Global Navigation Satellite Systems (GNSS) provides critical technical support for meteorological disaster warning and monitoring through real-time atmospheric data observation. Tropospheric delay, as one major errors source of GNSS, has been applied in meteorological disaster prediction. Understanding tropospheric delay variations is essential for improving GNSS-based disaster monitoring systems. Traditional prediction models rely heavily on empirical formulas, resulting in insufficient accuracy in their predictions, and some of them require measured meteorological data as input or large dataset for training. To deal with these issues, we propose a GNSS tropospheric delay prediction model based on Group Method of Data Handling (GMDH). By utilizing simple input parameters including latitude, longitude, altitude, and day of the year, zenith tropospheric delay (ZTD) is predicted with the GMDH model. Experimental results indicate that the proposed model outperforms Saastamoinen model, UNB3M model and LSTM. In Experiment 1, using the same 12 stations for both research and model training, the proposed model delivers ZTD prediction accuracy improvements of 47.53%, 56.98% and 37.58% compared to Saastamoinen model, UNB3M model and LSTM. In Experiment 2, using an additional 6 stations beyond those used for training, the improvements are 9.71%, 33.73% and 7.80%, respectively. Compared to LSTM, The running time is reduced by 5.00% and 4.14%, respectively.