Sentinel-2 and Landsat-8 potentials for high-resolution mapping of the shifting agricultural landscape mosaic systems of southern Cameroon

Abstract

Remotely sensed data is well-used for monitoring dynamics in land-use and land-cover (LULC). However, in previous studies, delineating small-size LULC within shifting agricultural landscape mosaic systems (SALMS) in the Congo basin has shown limitations when Landsat-7 is used, due to its spatial and spectral resolutions. Although Landsat-8 and Sentinel-2 data could provide a better contribution, their use in disaggregating these small-size LULC over degraded humid forest remains underexploited. This research evaluated the potential of Landsat-8 and Sentinel-2 data to support the disaggregation of small-size LULC within the SALMS. Statistical analyses of spectra data points were applied to (i) assess the spectral separability of LULC classes by bands, (ii) identify the suitable band-combinations, and (iii) compare the accuracy of both sensors for LULC mapping within the landscape. The results revealed that four Landsat-8 bands (2, 3, 5, and 6) and seven Sentinel-2 bands (2–4, and 7–11) showed good spectral separability. Two Landsat-8 band-combinations ([2–5-6] and [3–5-6]), with the highest Optimum Index Factor scores of 1416 and 1756 respectively, were suitable for colour-composition, while Sentinel-2 band-combination [8–11-4] did the same with a score of 473. At the reference level, there was no significant difference (p > 0.05) between Landsat-8 and Sentinel-2 generated maps. However, Sentinel-2, with 91 % accuracy, had a better capability to delineate small-size LULC within the landscape compared to Landsat-8 (85 % accuracy). This result provides an advantage towards accurate LULC monitoring with respect to carbon stock change estimates within the SALMS in the Congo, Amazonian and Southeast Asia basins.

Keywords

• Heterogeneous landscape
• Landsat-8
• Sentinel-2
• Small-size LULC disaggregation
• Spectra data points