TY - JOUR

T1 - Machine Learning Methods of Satellite Image Analysis for Mapping Geologic Landforms in Niger: A Comparison of the Aïr Mountains, Niger River Basin and Djado Plateau

AU - Леменкова, Полина Алексеевна

PY - 2024/12/24

Y1 - 2024/12/24

N2 - This study analyses geological landforms and land cover types of Niger using spaceborne data. A landlocked African country rich in geological structures, Niger is notable for contrasting environmental regions which were examined and compared: 1) lowlands (Niger River basin); 2) Aïr Mountains; and 3) Djado Plateau. The methodology is based on machine learning (ML) models and programming applied for Earth observation data. Spatio-temporal analysis was performed using Landsat 8-9 OLI-TIRS multispectral images classified by GRASS GIS. Data were processed by scripts using ML algorithms by modules r.random, r.learn.train, r.learn.predict, i.cluster, and i.maxlik. The algorithms of the probabilistic forecasting included support vector machine (SVM), random forest (RF), decision tree classifier and K neighbors classifier. Variations in landscapes caused by water deficit and soil erosion were analysed, and parallels between geologic and environmental setting were drawn. The intra-landscape variability of patches within Niger is revealed from 2014 to 2024. Landscape patterns are affected by drought periods in central Niger, geological setting of mountains, distribution of crust karst pits and sinkholes in eastern Niger. Western region of the Niger River basin shown land cover patterns linked to hydrological effects of soil erosion. This paper shown the use of ML methods for geological-environmental analysis.

AB - This study analyses geological landforms and land cover types of Niger using spaceborne data. A landlocked African country rich in geological structures, Niger is notable for contrasting environmental regions which were examined and compared: 1) lowlands (Niger River basin); 2) Aïr Mountains; and 3) Djado Plateau. The methodology is based on machine learning (ML) models and programming applied for Earth observation data. Spatio-temporal analysis was performed using Landsat 8-9 OLI-TIRS multispectral images classified by GRASS GIS. Data were processed by scripts using ML algorithms by modules r.random, r.learn.train, r.learn.predict, i.cluster, and i.maxlik. The algorithms of the probabilistic forecasting included support vector machine (SVM), random forest (RF), decision tree classifier and K neighbors classifier. Variations in landscapes caused by water deficit and soil erosion were analysed, and parallels between geologic and environmental setting were drawn. The intra-landscape variability of patches within Niger is revealed from 2014 to 2024. Landscape patterns are affected by drought periods in central Niger, geological setting of mountains, distribution of crust karst pits and sinkholes in eastern Niger. Western region of the Niger River basin shown land cover patterns linked to hydrological effects of soil erosion. This paper shown the use of ML methods for geological-environmental analysis.

KW - machine learning

KW - geology

KW - satellite image

KW - remote sensing

KW - cartography

U2 - 10.5937/podrad2445027L

DO - 10.5937/podrad2445027L

M3 - Article

VL - 45

SP - 27

EP - 47

JO - Podzemni Radovi

JF - Podzemni Radovi

SN - 0354-2904

IS - 1

ER -