12-17, 10:00–10:15 (Asia/Bangkok), Room34-1104
Spaceborne hyperspectral data can assist in estimating crop yields, predicting crop outcomes, and monitoring crops, which ultimately contributes to loss prevention and food security. Recently, EnMAP hyperspectral imagery has become available, starting from 2022. This study aims to analyze and classify oil palm and nipa palm plantations using hyperspectral images combined with machine learning algorithms. The Random Forest classifier, CatBoost classifier, and LightGBM classifier were utilized to automatically map the oil palm and nipa palm areas. The fully workflows of the hyperspectral imaging process were performed in EnMAP-Box plugin on QGIS software. The overall accuracy of three ML classfiers provides greater than 90% especially in oil palm and nipa palm plantations. Machine learning can find out the hidden information about spectral characteristics.
This study utilized hyperspectral imagery from the Environmental Mapping and Analysis Program (EnMAP) by the German Space Agency to classify oil palm and nipa palm plantations using three different machine learning methods. We also analyzed the spectral characteristics of various palms using spaceborne images with a 30-meter spatial resolution and 224 spectral bands, including VNIR and SWIR (420-2450 nm). The EnMAP images, available as free open data since 2022, were acquired through https://www.enmap.org/. The EnMAP hyperspectral data were visualized and processed using the EnMAP-Box, a free and open-source plug-in for QGIS. This plug-in offers several algorithms for processing high-dimensional imagery, such as band selection, dimensionality reduction, classification, and regression.
In addition, we employed Random Forest, CatBoost, and LightGBM classifiers, which are effective in handling the complexities of hyperspectral datasets. We defined classes for oil palm and nipa palm, along with other land cover types such as built-up areas, water bodies, mangroves, and forests, to create training and testing datasets. Multiple EnMAP images from Surat Thani and Nakhon Si Thammarat provinces in Thailand were used to classify oil palm and nipa palm areas. Additionally, Landsat 8 & 9 images with similar 30-meter spatial resolution were utilized to compare classification results between spaceborne multispectral and hyperspectral imagery. The results of this study provide accurate information on oil palm and nipa palm plantations, which are crucial for crop monitoring and management decision-making systems. Also, it can lead to large-scale area management of oil palm and nipa palm plantation.
