Resource information
Dust emission and deposition are associated with several factors such as surface roughness, land cover, soil properties, soil moisture (SM), and wind speed (WS). A combination of land surface and remote-sensing models has recently been investigated for dust detection and monitoring. The thermal bands of the Meteosat Second Generation Spinning Enhanced Visible and Infrared Imager (MSG/SEVIRI) satellite are widely used for qualitative detection of dust over desert because of their high spectral and temporal resolutions. In this work, the contribution of ground-measured WS data and satellite-measured SM data on aerosol optical thickness (AOT) retrieval was investigated using an artificial neural network (ANN) model. ANNs have been applied in similar applications and have shown a higher performance than simple multiple-regression models. This performance is mainly due to the ANN's ability to capture complex and non-linear relationships between inputs and outputs. A combination of MSG/SEVIRI brightness temperature (BT)/brightness temperature differences (BTDs), BTD₃.₉–₁₀.₈, BTD₈.₇–₁₀.₈, BTD₁₀.₈–₁₂, and BT₃.₉, was used as input to the base ANN model while Aerosol Robotic Network (AERONET) AOT (level 2) data at 0.5 μm were used as output. These input/output sets were obtained from two stations (Hamim and Mezaira) lying in the inland desert of the United Arab Emirates (UAE). About 3800 observations were collected, of which two-thirds were used to train the ANN model and the remaining third was kept as an independent set to assess the accuracy of the trained model. Later, Advanced Microwave Scanning Radiometer Earth Observing System (AMSR-E) SM data and ground-measured WS data were used as additional inputs to the base model to investigate their contribution to the AOT retrieval. SM data consist of daytime AMSR-E-derived daily and collected from a National Snow and Ice Data Centre (NSIDC)-archived database. Hourly average WS data were also collected at 10 m height in the same AERONET sites from two stations managed by the UAE National Centre of Meteorology and Seismology. All ground and satellite measurements were extracted for the closest time to AERONET measurements. The use of these additional inputs has been shown to have a positive impact on the accuracy of simulated AOT. The addition of these inputs to the base ANN increased R² from 0.68 to 0.76 and reduced root mean square error from 0.113 to 0.09.
