More Than Meets the Eye: Using Sentinel-2 to Map Small Plantations in Complex Forest Landscapes

Abstract

Many tropical forest landscapes are now complex mosaics of intact forests, recovering forests, tree crops, agroforestry, pasture, and crops. The small patch size of each land cover type contributes to making them difficult to separate using satellite remote sensing data. We used Sentinel-2 data to conduct supervised classifications covering seven classes, including oil palm, rubber, and betel nut plantations in Southern Myanmar, based on an extensive training dataset derived from expert interpretation of WorldView-3 and UAV data. We used a Random Forest classifier with all 13 Sentinel-2 bands, as well as vegetation and texture indices, over an area of 13,330 ha. The median overall accuracy of 1000 iterations was >95% (95.5%-96.0%) against independent test data, even though the tree crop classes appear visually very similar at a 20 m resolution. We conclude that the Sentinel-2 data, which are freely available with very frequent (five day) revisits, are able to differentiate these similar tree crop types. We suspect that this is due to the large number of spectral bands in Sentinel-2 data, indicating great potential for the wider application of Sentinel-2 data for the classification of small land parcels without needing to resort to object-based classification of higher resolution data.

Oil palm concessions in southern Myanmar consist mostly of unconverted forest

“…NDVI was selected over other indices (e.g. EVI, LSWI, SATVI) based on our previous study, which successfully classified similar landscapes in this region with high accuracy 34 . Including other indices did not result in significant changes in accuracy for our study.…”

“…We estimated the area of oil palm as well as other land cover classes by conducting a machine learning classification on Sentinel-1 and Sentinel-2 satellite data from 2018–2019 using thousands of reference data points. Sentinel data are provided at a high resolution (~10 m for Sentinel-1 and 10–60 m for Sentinel-2) with very narrow bands that enable some differentiation of spectral responses from vegetation that was previously only possible using hyperspectral sensors (Sentinel-2’s B5-7, 8A), which is necessary to classify the region with complex landscapes with small patches of plantations and forests 34 . Furthermore, their frequent revisits (6/12-day for Sentinel-1, 5-day for Sentinel-2) made it possible to create a high-quality composite based on the average of many scenes, reducing radar speckle (S1), sun-angle, and seasonal effects (S2).…”

Acquisition of Forest Attributes for Decision Support at the Forest Enterprise Level Using Remote-Sensing Techniques—A Review

“…Results of automatic pixel based classification was evaluated as 98.47% accurate when compared to manual delineation of aerial imagery. Another recent work [27] showed the possibility of mapping small plantations inside forested regions deploying Sentintel-2 images and Random Forest method achieving median overall accuracy larger than 95% (95.5%-96.0%) against independent test data.…”

Unmanned aerial vehicle (UAV) derived structure-from-motion photogrammetry point clouds for oil palm (Elaeis guineensis) canopy segmentation and height estimation

“…The challenge with using readily available satellite data for oil palm science is the limited spatial and/or temporal resolution of such data. The demand for finer spatial resolution data is motivated by the ability to resolve individual palm canopies, which can be used as reference data to improve land-cover classifications (Nomura and Mitchard 2018) and allows for automated identification and parameter retrieval to provide information about palm structure and status. These parameters are not only of interest for plantation management but are central to the estimation of plantation carbon stocks, e.g.…”

“…Manual identification based on spatial data products is still considered the most accurate and cost-effective method to generate inventories in commercial applications and the selection of training data in recent scientific applications (e.g. Nomura and Mitchard 2018). There are however promising first demonstrations of machine learning techniques on both fine spatial resolution satellite and UAV image data of oil palms and date palms respectively (Weijia Li et al 2016b;Malek et al 2014) as well as software packages for the operational implementation of object-based segmentation (eCognition, Trimble, California, USA) for palm identification.…”