Identifying Origins of Atmospheric Aerosols using Remote Sensing and Data Mining (Case study: Yazd province)

Subject Areas : Natural resources and environmental management

Mohamad Kazemi ¹ , Ali Reza Nafarzadegan ² , Fariborz Mohammadi ³ , Ali Rezaei Latifi ⁴

1 - Assistant Professor, Hormoz Studies and Research Center, University of Hormozgan, Bandar Abbas, Iran
2 - Assistant Professor, Department of Natural Resources Engineering, Faculty of Agriculture and Natural Resources, University of Hormozgan, Bandar Abbas, Iran
3 - Assistant Professor, Hormoz Studies and Research Center, University of Hormozgan, Bandar Abbas ; Assistant Professor, Department of Water Sciences & Engineering, Minab Higher Education Complex, University of Hormozgan, Minab, Iran
4 - Assistant Professor, Hormoz Studies and Research Center, University of Hormozgan, Bandar Abbas ; Assistant Professor, Physics Department, Faculty of Sciences, University of Hormozgan, Bandar Abbas, Iran

Received: 2020-01-13 Accepted : 2020-09-19 Published : 2021-03-21

Keywords: zoning, Machine Learning, Aerosol optical depth, Spatial variables,

Abstract :

Background and ObjectiveThe Middle East is one of the most important regions in the world for dust production. Iran, located in the Middle East, is exposed to numerous local and trans-regional dust systems due to its location in the arid and semi-arid regions of the world. Dust storms, in addition to covering arable land and plants with wind-blown materials, destroy fertile lands and reduce biological production and biodiversity, and severely affect the survival of residents. Dust storms are involved in the transmission of dangerous pathogens to humans, air pollution, and damage to respiratory function. Dust storms in Yazd province are relatively common and the average number of days with dust storms in the province reaches 43 days a year. This phenomenon has caused many problems for the people of the province. The main indicators of air quality are the concentration of suspended particles and the aerosol optical depth (AOD) following the occurrence of dust events. Numerous studies have been conducted in the world to identify the centers of dust collection and their origin. However, to the best of the authors’ knowledge, there is no study on the spatial zoning of dust conditions using three algorithms of CART, MARS and TreeNet algorithms as the predictive models. The purpose of this study is to forecast and zoning the potential of different areas for the production of dust aerosols using remote sensing data and data mining models as well as to specify the most important variables on this phenomenon in Yazd province. Materials and Methods The Yazd province lies in a dry region of Central Iran. The province experienced average annual rainfall of about 57 mm and an average annual temperature of about 20 ºC. The maximum temperature experienced in the warmest month of the province is close to 46 ºC. The maximum wind speed in this province is up to 120 kilometres per hour. The Google Earth Engine (GEE) interface (Javascript editor) was applied to collect remote sensing data in order to form three data sets that contain features related to topography, climate, and land surface conditions. These features were employed as the independent variables of the models, which is built by taking advantage of three data mining algorithms, classification and regression tree (CART), multivariate adaptive regression splines (MARS), and TreeNet, to specify the potential of areas for dust production. The dependent variable (target variable) of the models was the aerosol optical depth (AOD), which was acquired from MOD04 AOD retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard NASA's Terra satellite. The outcomes of the three models for classifying areas with different dust potentials were evaluated under performance criteria, such as R-squared, mean absolute deviation (MAD), the mean square error (MSE), the mean relative absolute deviation (MRAD), and the root means square error (RMSE). Results and Discussion The results showed the variables mostly affecting the dependent variable (AOD) in the MARS model were actual evapotranspiration, soil moisture, and the Palmer drought severity index. The values of R2 and RMSE in the MARS model were equal to 0.72 and 0.02, respectively. Similarly, the features with the highest relative importance according to the TreeNet model were soil moisture, Palmer drought severity index, and actual evapotranspiration. The values of R2 and RMSE in the TreeNet model were equal to 0.75 and 0.019, respectively. The results revealed that the CART model with R2 =0.85, MAD = 0.011, MSE =0.002, MRAD =0.262, and RMSE =0.014 had the best performance compared with the other two data mining models. The soil moisture, elevation, reference and actual evapotranspiration, minimum and maximum temperature, Palmer drought severity index, downward shortwave solar radiation, and wind speed were the most important variables in forecasting the potential of areas for dust production, respectively. Also, the areas with very high, high, moderate, low and very low susceptibility were occupied about 16%, 19%, 26%, 20% and 20% of the Yazd province, respectively. Conclusion All three models, which were based on three data mining algorithms, CART, MARS, and TreeNet, had a good agreement in specifying the most important variables affecting the optical depth of the dust aerosols in the study area. However, these models indicated different priority order for the identified variables in terms of relative importance; Besides, there was a difference in their performance criteria. As mentioned above, the CART model was the best-performing model, of the current study, for specifying the potential of areas for the generation of dust aerosols. According to this model, 25.8% of the province was classified as the moderate-risk of aerosol production, 18.6% of the province as the high-risk of aerosol production, and 16.0% of the study region as the very high-risk of dust aerosols. The high-risk areas are mostly spread in the western and southwestern regions of the Yazd province. Palmer United States golfer (born in 1929) More (Definitions, Synonyms, Translation). http://dorl.net/dor/20.1001.1.26767082.1400.12.1.4.5

References:

Ahmadlou M, Delavar M. 2015. Multiple land use change modeling using multivariate adaptive regression spline and geospatial information system. Journal of Geomatics Science and Technology, 5(2): 131-146. (In Persian).

Ali M, Asklany SA, El-wahab M, Hassan M. 2019. Data Mining Algorithms for Weather Forecast Phenomena Comparative Study. International Journal of Computer Science and Network Security, 19(9): 76-81.

Alibakhshi T, Azizi Z, Vafaeinezhad A, Aghamohammadi H. 2020. Survey of Area Changes in Water Basins of Shahid Abbaspour Dam Caused by 2019 Floods Using Google Earth Engine. Iranian Journal of Ecohydrology, 7(2): 345-357. (In Persian).

Bari Abarghuei H, Tabatabaei Aghda S, Tavakoli M, Najjar Hadashi N. 2006. The origin of Yazd storms and the damages caused by it. 1st National Conference on Wind erosion and dust storms. Paper presented at the 21 January, Yazd University, Yazd, Iran. (In Persion).

Boroughani M, Pourhashemi S. 2019. Susceptibility Zoning of Dust Source Areas by Data Mining Methods over Khorasan Razavi Province. Environmental Erosion Research Journal, 9(3): 1-22. (In Persian).

Danesh Shahraki M, Shahriari A, Gangali M, Bameri A. 2017. Seasonal and Spatial Variability of Airborne Dust Loading Rate over the Sistan plain cities and its Relationship with some Climatic Parameters. Journal of Water and Soil Conservation, 23(6): 199-215. (In Persian).

Ebrahimi-Khusfi Z, Ruhollah T-M, Maryam M. 2021. Evaluation of machine learning models for predicting the temporal variations of dust storm index in arid regions of Iran. Atmospheric Pollution Research, 12(1): 134-147. doi:https://doi.org/10.1016/j.apr.2020.08.029.

Friedman JH, Meulman JJ. 2003. Multiple additive regression trees with application in epidemiology. Statistics in Medicine, 22(9): 1365-1381. doi:https://doi.org/10.1002/sim.1501.

Fridedman J. 1991. Multivariate adaptive regression splines (with discussion). Ann Stat, 19(1): 79-141.

Gholami H, Aliakbar M, Adrian LC. 2020. Spatial mapping of the provenance of storm dust: Application of data mining and ensemble modelling. Atmospheric Research, 233: 104716. doi:https://doi.org/10.1016/j.atmosres.2019.104716.

Gordon L. 2013. Using classification and regression trees (CART) in SAS^® enterprise miner TM for applications in public health. SAS Global Forum 2013, San Francisco, California.

Halabian A, Javari M, Akbari Z, Akbari G. 2017. Evaluating the performance of decision tree model in estimating the suspended sediments of river (A case study on the basin of Meimeh river). Geography And Development Iranian Journal, 15(49): 81-96. (In Persian).

Hojati M. 2017. Artificial neural network based model to estimate dust storms PM10 content using MODIS satellite images. Journal of Environmental Studies, 42(4): 823-838. (In Persian).

Hunter H, Cervone G. 2017. Analysing the influence of African dust storms on the prevalence of coral disease in the Caribbean Sea using remote sensing and association rule data mining. International Journal of Remote Sensing, 38(6): 1494-1521. doi:https://doi.org/10.1080/01431161.2016.1277279.

Karimi K, Taheri Shahraiyni H, Habibi Nokhandan M, Hafezi Moghadas N. 2011. Identifying sources of origin for producing dust storms in Middle East using remote sensing. Journal of Climate Research, 2((7-8)): 57-72. (In Persian).

Khalighi Sigaroudi S, Shahbandari R, Dadfar R, Kamrani F. 2011. Investigation of the relationship between drought and dust storms (Case study: Yazd province). Paper presented at the 2nd National Conference on Wind Erosion and Dust Storms. Yazd University, Yazd, Iran. (In Persian).

Loh WY. 2011. Classification and regression trees. Wiley interdisciplinary reviews: data mining and knowledge discovery, 1(1): 14-23.

Mirakbari M, Ganji A, Fallah S. 2010. Regional bivariate frequency analysis of meteorological droughts. Journal of Hydrologic Engineering, 15(12): 985-1000. doi:https://doi.org/10.1061/(ASCE)HE.1943-5584.0000271.

Mohammad Khan S. 2017. The study of the status and trend of changes in dust storms in Iran during the period from 1985 to 2005. Irrigation and Watershed Management (Iranian Journal of Natural Resources) 2(3): 495-514. (In Persian).

Panahi M, Mirhashemi SH. 2015. Assessment among two data mining algorithms CART and CHAID in forecast air temperature of the synoptic station of Arak. Environmental Sciences, 13(4): 53-58. (In Persian).

Pourhashemi S, Amirahmadi A, Zangane Asadi MA, Salehi M. 2018. Identifying and determining the characteristics of dust centers in Khorasan Razavi province. Arid Regions Geography Studies, 9(34): 1-9. (In Persian).

Pourhashemi S, Boroghani M, Amirahmadi A, Zanganeh Asadi M, Salhi M. 2019. Dust source prioritization with using statistical models (Case study: Khorasan Razavi provience). Journal of Range and Watershed Managment, 72(2): 343-358. (In Persian).

Rashki A, Kaskaoutis D, Rautenbach CJW, Eriksson P, Qiang M, Gupta P. 2012. Dust storms and their horizontal dust loading in the Sistan region, Iran. Aeolian Research, 5(3): 51-62. doi:https://doi.org/10.1016/j.aeolia.2011.12.001.

Rezazadeh M, Irannejad P, Shao Y. 2013. Climatology of the Middle East dust events. Aeolian Research, 10: 103-109. doi:https://doi.org/10.1016/j.aeolia.2013.04.001.

Rokach L, Maimon OZ. 2014. Data mining with decision trees: theory and applications, vol 81. World scientific. 244 p.

Sharma H, Kumar S. 2016. A survey on decision tree algorithms of classification in data mining. International Journal of Science and Research (IJSR), 5(4): 2094-2097.

Sobhani B, Safarian Zengir V, Faizollahzadeh S. 2020. Modeling and prediction of dust in western Iran. Physical Geography Research Quarterly, 52(1): 17-35. (In Persian).

Soleimanpour S, Mesbah S, Hedayati B. 2018. Application of CART decision tree data mining to determine the most effective drinking water quality factors (case study: Kazeroon plain, Fars province). Iranian Journal of Health and Environment, 11(1): 1-14. (In Persian).

Tsolmon R, Ochirkhuyag L, Sternberg T. 2008. Monitoring the source of trans-national dust storms in north east Asia. International Journal of Digital Earth, 1(1): 119-129. doi:https://doi.org/10.1080/17538940701782593.

Zha W, Chan W-Y. 2005. Objective Speech Quality Measurement Using Statistical Data Mining. EURASIP Journal on Advances in Signal Processing, 2005(9): 721258. doi:10.1155/ASP.2005.1410.