Beamspace domain super-resolution methods for elevation estimation in multipath environment has attracted significant attention. However, the difference beam is rarely used in super-resolution methods, especially in low elevation estimation. In order to improve the accuracy of beamspace domain maximum likelihood algorithm for elevation estimation, we apply the difference beam to the beamformer. Therefore, we develop a sum and difference beamspace domain maximum likelihood (SDBDML) algorithm for elevation estimation, which has good estimation accuracy and is not sensitive to antenna height and reflection coefficient errors. In order to further improve the estimation performance, we take the refined signal model into account in the SDBDML algorithm, and develop a sum and difference beamspace domain refined maximum likelihood (SDBDRML) algorithm. Compared with the SDBDML algorithm, the SDBDRML algorithm has higher estimation accuracy and lower computational burden, although it is sensitive to antenna height and reflection coefficient errors. The theoretical target elevation angle root mean square error (RMSE) and the computational complexity of proposed algorithms are analyzed. Finally, our computer simulations and real data processing results demonstrate the effectiveness of the proposed algorithms.