for single method inversion and joint. You will see where model is well resolved and where is not.

you are right if you are overconfident in data with poor sensitivity, i.e. if you underestimate error/correlation in your data. If your error statistics is accurate on both datasets, more information translates in better resolution.

Joint inversion can increase the numbers of relevant parameters of a model in which the individual method cannot resolve separately. In the resolved inversion there is element of goodness of fit and also there is misfit. Correlation of the retained model in joint inversion and individual inversion of data can give a clue to the evaluation as this informs about the level of resolution

I think the best way to evaluate the performance, is to do the following:

1. Examine and evaluate each method separately and extensively.

2. Identify the strength and weaknesses of each inverse problem.

3. Compare your objectives/goals with (2), above.

4. Re-evaluate analyze the problem with each method separately.

5. Constrain the weaknesses of the iterations with the strength of the other method and vice versa.

6. Compare the results. Sometimes optimisation method p

Hi,

I agree with M. Moorkamp . The question is how different combinations of the data (to be jointly inverted) can improve the results? In the case multi unknown parameters, one may analyze the sensitivity of the data to each one. The objective function itself is important. One may apply a constraint regularization to inject primary information into the problem or stabilize the inversion. To assess the results based on different scenarios, you may perform uncertainty quantification. For example, see the following papers:

https://academic.oup.com/gji/article-abstract/218/1/666/5420754?redirectedFrom=fulltext

Article Uncertainty and Resolution Analysis of 2D and 3D Inversion M...