TY - JOUR

T1 - AI techniques for feature extraction, simplification and aggregation geospatial data

AU - Леменкова, Полина Алексеевна

PY - 2026/8/9

Y1 - 2026/8/9

N2 - This study proposes an Artificial Intelligence (AI) application to cartographic workflow with the purpose of environmental mapping. The Random Forest (RF) framework of Machine Learning (ML) was applied to satellite image analysis for detection and robust attribution of biodiversity changes in Africa. Changes in land cover types were investigated on Landsat using comparison of several algorithms integrated in data processing and cartographic scripts using Geographic Resources Analysis Support System (GRASS) Geographic Information System (GIS). The analysis of time series highlights biodiversity changes in north Namibian landscapes in relation to presumed impacts of multiple potential environmental drivers: climate-related factors in desert setting (high fluctuations in temperature, salt crust in the saline lake and low precipitation) and anthropogenic forces (land use, abandoned areas and agriculture). The computational results demonstrated following outcomes. According to the calculated number of pixels corresponding to the cells in the raster images covering the Etosha landscapes, the land use types demonstrated the following changes: 1. Mosaic vegetation and cropland (8.42%), Aquatic or regularly flooded vegetation decreased on 13,59% due to seasonal fluctuations, areas of artificial surfaces declined insignificantly (2.40%), Mixed broadleaved deciduous forests experienced slight decreasd (1.18%), and the class of open grasslands increased to 5.32 %. Moreover, rainfed croplands demonstrated stable development with 4.58% of changes, while salt hardpans changed significantly to 12, 73%. Finally, water bodies including Etosha basin decreased significantly due to the seasonal effects (21.73%), Herbaceous vegetation, shrubland and thickets developed relatively stable with minor changes not exceeding 3.25 %, and sparse savannah grasslands occupied similar areas with fluctuations not exceeding 4,72%. The accuracy of image classification was evaluated by kappa statistics and chi-square against the ground truth data obtained from the land cover and soil data inventory. The original data with Land Use Land Cover (LULC) thematic classes were reclassified into 10 categories. The results show variations in land cover change due to arid desert climate in northern Namibia. This provides evidence of biodiversity changes in Africa detected using AI-driven data analysis using multispectral satellite images.

AB - This study proposes an Artificial Intelligence (AI) application to cartographic workflow with the purpose of environmental mapping. The Random Forest (RF) framework of Machine Learning (ML) was applied to satellite image analysis for detection and robust attribution of biodiversity changes in Africa. Changes in land cover types were investigated on Landsat using comparison of several algorithms integrated in data processing and cartographic scripts using Geographic Resources Analysis Support System (GRASS) Geographic Information System (GIS). The analysis of time series highlights biodiversity changes in north Namibian landscapes in relation to presumed impacts of multiple potential environmental drivers: climate-related factors in desert setting (high fluctuations in temperature, salt crust in the saline lake and low precipitation) and anthropogenic forces (land use, abandoned areas and agriculture). The computational results demonstrated following outcomes. According to the calculated number of pixels corresponding to the cells in the raster images covering the Etosha landscapes, the land use types demonstrated the following changes: 1. Mosaic vegetation and cropland (8.42%), Aquatic or regularly flooded vegetation decreased on 13,59% due to seasonal fluctuations, areas of artificial surfaces declined insignificantly (2.40%), Mixed broadleaved deciduous forests experienced slight decreasd (1.18%), and the class of open grasslands increased to 5.32 %. Moreover, rainfed croplands demonstrated stable development with 4.58% of changes, while salt hardpans changed significantly to 12, 73%. Finally, water bodies including Etosha basin decreased significantly due to the seasonal effects (21.73%), Herbaceous vegetation, shrubland and thickets developed relatively stable with minor changes not exceeding 3.25 %, and sparse savannah grasslands occupied similar areas with fluctuations not exceeding 4,72%. The accuracy of image classification was evaluated by kappa statistics and chi-square against the ground truth data obtained from the land cover and soil data inventory. The original data with Land Use Land Cover (LULC) thematic classes were reclassified into 10 categories. The results show variations in land cover change due to arid desert climate in northern Namibia. This provides evidence of biodiversity changes in Africa detected using AI-driven data analysis using multispectral satellite images.

KW - Remote sensing

KW - artificial intelligence

KW - machine learning

KW - random forest

KW - cartography

U2 - 10.35208/ert.1685945

DO - 10.35208/ert.1685945

M3 - Article

VL - 9

SP - 677

EP - 692

JO - Environmental Research and Technology

JF - Environmental Research and Technology

SN - 2636-8498

IS - 4

ER -