Dust detection and susceptibility mapping by aiding satellite imagery time series and integration of ensemble machine learning with evolutionary algorithms

Razavi-Termeh, Seyed Vahid; Sadeghi‐Niaraki, Abolghasem; Choi, Soo-Mi

doi:10.1016/j.envpol.2023.122241

Cited by 5 publications

(1 citation statement)

References 86 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The closest AUC=0.832 was found for LDA algorithm with the earlier study illustrates that the selection of different models, such as MLP and RF-FPA, used in current study can provide a better result. In a recent work the authors have used RF-FPA along with another two models for dust detection and susceptibility and the prediction accuracy for RF-FPA was recorded AUC = 0.981(Razavi-Termeh et al, 2023) which is very close to prediction accurate for the RF-FPA used in current work.…”

supporting

confidence: 72%

Spatial Distribution of Drought Vulnerability Mapping: Introducing a new methodology

Li,

Meng,

Arabameri

et al. 2023

Preprint

View full text Add to dashboard Cite

Droughts as a natural calamity have wreaked havoc on human health, environment, and the economy around the world. Due to its complex and multi-faceted nature, the risk assessment of drought requires the analysis of diverse parameters and machine learning techniques provide an effective tool to approach this problem. In the present work, we have employed four machine learning models, Naïve Bayes (NB), Rotational tree- Forest by Penalizing Attributes (RF-FPA), Multi-Layer Perceptron (MLP), and Linear Discriminant Analysis (LDA) for the drought vulnerability mapping in the Najafabad watershed, Isfahan Province, Iran. The country faces serious challenge from hydrological and meteorological drought conditions. A total of 20 conditioning factors comprising of 3 topographical (slope, elevation, geomorphology), 6 environmental (NDVI, soil depth, LU/LC, soil texture, EC, soil moisture), 4 hydrological (groundwater level, drainage density, distance to stream, TWI), 4 meteorological (annual precipitation and temperature, evaporation, humidity), and 3 socio-economic (ADP, deep tune, population density) were included for the drought vulnerability mapping. The collinearity effects were checked with multicollinearity analysis prior to the spatial modelling. The variable importance of the different parameters was analysed using AdaBoost model. The results show that soil moisture is the most important parameters among all variables. It also came into the results that the RF-FPA, among all four models, is the most successful model during training (AUC = 0.976) and validation (AUC = 0.968).

show abstract

supporting

confidence: 72%