Integrated modelling for mapping spatial sources of dust in central Asia - An important dust source in the global atmospheric system

https://doi.org/10.1016/j.apr.2021.101173Get rights and content

Highlights

  • An new integrated modeling approach used to map dust source spatially.

  • Leave one feature out (LOFO) applied for the feature selection.

  • Game theory used to interpretability of the predictive models output.

  • The hybrid copula-gcForest model was identified as the most accurate.

  • Bulk density has the highest contribution to the predictive model's output.

Abstract

Spatial mapping of dust sources in arid and semi-arid regions is necessary to mitigate on-site and off-site impacts. In this study, we apply a novel integrated modelling approach including leave one feature out (LOFO) – as a technique for feature selection -, deep learning (DL) models (gcForest and bidirectional long short-term memory (Bi-LSTM)), game theory (GT) and a Gaussian copula-based multivariate (GCBM) model for mapping dust sources in Central Asia (CA). Eight factors (precipitation, cation exchange capacity, bulk density, wind speed, slope, silt content, lithology and coarse fragment content) were selected by LOFO as effective for controlling dust emissions, and were used in the novel modelling process. Six statistical indicators were utilized to assess the performance of the two DL models and a hybrid copula-gcForest model, while a sensitivity analysis of the models was also carried out. The hybrid copula-gcForest model was identified as the most accurate, predicting that 16%, 7.1%, 9.5% and 67.4% of the study area is grouped at low, moderate, high and very high susceptibility classes for dust emissions, respectively. Based on permutation feature importance measure (PFIM) and Shapely Additive exPlanations (SHAP), bulk density, precipitation and coarse fragment content were evaluated as the three most important factors with the highest contributions to the predictive model output. The study area suffers from intense wind erosion and the generated spatial maps of dust sources may be helpful for mitigating and controlling dust phenomena in CA.

Introduction

Dust storms are one of the most important consequences of wind erosion in arid and semi-arid regions (Lal, 2003; Xiao et al., 2017; Kouchami-Sardoo et al., 2019; Xu et al., 2019) - and have many on-site (e.g., loss of organic matter, degradation of water quality or topsoil structure, burial of plants, abrasion of crop tissue) and off-site (e.g., reduced visibility and traffic safety, deposition of dust in houses, or on agricultural or industrial land) impacts (Goossens, 2003; Miri et al., 2009). Dust storms generated by wind erosion affect terrestrial and marine ecosystems, as well as weather phenomena and climate dynamics (Jish Prakash et al., 2015; Tyagi et al., 2019; Suresh et al., 2021; Francis et al., 2021).

Frequent wind-blown dust events result from severe wind erosion in Central Asia (CA) (Pi et al., 2017; Shen et al., 2016; Wang et al., 2020; Zhang et al., 2020) and may contribute to 17–20% of global dust emissions (Xi and Sokolik, 2015). The highest rates of dust deposition globally have been simulated over CA (Groll et al., 2013; Zhang et al., 2017). The significance of CA dust emissions on local socio-economic performance, human health and ecosystems has been recognized in several previous studies (Indoitu et al., 2015; Issanova and Abuduwaili, 2017; Rupakheti et al., 2020; Li and Sokolik, 2018), which, thereby, motivate further research on the dust problem and associated atmospheric circulation patterns and dynamics (Groll et al., 2013; Shi et al., 2020; Li et al., 2019, 2021a). More specifically, determining land susceptibility classes to wind erosion and/or dust emissions and providing spatial maps for the provenance of dust are essential to help mitigate the deleterious effects of dust storms on the environment, socio-economic affairs and human health in CA (Li et al., 2020).

To address this issue, machine learning algorithms (e.g., random forest, support vector machine, cubist, classification and regression tree) (Gholami et al., 2021a; Boroughani et al., 2020) and deep learning (DL) techniques, including simple recurrent neural network (RNN) and restricted Boltzmann machine (RBM) (Gholami et al., 2021b), have been widely used in aeolian geomorphology for spatial modelling purposes. Many techniques such as copulas, variational autoencoders and generative adversarial networks have been used to generate synthetic data (Patki et al., 2016; Wan et al., 2017; Xu and Veeramachaneni, 2018). In comparison with black-box DL models, training copulas are easy to use and robust. Copula models provide a direct representation of the statistical distribution, which makes them easier to interpret and tweak after training. Overall, the copula-based models can be used effectively for generating synthetic data that are in close agreement with real data (Patki et al., 2016).

The susceptibility of land surfaces to wind erosion hazard is determined by a diverse range of environmental and bioclimatic variables (Parajuli et al., 2016, 2019; Jebali et al., 2021), and therefore, identifying the controlling factors through DL models is essential for a successful modelling exercise. A wide range of machine learning models including random forest, MARS, extreme gradient boosting, maximum entropy, Bayesian additive regression trees, multiple linear regression model and boosted regression trees has been used to identify effective factors for various environmental hazards such as gully erosion (Azareh et al., 2019), landslides (Chen et al., 2017), surface solar radiation (Wang et al., 2021; Yu et al., 2020) and dust emissions from arid land surfaces (Gholami et al., 2020b). Most recently, Mohammadifar et al. (2021a) and Gholami et al. (2021b) reported successful applications of game theory and the boruta algorithm to identify and quantify the relative importance of effective factors for soil erosion by water, and wind from arid land surfaces, respectively.

This study examines the factors that contribute to land degradation, wind erosion and dust emissions in CA and also develops new high-resolution spatial maps for dust sources over the region, by applying two individual novel DL models (i.e., the gcForest and the bidirectional LSTM –Bi-LSTM) and a hybrid copula-gcForest model based on a gcForest and Gaussian copula model. Evaluation of the performance of the individual DL models and a hybrid model is undertaken by applying six statistical indicators namely; accuracy, recall, precision, f1 score, cohens kappa and the receiver operating characteristic-area under curve (ROC-AUC). The analysis quantifies the relative importance of factors controlling dust emissions in CA, and their interactions and contributions to prediction, and finally, generates synthetic data with a copula Gaussian multivariate model. Calibration of the best model is undertaken using synthetic data. High resolution spatial mapping of the areas susceptible to wind erosion over CA may help in mitigating deleterious impacts of dust storms on the atmospheric environment, socio-economics and population health.

Section snippets

Study area description

CA, extending from the Caspian Sea to the East Tianshan Mountains, with an area of about 7.75 M km2 (Cheng et al., 2016), is an arid region with several deserts and thick and expansive loess deposits (Li et al., 2020, 2021a), located in the eastern end of the dust belt (Shi et al., 2021) (Fig. 1). CA includes several deserts such as the Karakum, Kyzylkum, Aralkum, Caspian depression, Balkhash arid land and Taklimakan desert in the Tarim Basin (Zhang et al., 2020; Shi et al., 2021). CA

Variables controlling dust emissions from the land surfaces

Many factors including climatic variables, soil characteristics and physiographic-topographic factors control dust emissions from land surfaces (Shao, 2008; Ge et al., 2016). Here, we examined 14 potential factors that affect land susceptibility to wind erosion and mineral dust emissions, comprising climatic factors (e.g., precipitation and wind speed), soil properties (bulk density, cation exchange capacity -CEC), silt, coarse fragment, sand, clay and organic carbon content, elevation (from a

Feature selection by LOFO importance

The results of feature selection by LOFO are presented in Fig. 5, and eight out of the 14 factors were considered effective for controlling dust emissions in CA, namely; precipitation, bulk density, slope, lithology, CEC, wind speed, silt content and coarse fragment content. On the contrary, six factors comprising sand content, land use, aspect, elevation, clay content and organic carbon content were removed from further analysis. Based on the genetic algorithm, wind speed, elevation, organic

Interaction among the features controlling dust emissions

The results of assessing the interaction of variables controlling dust emissions in CA suggested that bulk density had the strongest interaction strength with slope, coarse fragment content and precipitation, whereas CEC exhibited the strongest interaction with silt content, slope and wind speed. The strongest interaction strength for coarse fragment content was found to be with silt content, precipitation and bulk density, whereas as that for lithology was identified as being with silt

Conclusions

The novel scientific contribution of this study is the introduction of a new integrated modelling approach (LOFO – DL – GCBM – GT approach) for generating spatial maps of the dust sources and assessing the interpretability of the spatial maps over CA. Based on the statistical indicators of model performance, the hybrid copula-gcForest model performed slightly better than both individual DL models in generating spatial maps of dust sources in CA. According to the current modelling results,

Declaration of competing interest

The authors declare that there is no conflict of interests regarding the publication of this article.

Acknowledgements

The authors would like to thank the Faculty of Agriculture and Natural Resources, University of Hormozgan, Iran, for supporting this joint research project. The research was also supported by a collaborative agreement on “fingerprinting sources of aeolian dust and climate and environmental change in the core zone of the Silk Road economic belt based on loess deposits” between the University of Hormozgan and the Institute of Earth Environment, Chinese Academy of Sciences. D.G.K. acknowledges

References (91)

  • M. Groll et al.

    Spatial and temporal distribution of the dust deposition in Central Asia–results from a long term monitoring program

    Aeolian Research

    (2013)
  • X. Guan et al.

    Spatial distribution, temporal variation, and transport characteristics of atmospheric water vapor over Central Asia and the arid region of China

    Global Planet. Change

    (2019)
  • J. Huo et al.

    Thermodynamic and dynamic behavior in a recent dust precipitation event in north China

    Atmospheric Pollution Research

    (2021)
  • R. Indoitu et al.

    Dust storms in central asia: spatial and temporal variations

    J. Arid Environ.

    (2012)
  • K.U. Jaseena et al.

    Decomposition-based hybrid wind speed forecasting model using deep bidirectional LSTM networks

    Energy Convers. Manag.

    (2021)
  • S. Karami et al.

    Numerical simulations of dust storms originated from dried lakes in central and southwest Asia: the case of Aral Sea and Sistan Basin

    Aeolian Research

    (2021)
  • D.G. Kaskaoutis et al.

    Meteorological regimes modulating dust outbreaks in southwest Asia: the role of pressure anomaly and Inter-Tropical Convergence Zone on the 1–3 July 2014 case

    Aeolian Research

    (2015)
  • D.G. Kaskaoutis et al.

    The Caspian Sea–hindu kush index (CasHKI): a regulatory factor for dust activity over southwest asia

    Global Planet. Change

    (2016)
  • I. Kouchami-Sardoo et al.

    Application of a Bayesian belief network model for assessing the risk of wind erosion: a test with data from wind tunnel experiments

    Aeolian Research

    (2019)
  • R. Lal

    Soil erosion and the global carbon budget

    Environ. Int.

    (2003)
  • B. Laurent et al.

    Modeling mineral dust emissions from Chinese and Mongolian deserts

    Global Planet. Change

    (2006)
  • Y. Li et al.

    Atmospheric dust dynamics in southern Central Asia: implications for buildup of Tajikistan loess sediments

    Atmos. Res.

    (2019)
  • Y. Li et al.

    Source fingerprinting loess deposits in Central Asia using elemental geochemistry with Bayesian and GLUE models

    Catena

    (2020)
  • Y. Li et al.

    Aeolian dust dynamics in the Fergana Valley, Central Asia, since~ 30 ka inferred from loess deposits

    Geoscience Frontiers

    (2021)
  • B.A. Maher et al.

    Global connections between aeolian dust, climate and ocean biogeochemistry at the present day and at the last glacial maximum

    Earth Sci. Rev.

    (2010)
  • A. Mohammadifar et al.

    Assessment of the interpretability of data mining for the spatial modelling of water erosion using game theory

    Catena

    (2021)
  • A. Rashki et al.

    Assessment of dust activity and dust-plume pathways over Jazmurian Basin, southeast Iran

    Aeolian Research

    (2017)
  • A. Rashki et al.

    Statistical evaluation of the dust events at selected stations in southwest asia: from the Caspian Sea to the arabian sea

    Catena

    (2018)
  • D. Rupakheti et al.

    Columnar aerosol properties and radiative effects over Dushanbe, Tajikistan in Central Asia

    Environ. Pollut.

    (2020)
  • D. Saadoud et al.

    Application of fuzzy logic approach for wind erosion hazard mapping in Laghouat region (Algeria) using remote sensing and GIS

    Aeolian research

    (2018)
  • F. Shahid et al.

    Predictions for COVID-19 with deep learning models of LSTM, GRU and Bi-LSTM. Chaos

    Solitons & Fractals

    (2020)
  • F. Shahid et al.

    A novel genetic LSTM model for wind power forecast

    Energy

    (2021)
  • L. Shi et al.

    Temporal variation of dust emissions in dust sources over Central Asia in recent decades and the climate linkages

    Atmos. Environ.

    (2020)
  • L. Shi et al.

    Drivers to dust emissions over dust belt from 1980 to 2018 and their variation in two global warming phases

    Sci. Total Environ.

    (2021)
  • Y. Song et al.

    Spatio-temporal distribution of quaternary loess across central asia

    Palaeogeogr. Palaeoclimatol. Palaeoecol.

    (2021)
  • X. Su et al.

    A high-precision aerosol retrieval algorithm (HiPARA) for advanced himawari imager (AHI) data: development and verification

    Rem. Sens. Environ.

    (2021)
  • J. Sun

    Source regions and formation of the loess sediments on the high mountain regions of northwestern China

    Quat. Res.

    (2002)
  • K. Suresh et al.

    Provenance tracing of long-range transported dust over the Northeastern Arabian Sea during the southwest monsoon

    Atmos. Res.

    (2021)
  • Z. Wang et al.

    Investigating the all-sky surface solar radiation and its influencing factors in the Yangtze River Basin in recent four decades

    Atmos. Environ.

    (2021)
  • J. Xu et al.

    Computing payments for wind erosion prevention service incorporating ecosystem services flow and regional disparity in Yanchi County

    Sci. Total Environ.

    (2019)
  • C. Xu et al.

    Spatiotemporal variations and driving factors of dust storm events in northern China based on high-temporal-resolution analysis of meteorological data (1960–2007)

    Environ. Pollut.

    (2020)
  • X. Yang et al.

    Provenance of aeolian sediment in the Taklamakan Desert of western China, inferred from REE and major-elemental data

    Quat. Int.

    (2007)
  • M. Yang et al.

    Changes of the relationship between spring sand dust frequency and large-scale atmospheric circulation

    Atmos. Res.

    (2019)
  • J. Yin et al.

    Forecast of short-term daily reference evapotranspiration under limited meteorological variables using a hybrid bi-directional long short-term memory model (Bi-LSTM)

    Agric. Water Manag.

    (2020)
  • L. Yu et al.

    Clear-sky solar radiation changes over arid and semi-arid areas in China and their determining factors during 2001–2015

    Atmos. Environ.

    (2020)
  • Cited by (33)

    View all citing articles on Scopus

    Peer review under responsibility of Turkish National Committee for Air Pollution Research and Control.

    View full text