13235 Rate this article:

Multispectral vs. Hyperspectral Imaging – Enhancing Vegetation Mapping Accuracy


Data from multispectral satellite constellations like Landsat and SPOT have long been utilized for land use mapping and vegetation classification. Sentinel-2 builds on this legacy, providing similar moderate-resolution data acquired in broad spectral bands that cover visible, near infrared, and short-wave infrared wavelengths. Even though the MultiSpectral Instrument (MSI) payload of Sentinel-2 is tuned for vegetation mapping by sampling 13 well positioned spectral bands at spatial resolutions of 10m, 20m and 60m, in most cases hyperspectral imagery offers improved accuracy for this purpose. 

What is Hyperspectral Imaging?

While multispectral sensors capture images in a limited number of broad spectral bands, hyperspectral sensors acquire images with hundreds of narrow and contiguous spectral bands, ideally covering the entire electromagnetic spectrum. The increased spectral resolution of hyperspectral imagery enables the extraction of distinct spectral characteristics that may not be visible in multispectral images.

The additional spectral information provided by hyperspectral imaging allows for a deeper analysis of land surface features. It facilitates the identification and differentiation of similar objects, making it easier to classify vegetation types accurately.

A recent study by Jarocińska et al. published in Nature Scientific Reports (2023) investigated the extent to which additional spectral information improves the accuracy of identifying vegetation habitats with similar spectral properties. The study was conducted in five areas for non-forest EU Natura 2000 habitats and focused on four types of habitats: meadows, grasslands, heaths, and mires. (Figure 1).


Fig. 1: Images of analyzed vegetation ecosystems in Poland with habitat type and Natura 2000 code.

The study utilized multispectral data from the Sentinel-2 satellite and hyperspectral data from airborne HySpex sensors. Image preprocessing employed advanced spectral analytics using ENVI® software. ENVI is the industry standard for processing and analyzing hyperspectral data with the ability to capture the subtle spectral signatures in hyperspectral data. To ensure a fair comparison, the hyperspectral imagery was down-sampled to match the spatial resolution of Sentinel-2 (10m).

The results of the study demonstrated that hyperspectral data generally achieved higher classification accuracies compared to multispectral Sentinel-2 imagery, regardless of the habitat type. The F1 accuracy, on average, was 0.14 higher when using hyperspectral data (Figure 2). The authors conclude that the difference in accuracy was not constant, as it varied by area and habitat characterization. However, the authors emphasized that hyperspectral imagery was crucial for accurately mapping salt meadows (1340), Molinia meadows (6410), and lowland hay meadows (6510).


The study highlights the significant advantages of hyperspectral imaging over multispectral imaging for vegetation mapping. The increased spectral resolution enables more precise identification and differentiation of land surface features, ultimately improving classification accuracy.