نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی احیاء مناطق خشک و کوهستانی، دانشکده منابع طبیعی، دانشگاه تهران
2 استادیار، گروه مهندسی احیاء مناطق خشک و کوهستانی، دانشکده منابع طبیعی، دانشگاه تهران، کرج، ایران.
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Dust storms are considered a major type of extreme weather phenomenon, particularly in arid and semi-arid regions, and have significant impacts on health, the environment, and natural resource management. Accurate forecasting of this phenomenon remains a significant challenge. This study investigated the capability of modern deep learning models for predicting dust storms by comparing the performance of two individual models, Autoformer and FNO, and a hybrid framework based on combining both models for different time scales. To investigate the impact of the temporal structure of the inputs, four temporal combinations were formed, encompassing short-term to long-term forecasts, representing one season to one year ahead. According to the results obtained, the Autoformer model recorded the weakest performance. The FNO model, although showing better performance in detecting spatio-temporal patterns compared to Autoformer, did not show a significant difference and recorded a relatively weak performance. In contrast, the FNO-Autoformer hybrid model clearly provided the highest prediction accuracy across all evaluation metrics. High values of the NS index, a significant reduction in RMSE and MAE errors, as well as high correlation, indicated the better performance of the combined models compared to the individual models in predicting extreme weather phenomena. Furthermore, the findings of the present study proved that, in order to predict dust storms for points where an external factor causes the dust, short-term temporal combinations, especially the prediction combination with a one-season lag, perform best, and the prediction accuracy decreases with increasing time lag.
کلیدواژهها [English]
)