Under a number of practical operating scenarios, traditional moving target indicator (MTI) systems inadequately suppress ground clutter in airborne radar systems. Due to the moving platform, the clutter gains a nonzero relative velocity and spreads the power across Doppler frequencies. This obfuscates slow-moving targets of interest near the "direct current" component of the spectrum. In response, space-time adaptive processing (STAP) techniques have been developed that simultaneously operate in the space and time dimensions for effective clutter cancellation. STAP algorithms commonly operate under the assumption of homogeneous clutter, where the returns are described by complex, white Gaussian distributions. Empirical evidence shows that this assumption is invalid for many radar systems of interest, including high-resolution radar and radars operating at low grazing angles. We are interested in these heterogeneous cases, i.e., cases when the Gaussian model no longer suffices. Hence, the development of reliable STAP algorithms for real systems depends on the accuracy of the heterogeneous clutter models. The clutter of interest in this work includes heterogeneous texture clutter and point clutter. We have developed a cell-based clutter model (CCM) that provides simple, yet faithful means to simulate clutter scenarios for algorithm testing. The scene generated by the CMM can be tuned with two parameters, essentially describing the spikiness of the clutter scene. In one extreme, the texture resembles point clutter, generating strong returns from localized range-azimuth bins. On the other hand, our model can also simulate a flat, homogeneous environment. We prove the importance of model-based STAP techniques, namely knowledge-aided parametric covariance estimation (KAPE), in filtering a gamut of heterogeneous texture scenes. We demonstrate that the efficacy of KAPE does not diminish in the presence of typical spiky clutter. Computational complexities and susceptibility to modeling errors prohibit the use of KAPE in real systems. The computational complexity is a major concern, as the standard KAPE algorithm requires the inversion of an MNxMN matrix for each range bin, where M and N are the number of array elements and the number of pulses of the radar system, respectively. We developed a Gram Schmidt (GS) KAPE method that circumvents the need of a direct inversion and reduces the number of required power estimates. Another unavoidable concern is the performance degradations arising from uncalibrated array errors. This problem is exacerbated in KAPE, as it is a model-based technique; mismatched element amplitudes and phase errors amount to a modeling mismatch. We have developed the power-ridge aligning (PRA) calibration technique, a novel iterative gradient descent algorithm that outperforms current methods. We demonstrate the vast improvements attained using a combination of GS KAPE and PRA over the standard KAPE algorithm under various clutter scenarios in the presence of array errors.