When the global navigation satellite system (GNSS) is used for positioning in harsh environments with high occlusion and strong reflection, the signals are easily reflected, diffracted, or even blocked, which can lead to multipath and non-line-of-sight (NLOS) errors. Existing stochastic models do not fully consider the significant unmodeled errors, such as multipath and NLOS, in specific scenarios such as canyons. This paper systematically establishes the carrier-to-noise ratio (C/N0) template functions of the low-cost receiver for the first time and proposes an elevation stochastic model constrained by C/N0 (elevation-C/N0 model) by combining the two indicators (elevation angle and C/N0) with the idea of robust estimation. Then, real-time kinematic positioning (RTK) experiments are conducted to verify the effectiveness of the new model, and both static monitoring and urban kinematic situations are included. The results showed that: (a) in the static data, the average ambiguity fixed rate of the elevation-C/N0 model is 95.79%, which is 10.05%, 15.74%, and 12.57% higher than that of the equal-weight, elevation-based, and C/N0-based stochastic models, respectively. At the same time, only the new model consistently meets centimeter-level accuracy requirements in harsh environments. (b) In the kinematic data, compared with the three traditional stochastic models, the ambiguity fixed rate of the elevation-C/N0 stochastic model is increased by 58.33%, 19.79%, and 28.13%, respectively. The motion trajectory calculated by the elevation-C/N0 model is the smoothest out of all the models. In conclusion, compared with the traditional stochastic models, the elevation-C/N0 stochastic model is more applicable to both static and kinematic measurements in harsh environments. It can effectively mitigate the adverse effects of errors such as multipath and NLOS, increase the ambiguity fixed rate, and improve the positioning reliability to a certain extent, hence the proposed method has a better positioning performance.