Application of Global Dust Detection Index (GDDI) for Sand and Dust Storm Monitoring Over Kingdom of Saudi Arabia


 Sand and dust storm events are frequent natural hazards in the Kingdom of Saudi Arabia. Sand and dust storm monitoring is therefore essential to mitigate their environmental-related issues. Satellite remote sensing has been successfully used for sand and dust storm monitoring in various parts of the world. In the current endeavor, we are applying the Global Dust Detection Index (GDDI) on Moderate Resolution Imaging Spectroradiometer (MODIS) data onboard Terra satellite to monitor sand and dust storm activities over the Kingdom of Saudi Arabia. In the current study, fourteen sand and dust storm events are analyzed between the years 2000 to 2017. The GDDI based results are validated by using MODIS combined Dark Target (DT) and Deep Blue (DB) Aerosol Optical Depth (AOD) product, Meteosat satellite images, ground-based meteorological stations data, and AOD data from AERONET (Aerosol Robotic Network) stations in the study area. Also, GDDI based results are analyzed by determining algorithm accuracy, Probability Of Correct positive Detection (POCD), and Probability Of False positive Detection (POFD). Results of the study show that GDDI can successfully identify sand and dust storm events with various threshold values over the Kingdom of Saudi Arabia. It is envisaged that the outcome of this study would be very beneficial to understand sand and dust storm characteristics in the study region.


Introduction
Sand and dust storm events are frequent over arid and semi-arid regions of the world and are major contributors to the concentration of aerosols in the atmosphere. Sand and dust storm event has great impacts on environmental dynamics (Tegen et al. 2002;Jickells et al. 2005;Jones & Christopher 2008) as they can scatter and/or absorb electromagnetic radiation and thereby in uence the clouds microphysics and processes of atmospheric chemistry by modifying the ozone level. Sand and dust storm events can indirectly affect the weather systems of the globe and can also trigger global warming which has already received much attention worldwide (IPCC 2007;UNEP 2007;UNEP, WMO 2016;; UN 2019) due to their potentially catastrophic impacts (Yu et al. 2013;Hu et al. 2014). Thus, sand and dust storm events are one of the imperative environmental apprehensions being faced by both the developed and developing countries of the world. It is therefore crucial that sand and dust storm events should be monitored on regular basis to keep track of the related global environmental issues and adverse health concerns. The Kingdom of Saudi Arabia, which has some of the largest deserts of the world, is essentially vulnerable to the sand and dust storm events thus making their study an area of major concern among environmental scientists. The research on sand and dust storm event, however, has hampered as the spatial coverage of the meteorological stations are quite sparse on the ground in the country. Although, remote sensing, at present, can only provide near-global horizontal coverage with limited vertical resolutions , increasing concerns about environmental protection has caused the scientist to use satellite data for sand and dust storm monitoring Metcalfe and Buttle 1998;Johansson, Schmetz et al. 2002;Norman, and Gidhagen 2007;Schepanski et al. 2007;Martínez et al. 2009;Schepanski et al. 2012;Sannazzaro et al. 2014). This research aims to address sand and dust storm monitoring over the Kingdom of Saudi Arabia by using satellite and ground-based stations data.
During the last few decades, remote sensing has proven its major role for sand and dust storm monitoring in different parts of the world Schepanski et al. 2012;Sannazzaro et al. 2014;Butt et al. 2017;Butt & Mashat 2018). Remote sensing technology has great advantages over traditional measurements due to the extensive global coverage, satisfactory spatial resolution, and repetitive temporal resolutions (Scarratt et al. 2007;Jones & Christopher 2008). Many research groups have used data from various satellites and sensors for the estimation of aerosols quantity in the atmosphere as well as their relationship with sand and dust storm events (Washington et al. 2003;Kerkmann et al. 2005;Sabbah & Rybanský 2006;Li et al. 2007;Sabbah & Hasan 2008;Hutchison et al. 2008;Hoff & Christopher 2009;Badarinath et al. 2010;Muhammad Akhlaq et al. 2012;Ghasem et al. 2012;. Similarly, researchers have also used geostationary satellite data (for example, Meteosat satellite and Geosynchronous Meteorological Satellite) for sand and dust storm studies in different parts of the world (Liu & Lin 2004;Sannazzaro et al. 2014). However, various errors, especially around regions with low cloud cover (due to similar albedo for dust particles and neighborhoods of clouds) still occur during sand and dust storm monitoring.
Moderate Resolution Imaging Spectroradiometer (MODIS) onboard both Terra and Aqua satellites offer remotely sensed data through its 36 bands in the elds of Land surface, Cloud, Aerosols, Ocean Color Phytoplankton Biogeochemistry, Atmospheric Water Vapor, Atmospheric Temperature, and Ozone. MODIS data can be dedicatedly employed for the study of the environment over a region or the globe. ) developed the Dark Target (DT) method, which was later on modi ed by (Levy et al. 2010) in order to monitor atmospheric aerosol concentration.  found that the DT technique is unsuccessful over bright re ective surfaces (for example, the Kingdom of Saudi Arabia) and leave a big spatial gap in the estimation of Aerosol Optical Depth (AOD). Thus, the global AOD gap was lled by using Cloud-Aerosol Lidar, the Infrared Path nder Satellite Observation (CALIPSO), and the Multi-Angle Imaging Spectroradiometer (MISR) sensors data (Winker et al. 2009) until the development of MODIS Deep Blue (DB) technique proposed by (Hsu et al. 2006). Studies have been conducted which showed that MODIS DB product can successfully produce AOD maps with acceptable uncertainties over bright surfaces (Shi et al. 2011;Li et al. 2012;Shi et al. 2013;Butt et al. 2017). Recently, MODIS combined DT and DB product has also been prepared to produce AOD spatial distribution maps over dark and bright surfaces and is used in the current study.
For sand and dust storm monitoring, MODIS data has also been used successfully by various research groups for the development of algorithm (Li et al. 2007;Martínez et al. 2009). For example, ) presented the Normalized Difference Dust Index (NDDI) using MODIS data to detect sand and dust storm over China. They proposed different threshold values for the detection of various features along with sand and dust storm over the Gobi desert. Researchers have applied NDDI for the detection of sand and dust storm in different parts of the world and found that a slight adjustment in the threshold values is required from that proposed by . For example, in the detection of sand and dust storm by using NDDI, (Li & Song 2009) reported slightly lower threshold values over China whilst (Wu et al. 2006) estimated larger threshold values over the ocean than that proposed by . Similarly, (Butt & Mashat 2018) adjusted a much less threshold value for the detection of sand and dust storm over the Kingdom of Saudi Arabia with the help of NDDI. (Albugami et al. 2019) however concluded that threshold values of NDDI increase from 0.25 to 0.4 from the year 2000 to 2005. For the detection of sand and dust storm, (Samadi et al. 2014) proposed a new algorithm termed as Global Dust Detection Index (GDDI) using MODIS re ective and thermal Infrared bands data and applied it over Iran. They concluded that sand and dust storm events can be successfully discriminated from other features without any threshold values, both over land and water surfaces. Recently, (Zandkarimi et al. 2020) applied GDDI over Iran with some improvements in the detection of sand and dust storm in the region of study. Interestingly, they also did not propose any threshold values for the discrimination of various features. Many studies have been conducted in the past, which showed that MODIS data is highly suitable for sand and dust storm monitoring Ignatov et al. 2005;Remer et al. 2005;Butt & Mashat 2018).
In the current research, GDDI is applied on MODIS data over the Kingdom of Saudi Arabia and threshold values are proposed to discriminate sand and dust storm from other features. GDDI based results are validated by using in situ data from 28 ground stations in the Kingdom of Saudi Arabia as well as data from AERONET (Aerosol Robotic Network) stations in the country. In addition, the model-based results are also veri ed by using MODIS combined DT-DB products and Meteosat geostationary data over the study region. Finally, the performance of the GDDI is analyzed by calculating algorithm accuracy, Probability Of Correct positive Detection (POCD), and Probability Of False positive Detection (POFD).

Study Area
The study area, the Kingdom of Saudi Arabia, as shown in Fig. 1, is surrounded by the world's extremely arid regions that are also the main source of sand and dust storm events affecting the country. For example, deserts in Iraq, Jordan, and Syria are in the north, Iranian deserts in the northeast, deserts in UAE in the east, deserts in Yemen and Oman in the southeast, and deserts in African Sahara in the southwest of the country. In addition, the country also has some of the largest deserts in the world (for example, Rab al Khali, An Nafud, and Ad Dhana). Geographically less than 1 % area of the country is feasible for agricultural growth whilst the remaining area is arid or semi-arid. Climatologically, the country's annual average temperature is 25.4 ºC with less than 75 mm of annual rainfall. The Aziab wind (prefrontal when passes over Sahara desert and deserts in northern Saudi Arabia raise a large amount of dust) and Shamal wind (postfrontal slow-moving/semi-stationary low pressure as well as strong pressure gradients systems that can produce sand and dust storm events lasts from 2-5 days) are the major synoptic conditions over the country. Similarly, the dominant pressure systems (Sudanese Low, Siberian High, Indian Monsoon Low, and Azores High) are among the major contributor to the development of sand and dust storm events over the study area. For example, Siberian High Ridge is the most dominant system in the spring season (more dynamic escalation to strong southerly winds carrying hot air that causes sand and dust storm events) whist the Indian Monsoon Low and Azores High Ridge are the most dominant pressure systems during summer (produces a strong pressure gradient causing Shamal winds leading to the formation of sand and dust storm events). On the other hand, Siberian High, Indian Monsoon Low, and Azores High Ridge are playing an important part during autumn (with a weak pressure gradient located over north Africa resulting in a stable condition in the autumn season and thereby very few sand and dust storm events) while the most dominant pressure systems in winter are, Sudanese Low and Siberian High Ridge (these pressure systems are not vibrant enough for the development of sand and dust storm events).

Material
Near real-time detection of sand and dust storm events over a region can enable us to take precautionary measurements to minimize the related hazards. Thus, sand and dust storm events over the Kingdom of Saudi Arabia are monitored using MODIS data onboard the Terra satellite. Terra's orbit around the Earth is timed so that it passes from north to south across the equator in the morning. Terra is viewing the entire Earth's surface every 1 to 2 days, acquiring data in 36 spectral bands thereby can be very useful depending upon the data availability and application requirements and the research interest. Multispectral bands (36 bands) of MODIS ranged from 0.415 µm to 14.235 µm are an additional advantage in the environmental-related applications. The spatial resolution of MODIS is band dependent ranging from 250 m to 1 km. MODIS (MOD021KM) satellite data can be downloaded free of cost from https://modis.gsfc.nasa.gov/data/.
For the selection of MODIS sand and dust storm events, we have used station data in this study. The General Authority of Meteorology and Environmental Protection (GAMEP) is responsible to record and store various meteorological data in the Kingdom of Saudi Arabia as per the World Meteorological Organization (WMO) standards. Sand and dust storm data from 28 ground stations working under GAMEP is retrieved and assessed. Fourteen sand and dust storm events selected from GAMEP data are analyzed in the current study for which the MODIS images are available. Detail of the MODIS images used in the current study is given in Table 1. Table 1 Description of GAMEP station data, MODIS satellite data, MODIS DT-DB AOD product, Meteosat satellite data, and AERONET AOD data used in the current study.

Methodology
Considering the limitations in various developed methodologies (detection threshold change with the season, type of land, and the problem of discrimination from clouds, water, and land soil) (Samadi et al. 2014) developed GDDI using spectral signatures of sand in MODIS band 4 and band 7 for the detection of sand and dust storm events. In the development of GDDI (Samadi et al. 2014) collected training pixels of sand and dust storm over dark surfaces, bright surfaces, and water bodies from MODIS data. By using MODIS images from the Terra satellite they prepared the land cover map of three mainland cover classes (bright land cover, dark land cover, and water bodies) in the study area. They calculated the statistical mean and standard deviation for seven classes (clouds, bright surfaces, dust over bright surfaces, dark surfaces, dust over dark surfaces, water, and dust over water). Finally, they developed a relationship between band 4 and band 7 in order to differentiate dust from no-dust pixels over bright surfaces as given in equation 1.
where B4 and B7 are re ectances in bands 4 and 7, respectively. The spectral characteristic of sand suggests that strong sand and dust storm signals can be obtained using the difference between band 7 (2.13-µm) signals, which are high, and band 4 (0.555-µm) signals, which are relatively much lower.
In the current study, we have applied GDDI as given in equation (1)  GDDI based sand and dust storm event results are validated using MODIS combined DT-DB AOD product, Meteosat data, AERONET AOD data, and GAMEP data from meteorological stations. In the satellite-based sand and dust storm monitoring the underneath surfaces are divided into cold (mix signature including vegetation, water, etc.) and warm (large desert areas) surfaces. It is well documented that for the production of AOD spatial distribution maps DT and DB methods have been developed for cold and warm surfaces respectively. Considering the climatology and geography of the study region, the Kingdom of Saudi Arabia has been considered as a warm surface (Butt and Mashat 2018). However, it must be noted that sand and dust storm events entering in Kingdom of Saudi Arabia from the northeast and southwest are essentially crossing Arabian Gulf and Red sea respectively. Since the water area is considered as a cold surface while the Kingdom of Saudi Arabia is fundamentally a warm surface therefore we have used MODIS combined DT-DB based AOD product in order to validate GDDI based sand and dust storm events. Thus, monthly average as well as 5 days (± 2 days from the day of sand and dust storm event), MODIS combined DT-DB AOD products for the month (table 1) of sand and dust storm events are retrieved from Giovani. MODIS DT-DB AOD data in NetCDF format is processed using Grads software and monthly as well as 5 days maps are prepared. Similarly, Meteosat satellite data both from MFG and MSG (table 1) for 14 sand and dust storm events are analyzed in order to validate GDDI results. It is pertinent to mention here that data from MFG is a visible channel image in black and white from MVIRI (Meteosat Visible and Infrared Imager) instrument directly downloaded from https://www.eumetsat.int/case-studies in jpg format. On the other hand, MSG data from SEVIRI (Spinning Enhanced Visible and Infrared Imager) instrument is used to prepare an RGB color composite of dust products. MSG data downloaded from https://eoportal.eumetsat.int in NetCDF format is processed by using Python script. Once the NetCDF data is processed dust product is prepared by using band 10-band 9 in red, band 9-band 7 in green, and band 9 in blue colors. In the current study, we have also used AERONET station data in order to validate GDDI based sand and dust storm results. Many studies have been conducted in the past using AOD data measured by AERONET station (Sabbah and Hasan, 2008;Butt and Mashat, 2018). Therefore, we retrieved AERONET level 2.0 AOD data for 5 days (the fourth day is the sand and dust storm day) for 500 nm. Sand and dust storm events for which AERONET data were analyzed in order to validate GDDI results are given in Table 1. The GDDI based sand and dust storm events along with GAMEP ground station location are also analyzed using visual interpretation of location map ( gure 1). This will enable us to examine the authenticity of ground station data in the study region. In the current study, the performance of the GDDI algorithm is validated by using conventional statistical approaches, that is, accuracy, POCD, and POFD as given in equations 2, 3, and 4 respectively. A similar, approach has been used by Ciren and Kondragunta, (2014)

Results And Discussion
This study examines the application of GDDI using MODIS data for sand and dust storm monitoring in the Kingdom of Saudi Arabia from the year 2000 to 2017. A total of 14 sand and dust storm cases, shown in Fig. 2, are analyzed during the study period. Based on the spectral analysis of 14 images (Fig. 2) of sand and dust storm events we have proposed threshold values to ag different features in the study area. Since re ectance of water features in MODIS band 4 is higher than that in band 7 therefore negative GDDI values have been used to discriminate clouds/water from other features in the image. In the current study, we have classi ed the resultant GDDI image into 8 classes (shown in Figs. 5a, 6a,  For the validation of GDDI based sand and dust storm events, MFG and MSG images for the day of sand and dust storm events as shown in Fig. 3 are analyzed. Similarly, monthly (for the month of sand and dust storm event) and ve days (± 2 days from the day of sand and dust event) average MODIS based combined DT-DB AOD product as shown in Figs. 4a and 4b respectively are also analyzed in order to validate the results. However, during the study period, AERONET station data was available only for three sand and dust storm events. Thus, AOD data for two events from AERONET station at Solar Village and a single event at KAUST are analyzed. It is beyond the scope of this study to discuss in detail all 14 sand and dust storm events. Therefore, we have selected four sand and dust storm events (2003, 2011, 2012, and 2015) for detailed discussion that includes three events reported by AERONET station and one event reported by the highest number (11) Fig. 5a (GDDI results) that an intense sand and dust storm event (light brown color) is entering from the north and east sides of the study area. Our GDDI results are in accordance with GAMEP data, which indicates that ground stations located in the north (Guraiat, Hfr-Batin, Jouf, Qaisumah, Arar, Rafha, and Turaif), east (Ahsa, Dahran, Dammam), and south (Sharorah) regions of the study area have reported sand and dust storm event. Our analysis of Fig. 5a also indicates cloud cover (light aquamarine color) and cloud mixed with mild and moderate sand and dust (aquamarine mixed with light peach color and aquamarine mixed with peach color) over northern and eastern regions of the country. In addition, low to high concentrations of aerosols (light peach and peach color) can be seen (Fig. 5a) over northern, central, western, and southern regions of the study area. Once again, GDDI results are supported by GAMEP data, which shows that 13 stations (Gassim and Hail in the north, Riyadh in the center, Tabuk Fig. 4a). In addition, a high concentration of aerosols is also visible over a few parts of southern regions in the study area.
For the 30 April 2011 event (reported by 5 GAMEP stations), GDDI result, MSG image, ve days DT-DB AOD product, and AERONET station AOD product are shown in Figs. 6a, 6b, 6c, and 6d respectively. An intense sand and dust storm event (light brown color) along with cloud cover (light aquamarine color) can be seen in the eastern, central, and northern parts of the study area from the GDDI result (Fig. 6a).
The GDDI results for the 30 April 2011 event are in agreement with GAMEP data, which indicates that ground stations located in the east (Dammam), north (Qaisumah and Rafha), center (Riyadh), and west (Yanbu) of the study area have reported sand and dust storm event. Our analysis of Fig. 6a also highlights cloud cover (light aquamarine color) and clouds mixed with mild and moderate sand and dust (aquamarine mixed with light peach color and aquamarine mixed with peach color) in the northern and central regions of the study area. In addition, a low to a high concentration of aerosols (light peach and peach color) can be seen (Fig. 6a) over the northern, central, western, and southern regions of the study area. The GDDI results are again supported by GAMEP data, which shows that 9 stations (Gassim, Hail, Turaif, Jouf and Hfr Batin in the north, Riyadh in the center, Madinah and Jeddah in the west, and Najran in the south) have reported widespread of dust in the atmosphere on 30 April 2011. Once again GDDI based threshold values assigned to various features in the monitoring of sand and dust storm event are validated in the current study. Similarly, we have also validated GDDI results by MSG image (Fig. 6b) for 30 April 2011, which clearly shows clouds (dark brown color) as well as sand and dust storm event (pink color) in the eastern, central, and northern sides of the country. The analysis of ve days of DT-DB AOD product, as shown in Fig. 6c also indicates a very high concentration of aerosols (orange, red, and pink color) in the eastern, central, and northern regions of the study area as compared to the monthly average DT-DG AOD product (April 2011 case in Fig. 4a). In addition, a high concentration of aerosols is also visible over some western regions of the study area. Finally, 5 days AERONET station data retrieved from Solar Village in Riyadh (Fig. 6d) indicates a gradual increase in the concentration of aerosols from day 1 which attain maximum value on day 4 (the day of the event) and started to decrease on day 5.
For the 18 March 2012 event (reported by 5 GAMEP stations), GDDI result, MSG image, ve days DT-DB AOD product, and AERONET station AOD product are shown in Figs. 7a, 7b, 7c, and 7d respectively. It is evident from Fig. 7a (GDDI result) that an intense sand and dust storm event (light brown color) is entering from Arabian Gulf in the eastern part of the study area. Further analysis of Fig. 7a indicates sand and dust storm over the western part of the country as well. The GDDI results are once again in accordance with GAMEP data, which indicates that ground stations located in the eastern region (Ahsa, Dahran, Dammam) and western region (Jeddah and Yanbu) of the study area have reported sand and dust storm event. Our analysis of Fig. 7a also highlights a low to a high concentration of aerosols (light peach and peach color) over the northern, central, western, and southern regions of the study area. Once again, GDDI results are supported by GAMEP data, which shows that 20 stations (Arar, Gassim, Hail, Turaif, Tabuk, Qaisumah, Rafha, and Hfr Fig. 4a). In addition, moderate aerosols quantity is also visible over some northern parts of the study area. The widespread of aerosols covering the entire study area is also supported by the GAMEP data as 20 ground stations have reported moderate to a high concentration of aerosols. Similarly, 5 days AERONET station data retrieved from KAUST (Fig. 7d) indicates a gradual increase (from day 1) which turned into a large amount of aerosol concentration on day 4 (the day of the event) and ultimately start decreasing on day 5.
Unfortunately, a complete MODIS scene of the study area for the said date was unavailable however, moderate sand and dust storm traces (light brown mixed with peach color) can be seen over the eastern, central, and northeast parts (top right hand corner of Fig. 8a) of the study area. Our GDDI results are again in agreement with GAMEP data, which indicates that ground stations are located in the east (Dammam and Ahsa), north (Qaisumah and Rafha), and center (Riyadh) of the study area have reported sand and dust storm event. Our analysis of Fig. 8a also highlights a low to a high concentration of aerosols (light peach and peach color) over the western, northern, and southern regions of the study area.
Once again GDDI results are supported by GAMEP stations, which shows that 10 stations (Gassim, Hail, Arar, Tabuk, and Jouf in the north, Makkah, Wejh, and Yanbu in the west, and Wadi Dawaseer and Gizan in the south) have reported widespread of dust in the atmosphere. These are encouraging results that validated proposed GDDI based threshold values (label 3, 4, 5, 6, and 7) for the classi cation of various sand and dust storm-related features in the study area. The GDDI results are also validated by MSG image (Fig. 8b) for 01 April 2015, which clearly shows sand and dust storm (pink and dark pink color) events in the northern, eastern, and central regions of the country. The analysis of ve days DT-DB AOD product, as shown in Fig. 8c, indicates a high concentration (orange, red, and pink color) of aerosols in the eastern, central, and partly northern regions of the study area as compared to the monthly average DT-DB AOD product (April 2015 case in Fig. 4a). Finally, 5 days AERONET station data retrieved from Solar Village in Riyadh (Fig. 8d) indicates a gradual increase in the concentration of aerosols from day 1 which attain maximum value on day 4 (the day of the event) and started to decrease on day 5.

Conclusion
Every year, early warnings about natural calamity help get millions of people out of harm way before the danger comes. It is now clear that scienti c studies can protect us from these hazards by using data from different satellites and air quality information. More than 15 organizations, supported by WMO, currently provide daily sand and dust storm forecasts in different geographic regions of the world thus, integrating research and user communities (for example, medical, aeronautical, agricultural users, etc.). Their mission is to achieve comprehensive, coordinated, and sustained observations and modeling capabilities of sand and dust storm monitoring in order to increase the understanding and improve the monitoring and prediction capabilities.
The focus of current research is to evaluate the suitability of GDDI for sand and dust storm monitoring over the Kingdom of Saudi Arabia. Samadi et al. (2014) and Zandkarimi et al. (2020) concluded in different studies over Iran that no threshold values are required for the identi cation of sand and dust storm event by using the GDDI method. However, in current research, we have successfully identi ed threshold values for the identi cation of different features (for example, cloud cover, sand and dust mixed with clouds, low sand and dust in the atmosphere, moderate sand and dust storm, and heavy sand and dust storm) during the study period over Kingdom of Saudi Arabia. The GDDI threshold values assigned for the identi cation of different features is successfully validated by using GAMEP data. This study concluded that generally, GDDI can successfully identify sand and dust storm with a thunderstorm over the surface as well as water features in the study area. This study also concluded that thin clouds and outspread of thin dust can also be identi ed with the help of threshold values used in the application of GDDI.
It is pertinent that one day these outcomes could help millions of people in different parts of the world. Research forecasting products from atmospheric dust models may substantially contribute to risk reduction in many areas of societal bene t but will rely on real-time delivery of products to people in the Kingdom of Saudi Arabia with high precision.  Tables   Table 1. Description of GAMEP station data, MODIS satellite data, MODIS DT-DB AOD product, Meteosat satellite data, and AERONET AOD data used in the current study.

S. No
Sand and dust storm events reported by ground stations