The effectiveness of boundary integral equation and grid-free methods for radial grids, as well as solutions of classical and nonclassical modeling and recovery problems for geological fields are analyzed. It is shown that, as opposed to the methods employing a variational technique and radial basis neural networks, hybrid algorithms (fuzzy neural networks, genetic algorithms, and Kalman filter) for solving identification and recovery problems are more stable with respect to noise and give positive results even with conflicting data and significant measurement noise.