Global navigation satellite system (GNSS) velocity fields generally contain outliers due to environmental interference, local crustal activities, or strong seismic activities, which significantly affect the parameter estimation of the crustal deformation model. The robust M estimation can be used to improve the accuracy of the parameter estimation while retaining as much GNSS observation information as possible. The current robust estimation still uses the least-squares residuals instead of real errors as the initial values of the equivalent weight function and the breakdown point of parameters estimation is less than 50%. First, a new automatic selection strategy is proposed for quasi-accurate observations (observations that are reliable and do not have outliers but require confirmations) using quasi-accurate detection, and the outliers are roughly identified, almost independent of the breakdown point. Second, the variance of the unit weight of the estimate of real error of quasi-accurate observations is used as the initial variance factor and the estimate of real error of quasi-accurate observations is used as the initial value for the iterative calculation of the robust M estimation to accurately identify and detect large, medium, and small outliers. We evaluate the rigorousness of the least-squares estimation, conventional robust estimation, median method, and new algorithm from four perspectives using simulated GNSS velocity fields, i.e., model parameter estimation, variance factor, distribution of the observation weights, and accuracy of forward deformation fields. The new algorithm effectively eliminates outliers and estimates the model parameters. When outliers are below 50%, the robustness of the new algorithm is better than that of the other three. When the proportion of outliers is higher than 50% (60%, 70%, and 80%), the other three algorithms break down, while the new algorithm yields better parameter estimates with an increasing breakdown point.

由于环境干扰、局部地壳活动或强烈地震活动，全球导航卫星系统（GNSS）速度场通常包含异常值，这显着影响地壳变形模型的参数估计。稳健的 M 估计可用于提高参数估计的准确性，同时尽可能多地保留 GNSS 观测信息。目前的稳健估计仍然使用最小二乘残差而不是实际误差作为等效权重函数的初始值，并且参数估计的击穿点小于50%。首先，针对准准确观测（可靠且没有异常值但需要确认的观测）使用准准确检测提出了一种新的自动选择策略，并且粗略地识别异常值，几乎与故障点无关。其次，以准准确观测实际误差估计的单位权重方差作为初始方差因子，以准准确观测实际误差估计为初始值，迭代计算稳健性M 估计以准确识别和检测大、中和小异常值。我们使用模拟的GNSS速度场，从模型参数估计、方差因子、观测权重分布和前向变形精度四个角度评估最小二乘估计、常规鲁棒估计、中值法和新算法的严谨性。领域。新算法有效地消除了异常值并估计了模型参数。当异常值低于 50% 时，新算法的鲁棒性优于其他三种算法。当异常值的比例高于 50%（60%、70% 和 80%）时，其他三种算法都会崩溃，而新算法会随着崩溃点的增加产生更好的参数估计。