Article Sidebar
Published:
Aug 6, 2024Keywords:
Article Details
Authors who submit papers to this journal agree to the following terms:
a) Authors retain copyright over their work, while allowing the journal to place this work on the journal website under a Creative Commons Attribution License, which allows others to freely access, use, and share the work, with an acknowledgment of the work's authorship and its initial publication in this journal.
b) Authors are able to waive the terms of the CC license and enter into separate, additional contractual arrangements for the non-exclusive distribution and subsequent publication of this work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgment of its initial publication in this journal.
c) In addition, authors are encouraged to post and share their work online (e.g., in institutional repositories or on their website) at any point after publication on the journal website.
Main Article Content
Mapping smallholder maize farm distribution using multi-temporal Sentinel-1 data integrated with Sentinel-2, DEM and CHIRPS precipitation data in Google Earth Engine
Abstract
Mapping smallholder maize farms in complex and uneven rural terrain is a major barrier to accurately documenting the spatial representation of the farming units. Remote sensing technologies rely on various satellite products for differentiating maize cropland cover from other land cover types. The potential for multi-temporal Sentinel-1 synthetic aperture radar (SAR), Sentinel-2, digital elevation model (DEM) and precipitation data obtained from Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) version 2.0 was investigated for mapping maize crop distributions during the growing seasons, 2015–2021, in the Sekhukhune municipal area of Limpopo, a province in South Africa. Sentinel-1 variables, including monthly VH, VV, VV+VH (V = vertical, H = horizontal) polarization band data and data issuing from the principal component analysis of VH polarization were integrated with Sentinel-2-derived normalized difference vegetation index (NDVI), DEM terrain, and precipitation data. The random forest (RF) algorithm was applied to distinguish maize crops from four other land cover types, including bare soil, natural vegetation, built-up area, and water. The findings indicated that the models that used only Sentinel-1 data as input data had overall accuracies below 71%. The best performing models producing overall accuracies above 83% for 2015–2021 were those where Sentinel-1 (VV+VH) data were integrated with all the ancillary data. Overall, the McNemar test indicated enhanced performance for models where all other ancillary input data had been incorporated. The results of our study show considerable temporal variation in maize area estimates, with 59 240.84 ha in the 2018/2019 growing season compared to 18 462.51 ha in the 2020/2021 growing season. The spatial information gathered through these models proved to be valuable and is essential for addressing food security, one of the objectives of the Sustainable Development Goals.
