Scholarlink

<p>This study proposes an integrated high-resolution optical remote-sensing workflow for monitoring agro-ecosystem dynamics and transition signals in Senegal’s Sédhiou region from 2020 to 2025. Sentinel-2 imagery was preprocessed in Google Earth Engine (cloud masking, harmonization, and compositing) and used to derive three spectral indices (NDVI, SAVI, and NDWI). Field data collected in March 2025 were used to train and validate the 2025 land-cover map, while the 2020–2025 component focused on vegetation-index dynamics and transition signals rather than year-by-year validated land-cover conversions. A Random Forest classifier mapped five land-cover classes in 2025, and achieved 94% overall accuracy and kappa 0.98, with class-specific classification errors of 0.0%, 1.0%, 4.2%, 7.0%, and 10.2% for Water, Built-up, Forest, Cropland, and Shrubland, respectively. Monthly time-series analysis revealed marked intra- and interannual variability across land-cover classes and identified transition-signal hotspots (persistent departures in index trajectories) using rolling variability metrics and breakpoint screening. Forests were comparatively stable (mean NDVI = 0.53), while shrublands showed intermediate dynamics (mean NDVI = 0.49). Cropland areas exhibited the highest variability, consistent with crop–fallow cycles and rainfall sensitivity in rainfed systems; this variability alone does not constitute evidence of land-cover transition. The most pronounced break episodes were observed in 2022 and 2024, especially in the Cropland class, and are interpreted as transition signals that may reflect rainfall-driven variability and or land-use dynamics, thus requiring cautious attribution. Overall, the workflow provides an operational framework, including image preprocessing, feature extraction, classification, index time-series extraction, and breakpoint screening, for identifying variability hotspots and prioritizing areas for targeted field verification and sustainable agro-ecosystem management in Sédhiou.</p> Graphical Abstract <p></p> <p><?tk 2?>This study explores the monitoring of agro-ecosystem dynamics and transition signals in the Sédhiou region of Senegal using optical remote sensing and cloud computing. The graphical summary presented in the image outlines the four main stages of the process: data acquisition, preprocessing, data processing, and results analysis. We built monthly cloud-free composites and calculated indices such as NDVI, SAVI, and NDWI using Sentinel-2 images on Google Earth Engine. These datasets were used to characterize seasonal and interannual variability in vegetation-index trajectories. Field reference polygons collected in 2025 were used to train and validate a Random Forest classifier for the 2025 baseline land-cover map, distinguishing forest, cropland, shrubland, built-up, and water classes. The classifier produced land cover maps that separate forests, croplands, shrublands, and other areas. Further analysis includes spatiotemporal aggregation of biophysical indices, monthly multi-index climatology, and change point detection to analyze agro-ecosystem dynamics and transition signals. The time series highlights seasonal cycles and interannual fluctuations across land-cover classes. Monthly climatologies provide additional detail on class-specific temporal behaviour. Combining multi-index satellite data with machine learning and time-series, helped us trace variability hotspots and potential transition signals relevant to agro-ecosystem monitoring in the Sédhiou region.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Time Series Approach to Monitoring Agro-ecosystem Dynamics and Transition Signals Using Vegetation Indices in Sédhiou, Senegal

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Labaly Toure,
  • Amandine Carine Njeugeut Mbiafeu,
  • Asseypo Célestin Hauhouot,
  • Beatrice Asenso Barnieh,
  • Pauline Gluski

摘要

This study proposes an integrated high-resolution optical remote-sensing workflow for monitoring agro-ecosystem dynamics and transition signals in Senegal’s Sédhiou region from 2020 to 2025. Sentinel-2 imagery was preprocessed in Google Earth Engine (cloud masking, harmonization, and compositing) and used to derive three spectral indices (NDVI, SAVI, and NDWI). Field data collected in March 2025 were used to train and validate the 2025 land-cover map, while the 2020–2025 component focused on vegetation-index dynamics and transition signals rather than year-by-year validated land-cover conversions. A Random Forest classifier mapped five land-cover classes in 2025, and achieved 94% overall accuracy and kappa 0.98, with class-specific classification errors of 0.0%, 1.0%, 4.2%, 7.0%, and 10.2% for Water, Built-up, Forest, Cropland, and Shrubland, respectively. Monthly time-series analysis revealed marked intra- and interannual variability across land-cover classes and identified transition-signal hotspots (persistent departures in index trajectories) using rolling variability metrics and breakpoint screening. Forests were comparatively stable (mean NDVI = 0.53), while shrublands showed intermediate dynamics (mean NDVI = 0.49). Cropland areas exhibited the highest variability, consistent with crop–fallow cycles and rainfall sensitivity in rainfed systems; this variability alone does not constitute evidence of land-cover transition. The most pronounced break episodes were observed in 2022 and 2024, especially in the Cropland class, and are interpreted as transition signals that may reflect rainfall-driven variability and or land-use dynamics, thus requiring cautious attribution. Overall, the workflow provides an operational framework, including image preprocessing, feature extraction, classification, index time-series extraction, and breakpoint screening, for identifying variability hotspots and prioritizing areas for targeted field verification and sustainable agro-ecosystem management in Sédhiou.

Graphical Abstract

This study explores the monitoring of agro-ecosystem dynamics and transition signals in the Sédhiou region of Senegal using optical remote sensing and cloud computing. The graphical summary presented in the image outlines the four main stages of the process: data acquisition, preprocessing, data processing, and results analysis. We built monthly cloud-free composites and calculated indices such as NDVI, SAVI, and NDWI using Sentinel-2 images on Google Earth Engine. These datasets were used to characterize seasonal and interannual variability in vegetation-index trajectories. Field reference polygons collected in 2025 were used to train and validate a Random Forest classifier for the 2025 baseline land-cover map, distinguishing forest, cropland, shrubland, built-up, and water classes. The classifier produced land cover maps that separate forests, croplands, shrublands, and other areas. Further analysis includes spatiotemporal aggregation of biophysical indices, monthly multi-index climatology, and change point detection to analyze agro-ecosystem dynamics and transition signals. The time series highlights seasonal cycles and interannual fluctuations across land-cover classes. Monthly climatologies provide additional detail on class-specific temporal behaviour. Combining multi-index satellite data with machine learning and time-series, helped us trace variability hotspots and potential transition signals relevant to agro-ecosystem monitoring in the Sédhiou region.