The efficiency of remote sensing and machine learning algorithms in the zoning of susceptible areas to dust in Isfahan province

Document Type : Original Article

Authors

1 PhD candidate, Desert control and management, Department of Natural Resources and Earth Science, University of Kashan, Kashan, Iran.

2 Associate professor, Department of Natural Resources and Earth Science, University of Kashan, Kashan, Iran.

Abstract

Introduction

Dust is a phenomenon that mainly occurs in arid and semi-arid areas as a result of high wind speed and turbulence on the surface of the soil without vegetation and prone to erosion. Various factors such as wind speed, vegetation cover, soil characteristics, climatic factors, etc. are involved in the creation of wind erosion and the resulting dust, all of which are related to each other and lead to an increase or decrease in wind erosion and dust storms. The problems caused by dust storms are due to the lack of sufficient information about the prevailing conditions in the region, the way these conditions change, and the lack of knowledge of sensitive and prone areas to dust storms. To deal with this phenomenon and provide appropriate management solutions, it is necessary to know the areas prone to dust and the effective factors in the occurrence of this phenomenon. In this regard, remote sensing and modeling can be very effective in investigating the dust phenomenon. Numerous studies have also been conducted to investigate dust storms and dust sources and to model areas sensitive to this phenomenon using remote sensing data and machine learning.

Isfahan province is considered one of the most important geographical regions of the country, which is susceptible to successive drought, desertification, and dust storms due to its special geographical location, low rainfall, and proximity to the desert. So, it is necessary to carry out studies that will lead us to a correct understanding of dust-prone areas in this province. Therefore, In the current research, zoning of dust-prone areas in Isfahan province was done using meteorological codes related to dust, Aerosol optical depth values of MODIS sensor of Terra satellite (2001-2022), and machine learning algorithms including RF, BRT, SVM, and CART.



Materials and methods

Study area

Isfahan Province with an area of nearly 107017 km2 (6.4% of Iran area) is located between 30° 43′ to 34° 30′ N and 49° 38′ to 55° 31′ E in central Iran (Fig. 1). The mean annual precipitation of this province is between 40 mm and more than 800 mm and its mean annual temperature varies from 10 °C to 20 °C (Iran Meteorological Organization). According to the Torrent White method, the climate of Isfahan province is dry in 58.73% of its area (eastern, northeastern, and sub-central parts of the province), semi-arid in 28% of its area (central and northern parts of the province), and humid and semi-humid in 13.27% of its area (western and southern parts of the province).



Methodology

First, using AOD values, the occurrence and non-occurrence points of dust were determined. Ten various factors including land use, temperature, rainfall, erosive wind speed, slope, altitude, albedo, EVI, NDSI, and NDMI were determined as predictive factors. n the next step, the correlation between the predictive factors was calculated using the variance inflation factor (VIF). Using machine learning algorithms, spatial modeling of susceptible areas to dust was done and the importance of predictive factors in zoning was determined using the jackknife test. Finally, using the value of the area under the ROC curve (ROC-AUC), the validation of the model was done.



Results

The zoning map of dust-prone areas in Isfahan province showed that the low-altitude and flat parts of the north, parts of the northeast, southeast of the province, and the central areas towards the southwest and west of Isfahan province are vulnerable areas against the occurrence of dust.

The highest percentage of areas prone to dust in the RF model was in the very low class with a value of 21.36%, in the BRT model was in the medium class with a value of 22.66%, and in the SVM and CART models, it was in very high and low classes with values of 23.92% and 37.6%, respectively.

The results of validation illustrated that the RF model with AUC = 0.86 was the most efficient, followed by BRT, CART, and SVM models with AUC values of 0.82, 0.79, and 0.77 respectively. According to the results of the Jackknife test, in RF, BRT, and CART models, rainfall had the most effect in modeling while in the SVM model, temperature and then rainfall had the most effect in the modeling.



Discussion and conclusion

Based on the results, the most vulnerable areas to dust are assigned to salt land, barren lands, and rangeland with poor quality. These areas, which are mainly located in the northern, central, and parts of the eastern sides of the province, have the lowest amount of surface soil moisture, the lowest amount of rainfall, and the highest temperature, and as a result, they face the lack of vegetation or weak vegetation. Therefore, these areas are exposed to the occurrence of dust, and with winds blowing at a speed exceeding the threshold speed of wind erosion, dust storms will happen.

According to the validation results, the RF model has the best performance among the applied models, followed by the BRT, CART, and SVM models. Random forest algorithm is one of the advanced decision tree models used for classification and regression. This algorithm has a much more accurate performance compared to other simple regression trees or parametric statistical methods and is defined based on a large number of decision trees.

The results of the jackknife test introduced rainfall as the most important factor in RF, BRT, and CART models. In the RF model, as the best model, after the rainfall factor, the temperature and altitude factors are more important than other factors. Considering the low amount of rainfall in dust-prone areas, it can be said that low rainfall, soil dryness, and, as a result, the reduction of vegetation, increase the conditions for creating wind erosion and dust.

Keywords

Main Subjects


  1. Abolhasani, A., Zehtabian, G., Khosravi, H., Rahmati, O., Alamdarloo, E.H., & D'Odorico, P. (2022). A new conceptual framework for spatial predictive modelling of land degradation in a semiarid area. Land Degradation & Development, 33(17), 3358-3374. DOI: org/10.1002/ldr.4391
  2. Akhzari, D. & Haghighi, S. (2015). Effect of vegetation change of source area on dust storms occurrence in the west of Iran. Ecopersia, 3(4), 1133-1143. DOI: 1001.1.23222700.2015.3.4.3.3
  3. Alipour, N., Mesbahzadeh, T., Ahmadi, H., Malekian, A., & Jafari, M. (2018). Synoptic analysis of dust events and its relation with drought in Alborz and Qazvin provinces, Quarterly of Geography & Regional Planning, 8(30), 59-68. DOI: 1001.1.22286462.1397.8.2.4.3 [In Persian]
  4. Boali, A., Jafari, R., & Bashari, H. (2017). Wind erosion estimation and assessment using Bayesian belief networks in eastern Isfahan township. Desert Ecosystem Engineering, 6(14), 45-58. DOI: 22052/6.14.45 [In Persian]
  5. Boroghani, M., Moradi, H.R., Zangane Asadi, M.A., & Pourhashemi, S. (2019). Evaluation of the role of drought in frequency of dust in Khorasan Razavi province. Journal of Environment Science & technology, 21(5), 109-121. DOI: 22034/JEST.2019.10464 [In Persian]
  6. Boroughani, M., Pourhashemi, S., Hashemi, H., Salehi, M., Amirahmadi, A., Zangane Asadi, M.A., & Berndtsson, R. (2020). Application of remote sensing techniques and machine learning algorithms in dust source detection and dust source susceptibility mapping. Ecological Informatics, 56, 101059. DOI: org/10.1016/j.ecoinf.2020.101059
  7. Choubin, B., Abdolshahnejad, M., Moradi, E., Querol, X., Mosavi, A., Shamshirband, S. & Ghamisi, P. (2020). Spatial hazard assessment of the PM10 using machine learning models in Barcelona, Spain. Science of The Total Environment, 701, 134474. DOI: org/10.1016/j.scitotenv.2019.134474
  8. Darvand, S., Khosravi, H., Keshtkar, H.R., Zehtabian, Gh.R., & Rahmati, O. (2021). Comparison of machine learning models to prioritize susceptible areas to dust production. Range and Watershed Management, 74(1), 53-68. DOI: org/10.22059/jrwm.2021.321033.1580 [In Persian]
  9. Gareth, J., Daniela, W., Trevor, H., & Robert, T. (2013). An introduction to statistical learning: with applications in R. Spinger.
  10. Grinand, C., Vieilledent, G., Razafimbelo, T., Rakotoarijaona, J. R., Nourtier, M., & Bernoux, M. (2019). Landscape‐scale spatial modelling of deforestation, land degradation and regeneration using machine learning Land Degradation & Development, 31(13), 1699-1712. DOI: 10.1002/ldr.3526
  11. Ghomeshion, M., Vali, A.A., Ranjbar Fordoei, A., & Mousavi, H. (2022). Investigating the effect of land cover on dust spatial distribution in Southern Khuzestan province. ECOPERSIA, 10(3), 179-189. DOI: 1001.1.23222700.2022.10.3.2.9
  12. Jafari, M., Zehtabian, Gh.R., Ahmadi, H., Mesbahzadeh, T., & Norouzi, A.A. (2020). Detecting and routing of dust event using remote sensing and numerical modeling in Isfahan Province. Environmental Science, 18(1), 105-116. DOI: 29252/envs.18.1.105 [In Persian]
  13. Lin, X., Chang, H., Wang, K., Zhang, G., & Meng, G. (2020). Machine learning for source identification of dust on the Chinese Loess Plateau. Geophysical Research Letters, 47(21), 2020GL088950. DOI: 1029/2020GL088950
  14. Naeimi, M., Yousefi, M.J., Khosroshahi, M., Zandifar, S., & Ebrahimikhusfi, Z. (2020). Climatic factors affecting dune mobility in the west of Khorasan Razavi Province, Iran. Geographical Research on Desert Areas, 7(2), 25-45. [In Persian]
  15. Naghibi, S.A., Pourghasemi, H.R., & Dixon, B. (2016). GIS-based groundwater potential mapping using boosted regression tree, classification and regression tree, and random forest machine learning models in Iran. Environmental Monitoring and Assessment, 188(1), 1-27. DOI: 1007/s10661-015-5049-6
  16. Namdari, S., Karimi, N., Sorooshian, A., Mohammadi, GH., & Sehatkashani, (2018). Impacts of climate and synoptic fluctuations on dust storm activityover the Middle East. Atmospheric Environment, 173: 265-276. DOI: doi.org/10.1016/j.atmosenv.2017.11.016
  17. Pourhashemi, S.m., Amirahmadi, A., Zangane Asadi, M.A., & Salehi, M. (2019). Determination of Geomorphological and Land Use Features of Dust Harvesting Sources (Case Study: Khorasan Razavi Provience). Arid Region Geographic Studies, 9(34), 14-24. [In Persian]
  18. Rahmati, O., Panahi, M., Ghiasi, S.S., Deo, R.C., Tiefenbacher, J.P., Pradhan, B., Jahani, A., Goshtasb, H., Kornejady, A., Shahabi, H., & Shirzadi, A. (2020). Hybridized neural fuzzy ensembles for dust source modeling and prediction. Atmospheric Environment, 224, 117320. DOI: https://doi.org/10.1016/j.atmosenv.2020.117320
  19. Rezaei, M., & JavanNezhad, R. (2020). Modeling the Role of Climate in Distribution of two-spotted spider mite: Case study of Tehran province. Environmental Science Studies, 5(2), 2554-2559. [In Persian]
  20. Robinson, S. )2014(. Simulation: the practice of model development and use. 2nd edition. Palgrave Macmillan.
  21. Samadi, M., Jabbari, E., & Azamathulla, H.M. )2014(. Assessment of M5′ model tree and classification and regression trees for prediction of scour depth below free overfall spillways. Neural Computing and Applications, 24(2), 357-366. DOI: 1007/s00521-012-1230-9
  22. Sohrabi, T., Vali, A.A., Ranjbar Fordoei, A., & Mousavi, S.H. (2019). Quantitative analysis of vegetation feedback on the occurrence of dust in arid ecosystems (case study: Isfahan province). Range and Watershed Management, 71(4), 973-985. DOI: org/10.22059/jrwm.2019.273257.1337 [In Persian]
  23. Yesilnacar, E.K. )2005(. The application of computational intelligence to landslide susceptibility mapping in Turkey. The University of Melbourne, Department, 200.
  24. Zaker Esfahani, A.R. (2012). Combating with desertification process by an emphasis on capabilities of desert areas (case study: Isfahan province). Environmental Studies, 38(3), 155-164. DOI: 22059/jes.2012.29157 [In Persian]