How to identify oil palm growing areas using satellite remote sensing data

Oil palm is an important economic crop, and its distribution information is crucial for agricultural monitoring, land use assessment, and resource management. This paper proposes a multi-source remote sensing method combining Sentinel-1 and Sentinel-2 data for high-precision extraction of oil palm distribution information. Sentinel-1 provides all-weather, all-time radar imagery, unaffected by weather conditions; Sentinel-2 provides optical imagery containing rich information such as vegetation indices and texture features. The complementarity of these two sources provides favorable conditions for oil palm identification.

Research Methods

A specific region in Indonesia was selected as the study area (Figure 1). The planting area in this region in 2025 was extracted, and the data selection period was from June 1, 2024 to May 31, 2025. The data was preprocessed using the GEE platform, including band extraction, index calculation, and SAR data denoising, as shown in Figure 2. The data was then resampled to a 10-meter resolution, with the coordinate system being WGS84.

Figure 1 Study Area

Figure 2(a) VV

Figure 2(b) NDVI

Figure 3 shows the data features sorted using Gini coefficients. A total of 14 features from Sentinel-1 and Sentinel-2 data were selected for the Gini coefficients, with 8 features from Sentinel-2 and 6 from Sentinel-1. Using SAR data as the primary feature for classification, the VH and VV bands and their differences showed high correlation in the classification model, making these features crucial for detecting oil palm plantations. Furthermore, texture information and vegetation indices can also be used for vegetation classification; texture features improved the detection accuracy of oil palm more effectively than vegetation indices.

Figure 3. Ranking of Gini coefficients

Extraction results

VV, VH, B3, B8, NDVI, along with texture entropy and contrast, were selected for feature construction. Then, oil palm regions within the study area were extracted using random forest. The extraction results are shown in Figure 4.

Figure 4 Extraction results from oil palm plantation area

This paper uses a data fusion method combining Sentinel-1 and Sentinel-2 data to extract oil palm plantation areas. Through feature construction and random forest techniques, high-precision plantation area identification is achieved. The research indicates that multi-source remote sensing data fusion is an important direction for improving agricultural vegetation identification capabilities. Future work could further explore deep learning-based semantic segmentation methods to improve boundary accuracy and time-series dynamic analysis capabilities.