Power spectrum estimation from radar data is essential for target detection. For instance, microburst causes detrimental effects on airplane performance, and hence its detection is critical. We compare auto-regression (AR), Periodogram, Kaiser windowed Periodogram, and multiple-signal-classification (MUSIC) methods for microburst clutter spectrum estimation. Given a long train of returned signal, we are able to segment the signal to obtain multiple estimations of the parameters, which leads to a more accurate estimation after coefficient averaging. The estimated power spectrum is then integrated for clutter magnitude calculation to determine whether a target is present at certain cell. The magnitude of clutters in an ensemble from a wide region spatially or through time temporally is used to estimate the clutter map. We choose a K-distribution mixture model over the traditional Rayleigh distribution to better approximate the tail structure of the distribution to minimize the false alarm rate. We show that Dimension-Adaptive Particle Swarm Optimization (DA-PSO) is robust to sample size in estimating the K-distribution mixture model, which is desirable for real-time implementations.