Document Type : Research Article
Authors
1
Department of Geographical sciences, Faculty of Humanities, University of Hormozgan
2
Bandarabas, Iran
3
Department of Geographical sciences, University of Hormozgan
Abstract
One of the current weather phenomena that has affected Iran nationwide is the dust storm. Hormozgan providence, located in the country's south (in the global arid and semiarid region), is prone to wind erosion and dust storms due to its proximity to the Central Persian and Arabian deserts and its lack of vegetation cover. According to a review of the literature, earlier studies on simulating dust storms in the Hormozgan region have primarily used MODIS products on a daily time scale for the study area. The aim of this study was to model dust storms using spectral indices, such as the Normalized Difference Dust Index (NDDI) and Brightness Temperature Difference (BTD), based on the Artificial Neural Network (ANN) and Random Forest (RF) methods, two of the most well-known and effective machine learning techniques in the modeling and prediction fields, on an hourly time scale using Spinning Enhanced Visible and InfraRed Imager (SEVIRI) METEOSAT images.
The indicators were computed using METEOSAT images during the selected dates (21–24 November 2016 and 3–9 December 2016) for dusty days with visibility of less than 1000 m (8 images per day). In order to model and predict dust storms, the Artificial Neural Network (ANN) and Random Forest (RF) methods were used. NDVI and BT were used as dependent variables, and Air Temperature (AT), Wind Speed (WS), Air Pressure (P), and Absolut Humidity (AH) extracted from NASA GES DISK were used as dependent variables. Time scales for SIVIRI images and NASA GES DISK climate reanalysis data were 4 and 3 hours, respectively. Climate reanalysis data were extrapolated to 4 hours each day. The findings demonstrate that the NDDI's Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R2) were, respectively, 0.08, 0.31, and 0.24 based on ANN. While the BT index values were 0.42, 6 and 4.6, respectively. Based on the RF technique, the NDVI model had R2, RMSE, and MAE values of 0.55, 0.3, and 0.23, whereas the corresponding values for BT were 0.69, 4.4, and 2.8. The results show that RF models combined with climate reanalysis data have a good performance in modeling and predicting dust storms in the Iranian province of Hormozgan. BT index generated from SIVIEI images with a 4-hour time resolution and RF models are also of outstanding performance. further research is needed to evaluate the performance of The method used in this research in other regions of Iran.
Keywords
Main Subjects
)