Using the Boruta algorithm and deep learning models for mapping land susceptibility to atmospheric dust emissions in Iran

A - Papers appearing in refereed journals

Gholami, H., Mohammadifar, A., Golzari, S., Kaskaoutis, D.G. and Collins, A. L. 2021. Using the Boruta algorithm and deep learning models for mapping land susceptibility to atmospheric dust emissions in Iran. Aeolian Research. 50 (article), p. 100682. https://doi.org/10.1016/j.aeolia.2021.100682

AuthorsGholami, H., Mohammadifar, A., Golzari, S., Kaskaoutis, D.G. and Collins, A. L.
Abstract

Wind erosion have many negative effects on global terrestrial and aquatic ecosystems and these phenomena are controlled by several factors including climatic, meteorological, topographic, vegetation, surface and soil characteristics. This study applied, for the first time, the Boruta algorithm for identification of effective variables controlling wind erosion. The novelty of the study was increased further using application of two deep learning (DL) models comprising a simple recurrent neural network (RNN) and restricted boltzmann machine (RBM). Collectively, these tools were used to map land susceptibility to wind erosion in parts of Kerman province, southeastern Iran. Among 18 potential variables for controlling dust emissions via wind erosion, 4 and 14 were identified as non-important and important, respectively, by the Boruta algorithm, while three (precipitation, digital elevation model and soil organic carbon) were selected as the most important factors. An inventory map of the wind erosion confirmed using both a test dataset (30%) and a training dataset (70%) was used to construct predictive models of land susceptibility to wind erosion. Both DL predictive models exhibited highly satisfactory performance according to a Taylor diagram, but the simple RNN performed slightly better than RBM. Based on the simple RNN, 35.6%, 5%, 2.4%, 22.7% and 34.3% of the total study area were characterized by very low, low, moderate, high and very high susceptibility, respectively. Convergent prediction of the same susceptibility classes by intersecting the maps generated by both models classified 17.4%, 0.07%, 0.06%, 7.4% and 34% of the total study area as very low, low, moderate, high and very high susceptibility classes, respectively. We conclude that applying the Boruta algorithm and DL models as new methods in aeolian geomorphology, may provide accurate spatial maps of dust sources to help target mitigation of detrimental dust effects on climate, ecosystems and human health.

KeywordsWind erosion; Land susceptibility ; Convergent prediction ; Taylor diagram; Recurrent neural network ; Iran
Year of Publication2021
JournalAeolian Research
Journal citation50 (article), p. 100682
Digital Object Identifier (DOI)https://doi.org/10.1016/j.aeolia.2021.100682
Web address (URL)https://doi.org/10.1016/j.aeolia.2021.100682
Open accessPublished as non-open access
FunderBiotechnology and Biological Sciences Research Council
European Union
Funder project or codeS2N - Soil to Nutrition - Work package 3 (WP3) - Sustainable intensification - optimisation at multiple scales
PANhellenic infrastructure for Atmospheric Composition and climatE change” PANACEA (MIS 5021516)
Output statusPublished
Publication dates
Online18 Feb 2021
Publication process dates
Accepted25 Jan 2021
PublisherElsevier

Permalink - https://repository.rothamsted.ac.uk/item/9836v/using-the-boruta-algorithm-and-deep-learning-models-for-mapping-land-susceptibility-to-atmospheric-dust-emissions-in-iran

Restricted files

Publisher's version

Under embargo indefinitely

16 total views
0 total downloads
1 views this month
0 downloads this month