DTW-kNN and Sentinel-2 NDVI Time Series for Field-Scale Maize Identification and Multi-Crop Classification in an African Smallholder Landscape
Applied Sciences · 09.11.2026 · 16 (21) · 11065
DOI: 10.3390/app162111065 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 09.11.2026. Verfasst von Wonga Masiza und 2 weiteren Autor:innen. Die Arbeit nennt 77 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Earth Observation and Space Secretariat.
Abstract (Originalsprache)
Mapping crops at the field scale supports agricultural planning, production monitoring and resource allocation. However, this remains difficult in smallholder landscapes, where planting dates vary among farms, ground-reference data are limited, and fields are small, fragmented and irregularly shaped. This study evaluated Sentinel-2 Normalised Difference Vegetation Index (NDVI) time series combined with Dynamic Time Warping k-nearest neighbour (DTW-kNN) for maize identification and multi-crop classification in the O.R. Tambo District Municipality, South Africa.. A maize reference trajectory was first developed from known maize fields and then assessed against independent maize validation fields and candidate unknown fields. Known maize fields produced low DTW distances to the maize reference trajectory, while Days After Planting alignment showed a clear maize phenological pattern, with NDVI increasing after planting, peaking between approximately 90 and 120 DAP, and declining thereafter. For multi-crop classification, a simple mean-reference DTW baseline achieved 80.0% accuracy on independent validation fields. Retaining individual reference-field trajectories (DTW-kNN) instead of a single mean reference improved on this baseline. Training-only cross-validation selected DTW-kNN with k = 1 and a one-composite warping window; this configuration achieved 92.9% independent validation accuracy, comparable to benchmark Random Forest and Support Vector Machine classifiers (95.7%). DTW-kNN remained competitive while retaining direct temporal-similarity interpretation. When applied to 128 unlabelled fields, the selected DTW-kNN classifier predicted 106 fields as maize and 22 as soybean, with 65.6% of predictions classified with high confidence. The findings show that Sentinel-2 NDVI time series combined with DTW-kNN can support interpretable field-level crop classification in fragmented smallholder farming systems.
Förderung
- National Earth Observation and Space Secretariat
Zitieren
APA
Masiza, W., Khoboko, P. W., & Chirima, J. G. (2026). DTW-kNN and Sentinel-2 NDVI Time Series for Field-Scale Maize Identification and Multi-Crop Classification in an African Smallholder Landscape. Applied Sciences, 16(21), 11065. https://doi.org/10.3390/app162111065
Vancouver
Masiza W, Khoboko PW, Chirima JG. DTW-kNN and Sentinel-2 NDVI Time Series for Field-Scale Maize Identification and Multi-Crop Classification in an African Smallholder Landscape. Applied Sciences. 2026;16(21):11065. doi:10.3390/app162111065
BibTeX
@article{Masiza2026,
author = {Masiza, Wonga and Khoboko, Pitso Walter and Chirima, Johannes George},
title = {DTW-kNN and Sentinel-2 NDVI Time Series for Field-Scale Maize Identification and Multi-Crop Classification in an African Smallholder Landscape},
journal = {Applied Sciences},
year = {2026},
volume = {16},
number = {21},
pages = {11065},
doi = {10.3390/app162111065}
}
EUVIMED