Pansharpening largely preserves the normalized difference vegetation index: a multi-sensor comparison with PRISMA, Landsat 9, and field spectroscopy

Pansharpening is widely used to increase the spatial resolution of satellite imagery, yet its effects on vegetation indices derived from hyperspectral sensors remain poorly quantified. This study tests how five standard software pansharpening algorithms alter Normalized Difference Vegetation Index (NDVI) values derived from PRISMA hyperspectral and Landsat 9 multispectral data. We analysed 14 spectrally homogeneous plots within a 6.5 km² agricultural site in the Czech Republic in 2022 using acquisitions from Landsat 9 (04 August), PRISMA (07 August), ASD FieldSpec 4 field spectrometer, and UAV-borne MicaSense RedEdge-MX measurements (08 August). Across algorithms, deviations in NDVI relative to native imagery were statistically insignificant for both sensors (Landsat 9 and PRISMA); however, algorithm choice affected accuracy. The Principal Component (PC) Nearest Neighbor method had the largest errors (MAE = 0.093 for PRISMA; 0.078 for Landsat 9) and, for PRISMA, a positive bias relative to FieldSpec 4 (bias = +0.111). In contrast, NN Diffuse and Local Mean and Variance Matching (LMVM) produced the lowest errors. For PRISMA, NN Diffuse achieved an MAE of 0.049 and a bias of +0.040, while LMVM yielded an MAE of 0.050 and a bias of +0.016. For Landsat 9, NN Diffuse reached an MAE of 0.052 and a bias of +0.014, and LMVM produced an MAE of 0.059 and a bias of −0.018. UAV-derived NDVI agreed less with ASD FieldSpec 4 than the native satellite data (MAE = 0.056). Our results provide the first empirical assessment of pansharpening effects on PRISMA hyperspectral NDVI and demonstrate that, with robust algorithms and careful preprocessing, pansharpening does not materially distort NDVI. While the study is restricted to a single site, a single growing season, and one vegetation index, it highlights methodological considerations that can inform broader applications of hyperspectral pansharpening not only in vegetation monitoring.

1 Introduction

Hyperspectral satellite sensors capture finely resolved spectral information across hundreds of contiguous narrow bands (typical FWHM of ∼6–12 nm) across ∼400–2,500 nm (VNIR–SWIR) spectral region (depending on mission) (Gao et al., 2009), thus enabling precise discrimination of surface phenomena and traits with unique spectral signatures. The pioneer in this field was NASA’s Hyperion sensor aboard the EO-1 mission (2000–2017), which acquired 220 contiguous spectral bands with a spatial resolution of about 30 m (Pearlman et al., 2001; Mallinis et al., 2014). Current state-of-the-art missions include satellite systems such as Gaofen-5 (CNSA, China—launched in 2018) (Chen et al., 2022), PRISMA (ASI, Italy—launched in 2019) (Cogliati et al., 2021) and EnMAP (DLR, Germany—launched in 2022) (Erhard et al., 2017; Storch et al., 2023), and scientific missions deployed on the International Space Station (ISS), for example, the DESIS (DLR, Germany—launched in 2018) (Mahlayeye et al., 2024), the HISUI (JAXA, Japan—launched in 2019) (Matsunaga et al., 2017), and the EMIT (NASA, United States—launched in 2022) (Green et al., 2020). From the European environmental perspective, the most important are PRISMA and EnMAP satellites (Shaik et al., 2023; Qian, 2021). The Italian PRISMA mission employs a push-broom Hyper-Spectral imager that captures a continuum of 239 spectral bands, achieving spatial resolution of 30 m, complemented by a co-registered 5 m panchromatic band (Vangi et al., 2021). The German EnMAP mission also carries the Hyper-Spectral imager, which covers approximately 242 spectral channels, with 30 m spatial resolution (Guanter et al., 2015). Future capabilities are expected to expand substantially with the planned CHIME mission (ESA, Copernicus program—earliest launch ∼2028), designed to provide consistent, long-term global hyperspectral coverage at high radiometric and spectral fidelity (Nieke et al., 2023).

The PRISMA satellite is unique today due to the availability of a panchromatic band, which ensures its distinctive spatial resolution in the field of hyperspectral satellite systems (Alparone et al., 2024). To exploit the full potential of this data, it is necessary to use appropriate methods for fusing the high-resolution panchromatic image with the lower resolution hyperspectral data, resulting in a higher resolution image with improved spatial and spectral information (Loncan et al., 2015). Pansharpening is a widely used class of techniques for these tasks (Mookambiga and Gomathi, 2016). Many comprehensive reviews of different pansharpening approaches highlight their pros and cons (Amro et al., 2011; Sarp, 2014; Xu et al., 2014; Vivone et al., 2015; 2021; Pushparaj and Hegde, 2017; Dadrass Javan et al., 2021; Yilmaz et al., 2022; Ciotola et al., 2025). Gram-Schmidt, Principal Component (PC) (Cánovas-García et al., 2020; Toosi et al., 2020; Yilmaz et al., 2020), and the Nearest Neighbor Diffusion (NN Diffuse) pansharpening (Snehmani et al., 2017) are the most common algorithms for RGB and VNIR multispectral data processing, which are available in standard GIS and Remote Sensing (RS) software such as ENVI (NV5 Geospatial Software, Colorado, United States), QGIS (Orfeo Toolbox, QGIS Development Team), or ArcGIS Pro (ESRI, California, United States). However, pansharpening may not be only the domain of RGB and multispectral data, but more studies deal with hyperspectral (Ducay and Messinger, 2020; Ducay and Messinger, 2022; Seydi and Hasanlou, 2018; Ziaja et al., 2023; Alparone et al., 2024) or thermal-infrared imagery (Mushore et al., 2022). In recent years, hyperspectral pansharpening has evolved into a distinct research field, with advanced approaches such as Coupled Nonnegative Matrix Factorisation (CNMF), HySure, Bayesian fusion frameworks, and, more recently, deep learning architectures including convolutional neural networks (CNN), transformers, and diffusion models (Ciotola et al., 2025). By contrast, legacy techniques like Gram-Schmidt, PC, or NN Diffuse, which are readily available in ENVI, QGIS, or ArcGIS, are not specifically designed to preserve hyperspectral spectral integrity, which justifies testing their impact on spectral indices such as Normalized Difference Vegetation Index (NDVI). In the case of the PRISMA imagery, the pansharpening has been successfully used for marine plastic litter detection (Kremezi et al., 2021; Taggio et al., 2022), geo-archaeological prospections (Sech et al., 2024), classifying urban trees (Perretta et al., 2024), Local Climate Zones (Vavassori et al., 2024) and glacial lakes (Matta et al., 2024) mapping, insect pest monitoring (Sári-Barnácz et al., 2024), and asbestos detection (Shi et al., 2022). These studies show that image pre-processing relates to selecting an appropriate pansharpening method, which is crucial for accurate remote sensing-based analysis (Jawak et al., 2022).

Compared to raw imagery, derivative products (such as spectral indices) provide a more practical basis for applications by non-expert users, including farmers, foresters, and ecologists. Spectral indices are widely used as quantitative indicators derived from combinations of appropriate spectral bands acquired by RS sensors. Spectral indices are applicable to different environmental tasks, such as vegetation monitoring (Zeng et al., 2022), soil characteristics mapping (Lebrun et al., 2024), and land cover change detection (Xue and Su, 2017). From the list of available spectral indices (Bannari et al., 1995; Henrich et al., 2009), NDVI is very probably the most frequently employed index, which has gained a prominent position in remote sensing due to its calculation simplicity, strong theoretical foundation, and broad interpretability. Its long-term and repeated use across different sensor systems confirms its robustness and its essential role in ecological and environmental research at both local and global scales (Huang et al., 2021). In addition, NDVI can be calculated from broadband (represented mainly by multispectral) and narrowband (hyperspectral) sensors. NDVI has been extensively applied to assess vegetation health and stress (Pettorelli et al., 2005; Klouček et al., 2019; 2024), to support land use and land cover mapping and change detection (Zhu and Woodcock, 2014; Klouček et al., 2018), and to derive spatial patterns of selected soil properties, such as organic matter content and nutrient availability (Lebrun et al., 2024). Despite its enduring popularity and value for long-term monitoring, the NDVI has well-documented limitations that constrain its interpretability. It tends to saturate under high leaf area and biomass conditions, is sensitive to soil and background effects in sparsely vegetated areas, and its broadband formulation overlooks finer spectral features related to chlorophyll content and plant stress (Gao et al., 2023; Maluleke et al., 2024). The reliability of every spectral index is conditioned by the quality of the input data, particularly in terms of spectral and spatial uncertainty. In the case of radiometric credibility, represented by the surface reflectance, atmospheric corrections (Song et al., 2001; Moravec et al., 2021) and the selected pansharpening method play an irreplaceable role in ensuring data consistency and comparability across space and time (Lin et al., 2015).

While pansharpening effects on multispectral Landsat-based NDVI are documented, it is still unclear if these findings hold for the more spectrally complex PRISMA hyperspectral data, or how standard software algorithms would perform. Examples based on multispectral data highlight the importance of selecting an appropriate combination of pansharpening method and sensor for calculating vegetation indices. For instance, Johnson (2014) demonstrated, using the case of Landsat 8 multispectral pansharpening, that methods Fast Intensity-Hue-Saturation (FIHS) and Additive Wavelet Transform (AWT) did not introduce a significant bias in NDVI values. In contrast, Jovanović et al. (2016) showed that other methods, such as the Brovey Transform, can substantially affect the resulting NDVI, with RMSE values before and after transformation ranging from 0.21 to 0.30 depending on the satellite used (Landsat 7, Landsat 8, WorldView-2, or Ikonos).

However, to our knowledge, this research represents one of the first systematic evaluations of how standard software pansharpening algorithms affect hyperspectral PRISMA imagery-based NDVI accuracy. Although modern pansharpening methods based on advanced optimisation algorithms or deep learning achieve high accuracy in research studies, their practical application is often hindered by limited user accessibility. These methods are typically available only as research codes or prototype implementations, requiring substantial computational resources and careful parameter tuning. Consequently, for many practitioners, the application of these techniques remains challenging, which continues to favour the use of traditional mainstream algorithms. Our study aims to answer the question of how routine data pre-processing using standard pansharpening methods available in the mainstream software affects the values of the NDVI vegetation index calculated from recent hyperspectral (PRISMA) and multispectral (Landsat 9) imagery compared to in situ (ASD FieldSpec 4) and close-range (senseFly eBee X equipped by MicaSense RedEdge MX) reference measurements. The analysis is guided by four partial hypotheses: (a) The accuracy of tested pansharpening methods will differ significantly; (b) Standard pansharpening methods in mainstream software will perform better for multispectral than for hyperspectral pansharpening; (c) Depending on the method, NDVI derived from pansharpened data may deviate from NDVI derived from native PRISMA and Landsat 9 imagery; and (d) UAV-based multispectral orthomosaics (eBee X and MicaSense RedEdge-MX) can serve as a valuable reference for reflectance validation.

2 Materials and methods

2.1 Study area

The study area covers 6.5 km² (50 ° 06′N, 13 ° 51′E) and is situated in the Czech Republic, around 40 km west of Prague in the Central Bohemian Region (Figure 1). The central part is located at 450 m MSL and is occupied by the Czech University of Life Sciences Prague (CZU) farm called “Amalie.” Most of the study area is part of the protected landscape Křivoklátsko (Kuželková et al., 2024). From an agricultural point of view, there are perennial forage and fields where typical Central European crops such as wheat, barley and rapeseed are grown. For more details about the study area, see the Centre for Water, Soil and Landscape (CWSL) CZU initiative (CWSL, 2025).

Figure 1

Figure 1. Introduction of the study site. The background orthomosaic was acquired on 08 August 2022 by the RGB sensor of the senseFly Duet T camera placed on fixed wing eBee X. The coloured dots represent field spectroscopy measurements using ASD FieldSpec 4 performed on 08 August 2022.

2.2 Field measurement

We performed two field spectroscopy campaigns using the ASD FieldSpec 4 Standard-Res spectrometer (Malvern Panalytical, Malvern, United Kingdom). Spectral data were acquired on 04 July and 08 August 2022, selected to coincide with the acquisition dates of the complementary satellite data. Fourteen spectrally homogeneous plots representing different LULC categories were chosen across the study area (Figure 1). Ten spectroscopy measurements were simultaneously surveyed using a Trimble R580 RTK GNSS rover (Trimble, Sunnyvale, California, United States), see Figure 2, and the mean spectral reflectance per plot was calculated. Collected spectral data was processed using the supplied RS3 6.4 and ViewSpec Pro 6.2 software (Malvern Panalytical, Malvern, United Kingdom). Although field spectra were collected on two dates, the 08 August 2022 campaign, conducted between 09:00 and 12:00 GMT under a sunny and cloudless sky, provided the best temporal overlap with UAV, PRISMA and Landsat 9 at acquisitions (see details in chapters below). Therefore, the August dataset was used as the primary reference for inter-sensor comparison.

Figure 2

Figure 2. Field spectral reflectance measurements acquired with an ASD FieldSpec4 spectroradiometer on 08 August 2022. The main plot (A) displays spectral signatures for six different surface cover types: bare soil, perennial forage, standing crops, harvested crops, and their residues. The smaller, faceted plots (B) provide a detailed view of the characteristic reflectance magnitude and shape differences for each of these cover types across the 400–2,500 nm spectral range. The curves illustrate the distinct reflective properties of both vegetated and non-vegetated surfaces.

2.3 Acquisition and processing of UAV data

We performed two UAV missions in the 2022 vegetation season on 04 July (7:30–10:30 GMT) and 08 August (08:30–11:30 GMT) using a fixed-wing eBee X (senseFly, Cheseaux-sur-Lausanne, Switzerland) equipped with MicaSense RedEdge-MX (AgEagle Aerial Systems Inc., Wichita, Kansas, United States) and senseFly Duet T (senseFly, Cheseaux-sur-Lausanne, Switzerland) cameras, see Table 1. MicaSense RedEdge-MX is a multispectral camera with five narrow bands (Blue: 475 ± 16 nm; Green: 560 ± 13.5 nm; Red: 668 ± 8 nm; Red Edge: 717 ± 6 nm; and NIR: 842 ± 28.5 nm). Duet T is a hybrid sensor that captures simultaneously RGB and thermal infrared imagery. We planned and performed imagery acquisition in eMotion ground station software version 3.20 (senseFly, Cheseaux-sur-Lausanne, Switzerland) at approximately 300 m above ground level with 75% longitudinal and 70% lateral overlaps. Each mission consisted of three flights that took 50 min and covered an area of 6.5 km². The flight dates were selected based on weather conditions (eliminating strong wind, rain, etc.) and the PRISMA satellite overpass (Figure 3). On 04 July, weather conditions were calm with wind speeds of 1–3 m/s, air temperatures between 19 °C and 22 °C and relative humidity of 47%–49%. On 08 August, moderate wind conditions were recorded with wind speeds of 4–5 m/s, temperatures ranging from 21 °C to 22 °C and relative humidity 59%–61%.

Table 1

Table 1. List fixed-wing senseFly eBee X flights using MicaSense RedEdge-MX and senseFly Duet T cameras performed on 04 July and 08 August 2022, and their selected parameters relevant to study objectives.

Figure 3

Figure 3. UAV missions were performed on 04 July and 08 August using senseFly eBee X equipped with MicaSense RedEdge-MX and senseFly Duet T cameras ((A) – photo from 04 July), supported by measurement of 35 ground control points (GCPs) by a Trimble R580 RTK GNSS rover ((B) - photo from 04 July).

The imagery was processed using the Metashape image-matching software version 2.0.3 (Agisoft LLC, Saint Petersburg, Russia). The workflow includes standard Structure from Motion (SfM) processing steps, and, beyond UAV imagery, 35 ground control points (GCPs) surveyed by a Trimble R580 RTK GNSS rover (Trimble, Sunnyvale, California, United States) (Chakhvashvili et al., 2024) were used (Figure 3). UAV image processing resulting in two radiometrically corrected (using a calibrated reflectance panel MicaSense RP04 and downwelling light sensor—DLS) multispectral orthomosaics (used as the primary object in the subsequent analyses) and two RGB orthomosaics (used for visualisation and clarification purposes) in the Czech national SRS systems–S-JTSK/Krovak East North (EPSG: 5,514) and Baltic Sea vertical datum (EPSG: 8,357) with a sub-meter spatial resolution, see Table 1.

2.4 Downloading of Landsat products

In summer 2022, the multispectral imagery from Landsat 8 and Landsat 9 satellites (Operational Land Imager sensor - OLI; respectively OLI-2 in the case of Landsat 9) (Barsi et al., 2024) with nine spectral bands (0.43–2.29 μm) with 30 m spatial resolution and one panchromatic band (PAN) (0.50–0.68 µm) with 15 m spatial resolution were available. We downloaded a Landsat Collection 2 Level-2 Science Products (L2SP) calibrated on surface reflectance (Vermote et al., 2016) directly from the data provider (United States Geological Survey; USGS) through the EarthExplorer service. Due to frequent clouds, only three scenes from the eight available were usable. For study purposes, scene LC09_L2SP_192025_20220804_20230404_02_T1 was selected, see Table 2.

Table 2

Table 2. List of available and cloud-free (in bold) Landsat 8 and Landsat 9 scenes in July–August 2022.

Because the OLI-2 PAN band (Band 8) is not processed to surface reflectance in Collection 2 Level-2 products, we additionally retrieved the matching Level-1 Terrain Precision scene (L1TP) LC09_L1TP_192025_20220804_20230404_02_T1 to supply the 15 m PAN needed for pansharpening. In the fusion workflow, the spectral bands (Red and NIR) were taken from the L2SP surface-reflectance product, while the high-resolution PAN was taken from the corresponding L1TP product.

2.5 Tasking and downloading of PRISMA products

After registering an account and accepting the user license, we obtained PRISMA data directly from the ASI PRISMA portal (https://prisma.asi.it). We submitted tasking requests for the area of interest for the period July- August 2022. However, PRISMA imagery was only successfully acquired in August due to cloud cover or other objective reasons limiting ASi’s ability to obtain it. The standard Level-2D (L2D) products were made available for download from the same portal upon acceptance and acquisition. Level-2D is PRISMA’s geocoded, atmospherically corrected bottom-of-atmosphere (BOA) reflectance product generated by the ASI processor; it is delivered in Hierarchical Data Format 5 (HDF5) and is intended for direct application use. For this study, we used the Level-2D product acquired on 07 August 2022 (product ID: PRS_L2D_STD_20220807101606_20220807101610_0001). We used these L2D products for all analyses. The acquisition time for this scene was 10:16 GMT. The cloud cover percentage of the scene is 0.46%.

2.6 Selection of temporally coincident datasets

We centred the analysis on the field spectroscopy and UAV campaign of 08 August 2022 to ensure the closest possible temporal alignment across PRISMA and Landsat 9 sensors. For this period, we paired PRISMA L2D acquired on 07 August 2022 (1 day before field data), with Landsat 9 L2SP acquired on 04 August 2022 (4 days before, see Table 2). All satellite data were processed to surface reflectance (PRISMA L2D and Landsat Collection 2 Level-2 Science Product), co-registered to a standard grid, and cloud and shadow masked using provider Quality Assessment (QA) layers. We recorded the local overpass time and the solar/view geometry from the metadata for each acquisition to minimise directional effects. The 04–08 August window, where PRISMA, Landsat 9, and UAV data were nearly coincident, was used for the primary analyses. The 04 August Landsat scene was included as a sensitivity test within ±4 days of the UAV campaign and was analysed with its temporal offset explicitly reported (see Figure 4).

Figure 4

Figure 4. Multispectral (top row) and panchromatic images (bottom row) acquired by UAV sensors (MicaSense RedEdge-MX, SenseFly Duet T) and satellite systems (PRISMA, Landsat 9) during August 2022. The figure highlights differences in spatial resolution across the individual sensors in the “Amalie” study site.

2.7 Bandpass harmonization

We applied a bandpass harmonisation procedure to ensure comparability between the imaging spectrometer ASD FieldSpec 4, the hyperspectral satellite PRISMA, the UAV multispectral system (MicaSense RedEdge-MX) and the multispectral Landsat 9 measurements. In this study, the Landsat 9 OLI-2 Relative Spectral Response (RSR) functions were chosen as the reference because of their recent calibration, spectral stability (Mishra et al., 2016), and detailed documentation provided by the United States Geological Survey (NASA, 2025). For ASD FieldSpec 4, which provides 1 nm sampling, Landsat 9 synthetic band reflectance was computed as the weighted average of the high-resolution reflectance by the band-specific RSR across the wavelength interval where the RSR is non-zero. In practice, the RSR curve for the relevant band was sampled at each spectrometer wavelength, multiplied by the corresponding reflectance, and summed across the bandpass; division by the sum of RSR weights yielded the area-normalised band reflectance. For PRISMA, whose bands are spectrally broader and not perfectly aligned with Landsat 9, we handled partial band overlap explicitly. Each PRISMA band was represented by its central wavelength and full width at half maximum (FWHM). The intersection of this interval with the Landsat 9 bandpass was computed to obtain the fraction of the PRISMA band lying inside the Landsat 9 band. The Landsat 9 RSR was linearly interpolated and averaged within this overlapped segment to get the mean RSR overlap. The effective weight for the band was then defined as the product of the fractional overlap and the mean RSR. The “Landsat-like” reflectance for the band was calculated as the weighted mean of PRISMA band reflectances using these effective weights and normalised by their sum. This procedure was applied separately for the Red and NIR bands. PRISMA’s higher-resolution spectrum is effectively convolved to yield Landsat 9-equivalent Red and NIR reflectances that are directly comparable. For the UAV data, we harmonised the Red and NIR channels of the MicaSense RedEdge-MX. The Red channel is centred at 668 nm with a bandwidth of 14 nm (±7 nm) and the NIR channel at 842 nm with a bandwidth of 57 nm (±28.5 nm). For both channels, the Landsat 9 OLI-2 RSR was applied across defined bandwidths, and “Landsat-like” reflectances were derived following the same RSR-weighted averaging procedure as used for ASD FieldSpec 4 and PRISMA. This ensured that UAV Red and NIR measurements were directly comparable to the corresponding Landsat 9 bands. It is important to note that this harmonization procedure was a critical step to ensure a valid inter-sensor comparison, as it standardizes all primary NDVI calculations to the Landsat 9 OLI-2 bandpasses. We acknowledge that this approach means the main comparative analysis (presented in Sections 3.1–3.4) tests the pansharpening impact on a “Landsat-like” NDVI derived from PRISMA, rather than a native “narrowband” index. To investigate the implications of this methodological choice and test the performance on a true hyperspectral index, a parallel analysis was conducted. The methodology for this native “Narrowband” NDVI is detailed in Section 2.10.2, and the corresponding results are presented and discussed in Section 3.5.

2.8 Image co-registration, masking, and temporal pairing

All image products were orthorectified using the provider’s metadata and subsequently co-registered in software ENVI 5.6.3 (NV5 Geospatial Software, Colorado, United States). A set of 20 stable ground control points (GCPs) and tie points were manually selected, and geometric warping was applied to achieve sub-pixel alignment across sensors. The accuracy of the co-registration was evaluated visually at fixed features and confirmed by calculating residuals at the control points, resulting in a root mean square error (RMSE) of 0.364 pixels. Cloud and shadow were masked using the QA layers supplied with each product. For the August analysis window, ASD FieldSpec 4 spectra (08 August), MicaSense RedEdge-MX (08 August), PRISMA (07 August), and Landsat 9 (04 August) acquisitions were paired to minimise phenological and angular discrepancies.

2.9 Image pansharpening

To examine whether pansharpening affects spectral reflectance, five algorithms from different pansharpening categories were applied to the surface-reflectance products (Figure 5). Gram–Schmidt Nearest Neighbor, PC Nearest Neighbor, and NN Diffuse were executed in ENVI 5.6.3. Local Mean and Variance Matching (LMVM) bicubic and Bayesian fusion bicubic were executed using the Orfeo Toolbox (OTB) 9.1 plugin (CNES, Paris, France) in QGIS 3.40.10 “Bratislava” (QGIS Development Team). Gram–Schmidt and PC represent component-substitution approaches; NN Diffuse is a diffusion-based variant often grouped with component-substitution for operational use; LMVM is a multi-resolution method based on local mean–variance modelling; and Bayesian bicubic follows a Bayesian fusion framework with bicubic resampling. Parameter settings are detailed in Table 3. Optional fields left blank were handled with ENVI/OTB defaults.

Figure 5

Figure 5. Visual comparison of native hyperspectral PRISMA (top row) and multispectral Landsat 9 (bottom row) imagery with the results of several pansharpening techniques: Bayesian Bicubic, Gram–Schmidt, NN Diffuse, Local Mean and Variance Matching (LMVM), and Principal Component (PC). The native PRISMA scene was acquired on 07 August, whereas the Landsat 9 scene was acquired on 04 August. This figure illustrates the spatial and spectral differences between the original data and the outputs of the applied pansharpening algorithms.

Table 3

Table 3. Pansharpening algorithms, software versions, and key parameter settings applied to Landsat 9 and PRISMA.

2.10 Vegetation index computation and spatial sampling

2.10.1 “Landsat-like” NDVI calculation

Finally, NDVIs were calculated using the harmonised (Sections 2.6, 2.7) Red and NIR bands of five pansharpened PRISMA and Landsat 9 products at the panchromatic ground sampling distance, and native ASD FieldSpec 4, PRISMA, Landsat 9, and UAV datasets. Because the native pixel size of PRISMA and Landsat 9 is large relative to the ASD FieldSpec 4 and UAV, we extracted the NDVI only from the centre of one pixel for each of 14 plots. It was possible because larger spectrally homogeneous regions were selected as areas of interest during field measurements. In the case of UAV, the pixel size is much finer (ca. 24 cm); therefore, we calculated the mean surface reflectance in a 30 m diameter around the centre of every plot. For NDVI values extraction, the software ArcGIS Pro (ESRI, California, United States) was used.

2.10.2 Native narrowband NDVI calculation

To directly assess the pansharpening impact on a true hyperspectral index, a separate “narrowband NDVI” was calculated for the PRISMA sensor, distinct from the Landsat-harmonized “Landsat-like” NDVI used in the primary inter-sensor comparison. Based on established literature for optimal band selection (Haboudane et al., 2004), we selected native PRISMA bands corresponding to the chlorophyll absorption peak and the high-reflectance NIR plateau. This corresponded to Band 34 (centered at 669 nm) for Red and Band 47 (centered at 801 nm) for NIR. The same five pansharpening algorithms were applied to this specific pair of bands, and the resulting narrowband NDVI was extracted from the 14 plots for a parallel analysis, which was evaluated using the same statistical workflow described in Section 2.11.

2.11 Statistical analysis

We aimed to evaluate how selected pansharpening methods influence NDVI values compared to reference measurements. To do this, we tested four hypotheses: (a) The accuracy of tested pansharpening methods will differ significantly; (b) Standard pansharpening methods in mainstream software will perform better for multispectral than for hyperspectral pansharpening; (c) Depending on the method, NDVI derived from pansharpened data may deviate from NDVI derived from native PRISMA and Landsat 9 imagery; and (d) UAV-based multispectral orthomosaics (senseFly eBee X and MicaSense RedEdge-MX) can serve as a valuable reference for reflectance validation.

Statistical analyses were carried out on the August 2022 datasets, when ASD FieldSpec 4 (08 August), UAV-based MicaSense RedEdge-MX (08 August), PRISMA (07 August), and Landsat 9 (04 August) acquisitions were most nearly coincident. To evaluate the agreement between NDVI derived from different pansharpening methods and the reference NDVI, both accuracy metrics and statistical tests were applied. For each dataset, pixel-wise NDVI values were extracted and compared against the reference using mean absolute error (MAE), root mean square error (RMSE), bias, Pearson’s correlation coefficient (r), the slope, and intercept of an ordinary least-squares regression.

MAE and RMSE measure the average magnitude of error, with RMSE giving greater weight to larger deviations. Bias is the signed mean difference and indicates systematic over- or underestimation relative to ASD FieldSpec 4. Pearson’s r quantifies the linear correlation between pansharpened and reference NDVI. Slope and intercept are from an ordinary least-squares regression of pansharpened NDVI on ASD FieldSpec 4 NDVI and describe proportional and additive deviations, respectively. Lower MAE, RMSE, and Bias together with higher r and a slope close to 1 (and intercept near 0) indicate better agreement with the reference.

To assess the statistical significance of differences among NDVI values generated by various pansharpening methods, a rigorous two-stage analysis was conducted. For each sensor, we conducted two-sided paired Wilcoxon signed-rank tests twice: first, comparing NDVI from each pansharpening method against the sensor’s own surface-reflectance NDVI (no pansharpening), and second, comparing the same methods against the ASD FieldSpec 4 NDVI baseline resampled to the Red/NIR bands. We report only the Wilcoxon p-value for each method–baseline comparison (α = 0.05). All tests were run in R version 4.3.3 (R Core Team, Vienna, Austria) (R Core Team, 2025) using the base stats package (wilcox.test). An overview of the study workflow is shown in Figure 6.

Figure 6

Figure 6. Overview of the workflow for pansharpening assessment. UAV, satellite (Landsat 9, PRISMA), and field spectroscopy data collected in 2022 were spectrally harmonised and co-registered to the PAN imagery. Pansharpening methods were applied, and NDVI was extracted. Results were evaluated against ASD FieldSpec 4 and sensor surface reflectance baselines using standard agreement metrics (MAE, RMSE, bias, Pearson’s r, OLS slope and intercept. Statistical significance was tested with two-sided paired Wilcoxon signed-rank tests (α = 0.05), where p < 0.05 indicates a statistically significant difference between the methods.

3 Results

3.1 Accuracy of pansharpening methods

The significance of method-specific differences was first assessed using paired Wilcoxon signed-rank tests comparing NDVI from each pansharpened product and from the native surface-reflectance imagery against the ASD FieldSpec 4 reference. The paired Wilcoxon tests revealed clear method-specific patterns. For Landsat 9, none of the pansharpened NDVI products differed significantly from the ASD FieldSpec 4 reference (all p ≥ 0.103), indicating that any errors were unsystematic across plots. In contrast, for PRISMA, the PC–Nearest Neighbor method showed a statistically significant positive shift relative to the field reference (p = 0.017), whereas all other PRISMA methods were non-significant (p = 0.079–0.209), and the native PRISMA surface-reflectance product also did not differ from ASD FieldSpec 4 (p = 0.103). Collectively, these outcomes confirm that accuracy varies by pansharpening method—supporting Hypothesis (a)—and they underscore the relative robustness of the Bayesian, GS, LMVM, and NN Diffuse approaches compared with PC–Nearest Neighbor, which uniquely overestimated NDVI in PRISMA. Full p-values are reported in Table 4.

Table 4

Table 4. Statistical analysis of pansharpened NDVI for both Landsat 9 and PRISMA.

The statistical findings are reflected in Figure 7, which plots method-specific deviations from ASD FieldSpec 4 across plots. Landsat 9 methods cluster tightly around zero, consistent with the absence of significant differences, while PRISMA’s PC–Nearest Neighbor shows conspicuous positive excursions. This reinforces the conclusion that PC-based substitution can bias hyperspectral pansharpening, whereas local-contrast-preserving algorithms perform more reliably. Three plots (4 – Harvested crops, 11 – Residue, and 12 – Perennial forage) show markedly larger deviations from the FieldSpec 4 reference than the remaining sites. These anomalies are unlikely to reflect pansharpening effects alone and are instead attributed to two primary factors.

Figure 7

Figure 7. NDVI differences between pansharpened products and ASD FieldSpec 4 across 14 plots. Panels show land-cover classes by plot ID. Points represent method-specific departures (ΔNDVI) from field values, with the dashed line marking zero difference. Results from Landsat 9 are shown in red and results from PRISMA are shown in blue.

First, as shown in Figure 2, these specific land cover types exhibit considerable intra-class spectral diversity due to their heterogeneous mixture of soil and senescent vegetation. This plausibly results in a spectral-spatial scale mismatch, where the 10-point-averaged FieldSpec 4 measurement was not fully representative of the 900 m² satellite pixel’s average spectral condition. Second, we also acknowledge the impact of field conditions. Even with frequent calibration and clear skies, residual atmospheric effects (such as variable water vapor or aerosols) can subtly influence ground-based hyperspectral measurements. This introduces a potential, unavoidable discrepancy when comparing the field reference to the atmospherically-corrected satellite products. This interpretation is strongly supported by the evidence in Section 3.3 (Figure 8), which shows these same plots align well with their native (unsharpened) satellite data, confirming the anomaly originates from the FieldSpec-to-pixel comparison and is not an artifact of the pansharpening process itself.

Figure 8

Figure 8. NDVI differences between pansharpened products and each sensor’s native surface-reflectance NDVI across 14 plots. Panels show land-cover classes by plot IDPoints represent method-specific departures (ΔNDVI) from native values, with the dashed line marking zero difference. Results from Landsat 9 are shown in red and results from PRISMA are shown in blue.

3.2 Sensor-specific robustness of off-the-shelf methods

Agreement statistics showed that for Landsat 9, errors were modest across all pansharpening methods (MAE = 0.051–0.078) and correlations with ASD FieldSpec 4 were consistently strong (r > 0.90). In contrast, PRISMA exhibited greater variability among methods. NN Diffuse and LMVM performed best (MAE = 0.049–0.050), even slightly surpassing the native product (MAE = 0.053), while PC–Nearest Neighbor produced substantially larger errors (MAE = 0.093) and a biased regression slope against the field reference. Full agreement metrics are reported in Table 5.

Table 5

Table 5. Statistical analysis of pansharpened NDVI for both Landsat 9 and PRISMA.

This evidence supports Hypothesis (b). Standard pansharpening methods implemented in mainstream software prove more stable for multispectral fusion, as demonstrated by the consistency of Landsat 9 across land-cover classes. By contrast, hyperspectral fusion is more method-sensitive, with PC–Nearest Neighbor producing systematic biases for PRISMA, while LMVM and NN Diffuse yield more reliable outcomes.

3.3 NDVI from pansharpened vs. native products

The Wilcoxon signed-rank tests showed that for Landsat 9, none of the pansharpening methods produced NDVI values significantly different from the native product (all p > 0.05). The PC–Nearest Neighbor method approached the significance threshold (p = 0.060), while all other methods showed higher p-values (p = 0.223–0.860). For PRISMA, most methods likewise did not differ significantly (p = 0.183–0.450). However, the PC–Nearest Neighbor approach produced a statistically significant difference from the native product (p = 0.023). It is important to note that with a small sample size (n = 14), the statistical power of this analysis is limited. Therefore, p-values approaching the significance threshold do not confirm the absence of an effect, but rather that a statistically significant difference was not detected within this dataset. The full set of results is provided in Table 6.

Table 6

Table 6. Statistical comparison of five pansharpening methods against the sensor surface reflectance reference across 14 regions of interest.

Consistent with the non-parametric tests, plot-level departures of pansharpened NDVI from each sensor’s native product are generally small (Figure 8). All Landsat 9 methods remain close to zero; for PRISMA, deviations are still modest overall, though PC–Nearest Neighbor shows more frequent positive shifts. These outcomes suggest that pansharpening generally preserves NDVI consistency with native products at the plot scale, thereby only partially supporting Hypothesis (c). While most algorithms introduce no significant deviations, PC–Nearest Neighbor in PRISMA again shows a tendency toward bias. This is particularly evident for plot 4 (Harvested crops), plot 11 (Residue), and plot 12 (Perennial forage), which showed conspicuous deviations when compared to FieldSpec 4 (Figure 7) but remain better aligned with the native products in Figure 8. Their behaviour confirms that the earlier anomalies originate from uncertainties in the field reference rather than from pansharpening, reinforcing that, aside from the systematic uplift of PC–Nearest Neighbor in PRISMA, NDVI is largely preserved by the tested algorithms.

3.4 Utility of UAV orthomosaics as a reference

The comparison revealed that UAV NDVI was strongly correlated with ASD FieldSpec 4 (r = 0.978), yet its absolute errors were higher (MAE = 0.056; RMSE = 0.074) than those of PRISMA (MAE = 0.053) and Landsat 9 (MAE = 0.051). Although the UAV orthomosaic effectively captured spatial variability in NDVI, its agreement with field reference was weaker than that of the satellite products. This suggests that UAV data are valuable for fine-scale spatial analyses but less reliable as a primary quantitative benchmark compared with ASD FieldSpec 4 or satellite imagery. Detailed metrics are provided in Table 7.

Table 7

Table 7. Baseline agreement of PRISMA, UAV, and Landsat 9 surface reflectance NDVI with ASD FieldSpec 4.

3.5 Assessment of native hyperspectral narrowband NDVI (PRISMA)

To confirm that our main findings were not an artifact of the Landsat-band harmonization of PRISMA data, a parallel analysis was run using a native narrowband NDVI derived from PRISMA bands 34 (669 nm) and 47 (801 nm). The results of this hyperspectral-specific analysis, comparing it directly to the “Landsat-like” results, are presented in Figures 9, 10, and Tables 8–10. The findings were highly consistent with the main analysis. When compared against the ASD FieldSpec 4 reference (Figure 9; Tables 8, 9), the “Narrowband” NDVI showed the same performance pattern as the “Landsat-like” NDVI. Based on error metrics (Table 8), the PC–Nearest Neighbor method again produced the largest deviations (MAE = 0.089, RMSE = 0.129). In contrast, LMVM bicubic and the sensor surface reflectance (native) remained the most accurate, yielding the lowest errors (MAE = 0.044 for both). This pattern was strongly confirmed by the statistical analysis (Table 9), which showed that PC–Nearest Neighbor was the only method to produce NDVI values statistically different from the ASD FieldSpec 4 reference (p = 0.021). Conversely, all other methods, including LMVM bicubic (p = 0.090) and the native sensor surface reflectance (p = 0.124), were statistically indistinguishable from the ground-truth reference. Similarly, when comparing the pansharpened “Narrowband” NDVI against its native (unsharpened) counterpart (Figure 10; Table 10), the deviations were minimal for most methods. This mirrored the stability seen in the “Landsat-like” analysis. Once again, the PC–Nearest Neighbor method was the only approach to produce a statistically significant difference from the native unsharpened product (p = 0.023), while all other methods showed no significant difference (p = 0.183–0.450).

Figure 9

Figure 9. NDVI differences between pansharpened PRISMA “Landsat-like” and PRISMA “Narrowband” products and ASD FieldSpec 4 across 14 plots. Panels show land-cover classes by plot ID. Points represent method-specific departures (ΔNDVI) from field values, with the dashed line marking zero difference. Results from PRISMA “Narrowband” are shown in cyan and results from PRISMA “Landsat-like” are shown in blue.

Figure 10

Figure 10. NDVI differences between pansharpened PRISMA “Landsat-like” and PRISMA “Narrowband” products and sensor’s native surface-reflectance across 14 plots. Panels show land-cover classes by plot ID. Points represent method-specific departures (ΔNDVI) from field values, with the dashed line marking zero difference. Results from PRISMA “Narrowband” are shown in cyan and results from PRISMA “Landsat-like” are shown in blue.

Table 8

Table 8. Statistical analysis of pansharpened NDVI for PRISMA narrowband NDVI.

Table 9

Table 9. Statistical analysis of pansharpened NDVI for both PRISMA narrowband and PRISMA “Landsat-like.”

Table 10

Table 10. Statistical comparison of five pansharpening methods against PRISMA “Landsat-like” and PRISMA narrowband sensor surface reflectance reference across 14 regions of interest.

This parallel analysis confirms that the choice of algorithm is the dominant factor influencing pansharpening accuracy for PRISMA. The PC–Nearest Neighbor method was consistently the poorest performer, showing statistically significant deviations from both the ground-truth reference (Table 9) and the native sensor reflectance (Table 10). Conversely, robust methods like LMVM bicubic produced results statistically indistinguishable from the reference. This clear performance gap holds true regardless of whether a “Landsat-like” or “Narrowband” NDVI definition is used.

4 Discussion

The study was guided by four main hypotheses. The first hypothesis, that the accuracy of the tested pansharpening methods would differ significantly, was supported by the results. For both PRISMA and Landsat 9, the Principal Component (PC) Nearest Neighbor method consistently produced the highest errors (MAE of 0.093 for PRISMA, 0.078 for Landsat 9) and, in the case of PRISMA, was the only algorithm to show a statistically significant positive bias against the ASD FieldSpec 4 reference (p = 0.017). In contrast, algorithms like NN Diffuse and LMVM bicubic performed robustly, yielding the lowest errors and closely matching the native surface reflectance products. This confirms that not all accessible pansharpening techniques are equal in their ability to preserve spectral information for quantitative analysis (Loncan et al., 2015).

The second hypothesis posited that these standard pansharpening methods would perform better for multispectral than for hyperspectral pansharpening. PC-based approaches, although widely available in mainstream software, proved suboptimal for hyperspectral vegetation monitoring, showing the largest errors and a significant positive bias for PRISMA that did not occur with Landsat 9. In contrast, NN Diffuse and LMVM consistently produced NDVI values closely matching both native imagery and in situ reference measurements, confirming earlier findings that local contrast-preserving techniques outperform global transformations in vegetation applications. The stronger degradation observed for PRISMA underscores that the higher spectral complexity of hyperspectral data can amplify distortions introduced by less suitable methods, such as PC–Nearest Neighbor, which were originally designed with multispectral imagery in mind (Garzelli et al., 2023).

The third hypothesis, that NDVI from pansharpened data may deviate from NDVI calculated from native imagery, was also supported, though the deviations were generally small. When pansharpened products were compared against their own unsharpened, native-resolution, no method produced a statistically significant difference for either sensor. This is a critical practical finding, as it suggests that for most applications, the spatial enhancement from pansharpening does not fundamentally alter the NDVI values relative to what the sensor originally measured, thereby preserving the data’s utility for time-series analysis and change detection (Johnson, 2014).

Finally, the fourth hypothesis that UAV-based multispectral orthomosaics could serve as a valuable reference for reflectance validation was not supported by our data. The UAV-derived NDVI exhibited a slightly higher MAE (0.056) and RMSE (0.074) against the ASD FieldSpec 4 reference compared to the native satellite products from PRISMA and Landsat 9. This outcome may partly reflect limitations in the radiometric calibration of the RedEdge-MX sensor. Despite standard calibration using reflectance panels, UAV systems are more prone to residual uncertainties due to sensor drift and variations in illumination geometry. The very low flight altitude has likely enhanced bidirectional reflectance distribution function (BRDF) effects, as reflectance anisotropies are more pronounced in narrow-view imaging. Even under clear conditions, such geometric effects can cause deviations that complicate cross-sensor comparison (Jiang et al., 2022; Chakhvashvili et al., 2024). Achieving absolute radiometric accuracy with UAV data, therefore, remains challenging compared to well-calibrated, atmospherically corrected satellite products, which benefit from onboard calibration and standardised correction protocols (Slade et al., 2023).

A key methodological choice in this study was the harmonization of all datasets to the Landsat 9 bandpasses to ensure a fair inter-sensor comparison. This step, however, raised a valid question as to whether the findings would hold for a true hyperspectral “Narrowband” index, as our main analysis focused on a “Landsat-like” NDVI. To address this, we conducted a parallel analysis on a native PRISMA “Narrowband” NDVI with results presented in Section 3.5. This analysis confirmed that the performance of the pansharpening algorithms was highly consistent for both the “Landsat-like” NDVI and the true “Narrowband” (hyperspectral) NDVI. In both scenarios, the LMVM and NN Diffuse methods were among the most robust, while the PC–Nearest Neighbor method consistently produced the largest errors. This demonstrates that our findings are not an artifact of the band harmonization but are a robust reflection of how these algorithms handle PRISMA’s hyperspectral data, adding significant weight to our conclusions.

The findings of this study are constrained by certain limitations. The assessment was restricted to a single site, one phenological stage, and one vegetation index (NDVI) using a limited set of 14 spectrally homogeneous plots. Furthermore, the small sample size (n = 14), while appropriate for the use of non-parametric tests, must be acknowledged as a limitation due to its low statistical power. This increases the probability of a Type II error, meaning that some real, albeit small, differences between methods may not have been detected as statistically significant. Consequently, findings with p-values just above the 0.05 threshold should be interpreted with caution, as they do not definitively prove the absence of an effect but rather indicate that an effect was not detected with this sample. While NDVI is the most commonly used and robust vegetation metric, other indices sensitive to specific pigment or structural features may be more affected by spectral–spatial transformations introduced by pansharpening. Similarly, the findings may not generalise to heterogeneous landscapes, such as forests or urban areas, or to periods of rapid vegetation change, as the analysis relied on a single temporal snapshot in August 2022, with minor date differences between the sensor acquisitions (August 04, 07, and 08). While this window was intentionally chosen to minimise phenological changes, rapid vegetation dynamics could still influence comparisons.

Our study provides one of the first empirical assessments of how pansharpening influences vegetation indices derived from hyperspectral imagery. Using five widely available algorithms on PRISMA hyperspectral and Landsat 9 multispectral data, validated against FieldSpec 4 spectra and UAV measurements across 14 agricultural plots, we found that NDVI values from pansharpened products deviated only marginally from those of native imagery, with most differences statistically insignificant. This demonstrates that, when appropriate preprocessing is applied, pansharpening can generally be integrated into vegetation monitoring workflows without materially distorting NDVI.

Algorithm choice, however, proved critical. NN Diffuse and LMVM (bicubic) preserved NDVI fidelity most effectively, closely matching both native products and field spectra, while the Principal Component (PC) Nearest Neighbor method produced the largest errors and a significant positive bias for PRISMA, highlighting its unsuitability for hyperspectral applications. The stronger impact observed for PRISMA compared to Landsat 9 suggests that the higher spectral complexity of hyperspectral data amplifies distortions introduced by legacy component-substitution methods. Comparison with UAV RedEdge-MX imagery further confirmed that UAV-based NDVI, while offering superior spatial detail for field-scale diagnostics, requires meticulous radiometric processing to serve as an absolute reflectance benchmark. In contrast, the native and pansharpened satellite products exhibited closer agreement with field spectroscopy, reflecting their standardized calibration and atmospheric correction. Together, these findings lead to four key conclusions: (i) method choice matters, with robust local contrast-preserving approaches (NN Diffuse, LMVM) outperforming global component-substitution; (ii) pansharpening can be applied to both multispectral and hyperspectral data without materially altering NDVI, provided preprocessing is rigorous (bandpass harmonisation, co-registration, cloud/shadow masking); (iii) UAV NDVI should be used cautiously as a quantitative reference; and (iv) PC-based methods, while common in mainstream software, are poorly suited for hyperspectral NDVI retrieval. The study is, however, limited to one temperate agricultural site, a single growing season, and one vegetation index. Future work should extend testing across additional indices, landscapes, and phenological stages, and benchmark accessible methods against advanced Bayesian and deep-learning fusion schemes to evaluate practical accuracy–accessibility trade-offs. With hyperspectral missions such as PRISMA, EnMAP, and CHIME expanding data availability, these results provide a methodological foundation for the reliable integration of pansharpening into vegetation monitoring at finer spatial resolutions.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethics statement

Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

Author contributions

JP: Writing – original draft, Writing – review and editing, Project administration, Investigation, Data curation, Formal Analysis, Methodology, Validation, Visualization. JK: Writing – review and editing, Formal Analysis, Validation. DM: Methodology, Writing – review and editing, Conceptualization. JR: Visualization, Data curation, Writing – review and editing. TK: Supervision, Methodology, Writing – review and editing, Visualization, Formal Analysis, Writing – original draft, Conceptualization, Validation, Project administration, Data curation.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This study was supported by the Internal Grant Agency (IGA) of the Faculty of Environmental Sciences, Czech University of Life Sciences Prague, under Grant numbers 2024B0013 and 2022B0025.

Acknowledgements

We want to thank Petr Klápště and Ondřej Lagner for their help with UAV data acquisition. This article is supported by the EU COST (European Cooperation in Science and Technology) Action CA22136 “Pan-European Network of Green Deal Agriculture and Forestry Earth Observation Science” (PANGEOS). This study is conducted within the framework of the openskylab.space initiative.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The tool was used exclusively to improve the clarity and academic style of the text and to identify non-scientific errors, including typographical mistakes, grammatical inconsistencies, discrepancies in figure and table numbering, and cross-referencing issues. The AI system was not used to generate scientific ideas, perform data analysis, or interpret results. All study design, data processing, analyses, and conclusions are entirely the work of the authors, who reviewed and verified all manuscript content.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Alparone, L., Arienzo, A., and Garzelli, A. (2024). Spatial resolution enhancement of satellite hyperspectral data via nested hypersharpening with Sentinel-2 multispectral data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 17, 10956–10966. doi:10.1109/JSTARS.2024.3406762

CrossRef Full Text | Google Scholar

Amro, I., Mateos, J., Vega, M., Molina, R., and Katsaggelos, A. K. (2011). A survey of classical methods and new trends in pansharpening of multispectral images. EURASIP J. Adv. Signal Process 2011, 79. doi:10.1186/1687-6180-2011-79

CrossRef Full Text | Google Scholar

Bannari, A., Morin, D., Bonn, F., and Huete, A. R. (1995). A review of vegetation indices. Remote Sens. Rev. 13, 95–120. doi:10.1080/02757259509532298

CrossRef Full Text | Google Scholar

Barsi, J. A., Donley, E., Goldman, M., Kampe, T., Markham, B. L., McAndrew, B., et al. (2024). Prelaunch spectral characterization of the operational land Imager-2. Remote Sens. (Basel) 16, 981. doi:10.3390/rs16060981

CrossRef Full Text | Google Scholar

Cánovas-García, F., Pesántez-Cobos, P., and Alonso-Sarría, F. (2020). fusionImage: an R package for pan-sharpening images in open source software. Trans. GIS 24, 1185–1207. doi:10.1111/tgis.12676

CrossRef Full Text | Google Scholar

Chakhvashvili, E., Machwitz, M., Antala, M., Rozenstein, O., Prikaziuk, E., Schlerf, M., et al. (2024). Crop stress detection from UAVs: best practices and lessons learned for exploiting sensor synergies. Precis. Agric. 25, 2614–2642. doi:10.1007/s11119-024-10168-3

CrossRef Full Text | Google Scholar

Chen, L., Letu, H., Fan, M., Shang, H., Tao, J., Wu, L., et al. (2022). An introduction to the Chinese high-resolution Earth observation system: Gaofen-1∼7 civilian satellites. J. Remote Sens. (United States) 2022, 2022. doi:10.34133/2022/9769536

CrossRef Full Text | Google Scholar

Ciotola, M., Guarino, G., Vivone, G., Poggi, G., Chanussot, J., Plaza, A., et al. (2025). Hyperspectral pansharpening: critical review, tools, and future perspectives. IEEE Geosci. Remote Sens. Mag. 13, 311–338. doi:10.1109/MGRS.2024.3509139

CrossRef Full Text | Google Scholar

Cogliati, S., Sarti, F., Chiarantini, L., Cosi, M., Lorusso, R., Lopinto, E., et al. (2021). The PRISMA imaging spectroscopy mission: overview and first performance analysis. Remote Sens. Environ. 262, 112499. doi:10.1016/j.rse.2021.112499

CrossRef Full Text | Google Scholar

CWSL (2025). Smart landscape. Available online at: https://cvpk.czu.cz/en/r-17791-smart-landscape (Accessed September 23, 2025).

Google Scholar

Dadrass Javan, F., Samadzadegan, F., Mehravar, S., Toosi, A., Khatami, R., and Stein, A. (2021). A review of image fusion techniques for pan-sharpening of high-resolution satellite imagery. ISPRS J. Photogrammetry Remote Sens. 171, 101–117. doi:10.1016/j.isprsjprs.2020.11.001

CrossRef Full Text | Google Scholar

Ducay, R., and Messinger, D. W. (2020). “Radiometric assessment of four pan-sharpening algorithms as applied to hyperspectral imagery,” in Algorithms, technologies, and applications for multispectral and hyperspectral imagery XXVI. Editors M. Velez-Reyes, and D. W. Messinger (SPIE). doi:10.1117/12.25587411139203

CrossRef Full Text | Google Scholar

Ducay, R., and Messinger, D. (2022). “Hyperspectral-multispectral image fusion using nndiffuse: performance assessment using A pixel classification task,” in 2022 12th workshop on hyperspectral imaging and signal processing: evolution in remote sensing (WHISPERS), 1–5. doi:10.1109/WHISPERS56178.2022.9955122

CrossRef Full Text | Google Scholar

Erhard, M., Sornig, M., Fischer, S., Sang, B., Heider, B., Betz, M., et al. (2017). The hyperspectral instrument onboard ENMAP: overview and current status. SPIE-Intl Soc. Opt. Eng. 191, 191. doi:10.1117/12.2296178

CrossRef Full Text | Google Scholar

Gao, B. C., Montes, M. J., Davis, C. O., and Goetz, A. F. H. (2009). Atmospheric correction algorithms for hyperspectral remote sensing data of land and ocean. Remote Sens. Environ. 113, S17–S24. doi:10.1016/j.rse.2007.12.015

CrossRef Full Text | Google Scholar

Gao, S., Zhong, R., Yan, K., Ma, X., Chen, X., Pu, J., et al. (2023). Evaluating the saturation effect of vegetation indices in forests using 3D radiative transfer simulations and satellite observations. Remote Sens. Environ. 295, 113665. doi:10.1016/j.rse.2023.113665

CrossRef Full Text | Google Scholar

Garzelli, A., Zoppetti, C., Arienzo, A., and Alparone, L. (2023). “Spatial resolution enhancement of prisma hyperspectral data via nested hypersharpening with Sentinel-2 multispectral data,” in IGARSS 2023 - 2023 IEEE international geoscience and remote sensing symposium, 5997–6000. doi:10.1109/IGARSS52108.2023.10281961

CrossRef Full Text | Google Scholar

Green, R. O., Mahowald, N., Ung, C., Thompson, D. R., Bator, L., Bennet, M., et al. (2020). “The Earth surface mineral dust source investigation: an Earth science imaging spectroscopy mission,” in 2020 IEEE Aerospace Conference, 1–15. doi:10.1109/AERO47225.2020.9172731

CrossRef Full Text | Google Scholar

Guanter, L., Kaufmann, H., Segl, K., Foerster, S., Rogass, C., Chabrillat, S., et al. (2015). The EnMAP spaceborne imaging spectroscopy mission for earth observation. Remote Sens. (Basel) 7, 8830–8857. doi:10.3390/rs70708830

CrossRef Full Text | Google Scholar

Haboudane, D., Miller, J. R., Pattey, E., Zarco-Tejada, P. J., and Strachan, I. B. (2004). Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: modeling and validation in the context of precision agriculture. Remote Sens. Environ. 90 (3), 337–352. doi:10.1016/j.rse.2003.12.013

CrossRef Full Text | Google Scholar

Henrich, V., Götze, C., Jung, A., Sandow, C., Thürkow, D., and Cornelia, G. (2009). “Development of an online indices database: motivation, concept and implementation,” in 6th EARSeL imaging spectroscopy SIG workshop innovative tool for scientific and commercial environment applications Tel Aviv. Available online at: https://www.indexdatabase.de (Accessed June 20, 2022).

Google Scholar

Huang, S., Tang, L., Hupy, J. P., Wang, Y., and Shao, G. (2021). A commentary review on the use of normalized difference vegetation index (NDVI) in the era of popular remote sensing. J For Res (Harbin) 32, 1–6. doi:10.1007/s11676-020-01155-1

CrossRef Full Text | Google Scholar

Jawak, S. D., Wankhede, S. F., Luis, A. J., and Balakrishna, K. (2022). Impact of image-processing routines on mapping Glacier surface facies from Svalbard and the himalayas using pixel-based methods. Remote Sens. (Basel) 14, 1414. doi:10.3390/rs14061414

CrossRef Full Text | Google Scholar

Jiang, J., Johansen, K., Tu, Y. H., and McCabe, M. F. (2022). Multi-sensor and multi-platform consistency and interoperability between UAV, Planet CubeSat, Sentinel-2, and Landsat reflectance data. GIsci Remote Sens. 59, 936–958. doi:10.1080/15481603.2022.2083791

CrossRef Full Text | Google Scholar

Johnson, B. (2014). Effects of pansharpening on vegetation indices. ISPRS Int. J. Geoinf 3, 507–522. doi:10.3390/ijgi3020507

CrossRef Full Text | Google Scholar

Jovanović, D., Govedarica, M., Sabo, F., Važić, R., and Popovic, D. (2016). “Impact analysis of pansharpening Landsat ETM+, Landsat OLI, WorldView-2, and Ikonos images on vegetation indices,” in Proceedings of SPIE - The International Society for Optical Engineering, 968814. doi:10.1117/12.2241543

CrossRef Full Text | Google Scholar

Klouček, T., Moravec, D., Komárek, J., Lagner, O., and Štych, P. (2018). Selecting appropriate variables for detecting grassland to cropland changes using high resolution satellite data. PeerJ 6, e5487. doi:10.7717/peerj.5487

PubMed Abstract | CrossRef Full Text | Google Scholar

Klouček, T., Komárek, J., Surový, P., Hrach, K., Janata, P., and Vašíček, B. (2019). The use of UAV mounted sensors for precise detection of bark beetle infestation. Remote Sens. (Basel) 11, 1561. doi:10.3390/rs11131561

CrossRef Full Text | Google Scholar

Klouček, T., Modlinger, R., Zikmundová, M., Kycko, M., and Komárek, J. (2024). Early detection of bark beetle infestation using UAV-borne multispectral imagery: a case study on the spruce forest in the Czech Republic. Front. For. Glob. Change 7, 1215734. doi:10.3389/ffgc.2024.1215734

CrossRef Full Text | Google Scholar

Kremezi, M., Kristollari, V., Karathanassi, V., Topouzelis, K., Kolokoussis, P., Taggio, N., et al. (2021). Pansharpening PRISMA data for marine plastic litter detection using plastic indexes. IEEE Access 9, 61955–61971. doi:10.1109/ACCESS.2021.3073903

CrossRef Full Text | Google Scholar

Kuželková, M., Jačka, L., Kovář, M., Hradilek, V., and Máca, P. (2024). Tree trait-mediated differences in soil moisture regimes: a comparative study of beech, spruce, and larch in a drought-prone area of Central Europe. Eur J For Res 143, 319–332. doi:10.1007/s10342-023-01628-y

CrossRef Full Text | Google Scholar

Lebrun, M., Aguinaga, M., Zahid, Z., Šimek, P., Ouředníček, P., Klápště, P., et al. (2024). Manure-biochar blends effectively reduce nutrient leaching and increase water retention in a sandy, agricultural soil: insights from a field experiment. Soil Use Manag. 40, e13135. doi:10.1111/sum.13135

CrossRef Full Text | Google Scholar

Lin, C., Wu, C. C., Tsogt, K., Ouyang, Y. C., and Chang, C. I. (2015). Effects of atmospheric correction and pansharpening on LULC classification accuracy using WorldView-2 imagery. Inf. Process. Agric. 2, 25–36. doi:10.1016/j.inpa.2015.01.003

CrossRef Full Text | Google Scholar

Loncan, L., Almeida, L. B., Bioucas-Dias, J. M., Briottet, X., Chanussot, J., Dobigeon, N., et al. (2015). Hyperspectral pansharpening: a review. IEEE Geosci. Remote Sens. Mag. 3, 27–46. doi:10.1109/mgrs.2015.2440094

CrossRef Full Text | Google Scholar

Mahlayeye, M., Darvishzadeh, R., Jepkosgei, C., Mlawa, K. A., and Nelson, A. (2024). DESIS hyperspectral satellite data for cropping pattern classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 17, 17917–17929. doi:10.1109/JSTARS.2024.3457791

CrossRef Full Text | Google Scholar

Mallinis, G., Galidaki, G., and Gitas, I. (2014). A comparative analysis of EO-1 hyperion, quickbird and landsat TM imagery for fuel type mapping of a typical mediterranean landscape. Remote Sens. (Basel) 6, 1684–1704. doi:10.3390/rs6021684

CrossRef Full Text | Google Scholar

Maluleke, A., Feig, G., Brümmer, C., Rybchak, O., and Midgley, G. (2024). Evaluation of selected Sentinel-2 remotely sensed vegetation indices and MODIS GPP in representing productivity in semi-arid South African ecosystems. J. Geophys Res. Biogeosci 129, e2023JG007728. doi:10.1029/2023JG007728

CrossRef Full Text | Google Scholar

Matsunaga, T., Iwasaki, A., Tsuchida, S., Iwao, K., Tanii, J., Kashimura, O., et al. (2017). “Current status of hyperspectral imager suite (HISUI) onboard international space station (ISS),” in 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 443–446. doi:10.1109/IGARSS.2017.8126989

CrossRef Full Text | Google Scholar

Matta, E., De Luca, G., Pepe, M., Pogliotti, P., Bresciani, M., Genesio, L., et al. (2024). “Mapping glacial Lakes with prisma images,” in 2024 14th workshop on hyperspectral imaging and signal processing: evolution in remote sensing (WHISPERS), 1–6. doi:10.1109/WHISPERS65427.2024.10876525

CrossRef Full Text | Google Scholar

Mishra, N., Helder, D., Barsi, J., and Markham, B. (2016). Continuous calibration improvement in solar reflective bands: Landsat 5 through Landsat 8. Remote Sens. Environ. 185, 7–15. doi:10.1016/j.rse.2016.07.032

PubMed Abstract | CrossRef Full Text | Google Scholar

Mookambiga, A., and Gomathi, V. (2016). Comprehensive review on fusion techniques for spatial information enhancement in hyperspectral imagery. Multidimens. Syst. Signal Process 27, 863–889. doi:10.1007/s11045-016-0415-2

CrossRef Full Text | Google Scholar

Moravec, D., Komárek, J., Medina, S. L. C., and Molina, I. (2021). Effect of atmospheric corrections on NDVI: intercomparability of Landsat 8, Sentinel-2, and UAV sensors. Remote Sens. (Basel) 13, 3550. doi:10.3390/rs13183550

CrossRef Full Text | Google Scholar

Mushore, T. D., Mutanga, O., Odindi, J., Sadza, V., and Dube, T. (2022). Pansharpened landsat 8 thermal-infrared data for improved land surface temperature characterization in a heterogeneous urban landscape. Remote Sens. Appl. 26, 100728. doi:10.1016/j.rsase.2022.100728

CrossRef Full Text | Google Scholar

NASA (2025). OLI-2 relative spectral response. Available online at: https://landsat.gsfc.nasa.gov/satellites/landsat-9/landsat-9-instruments/oli-2-design/oli-2-relative-spectral-response/#rsr (Accessed September 17, 2025).

Google Scholar

Nieke, J., Despoisse, L., Gabriele, A., Weber, H., Strese, H., Ghasemi, N., et al. (2023). “The copernicus hyperspectral imaging mission for the environment (CHIME): an overview of its mission, system and planning status”. Sensors, systems, and next-generation satellites XXVII. Editors T. Kimura, S. R. Babu, and A. Hélière (Bellingham, United States: SPIE), 7, 7. doi:10.1117/12.2679977

CrossRef Full Text | Google Scholar

Pearlman, J., Carman, S., Segal, C., Jarecke, P., Clancy, P., and Browne, W. (2001). “Overview of the hyperion imaging spectrometer for the NASA EO-1 mission,” in IGARSS 2001. Scanning the Present and Resolving the Future. Proceedings. IEEE 2001 International Geoscience and Remote Sensing Symposium (Cat. No.01CH37217) 7, 3036–3038. doi:10.1109/IGARSS.2001.978246

CrossRef Full Text | Google Scholar

Perretta, M., Delogu, G., Funsten, C., Patriarca, A., Caputi, E., and Boccia, L. (2024). Testing the impact of pansharpening using PRISMA hyperspectral data: a case study classifying urban trees in Naples, Italy. Remote Sens. (Basel) 16, 3730. doi:10.3390/rs16193730

CrossRef Full Text | Google Scholar

Pettorelli, N., Vik, J. O., Mysterud, A., Gaillard, J. M., Tucker, C. J., and Stenseth, N. C. (2005). Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends Ecol. Evol. 20, 503–510. doi:10.1016/j.tree.2005.05.011

PubMed Abstract | CrossRef Full Text | Google Scholar

Pushparaj, J., and Hegde, A. V. (2017). Comparison of various pan-sharpening methods using Quickbird-2 and Landsat-8 imagery. Arabian J. Geosciences 10, 119. doi:10.1007/s12517-017-2878-3

CrossRef Full Text | Google Scholar

Qian, S. E. (2021). Hyperspectral satellites, evolution, and development history. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 14, 7032–7056. doi:10.1109/JSTARS.2021.3090256

CrossRef Full Text | Google Scholar

R Core Team (2025). The R project for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Available online at: https://www.r-project.org (Accessed September 23, 2025).

Google Scholar

Sári-Barnácz, F. E., Zalai, M., Milics, G., Tóthné Kun, M., Mészáros, J., Árvai, M., et al. (2024). Monitoring Helicoverpa armigera damage with PRISMA hyperspectral imagery: first experience in maize and comparison with Sentinel-2 imagery. Remote Sens. (Basel) 16, 3235. doi:10.3390/rs16173235

CrossRef Full Text | Google Scholar

Sarp, G. (2014). Spectral and spatial quality analysis of pan-sharpening algorithms: a case study in Istanbul. Eur. J. Remote Sens. 47, 19–28. doi:10.5721/EuJRS20144702

CrossRef Full Text | Google Scholar

Sech, G., Poggi, G., Ljubenović, M., Fiorucci, M., and Traviglia, A. (2024). “Pansharpening of PRISMA products for archaeological prospection,” in IGARSS 2024 - 2024 IEEE international geoscience and remote sensing symposium, 2234–2238. doi:10.1109/IGARSS53475.2024.10642261

CrossRef Full Text | Google Scholar

Seydi, S. T., and Hasanlou, M. (2018). Sensitivity analysis of pansharpening in hyperspectral change detection. Appl. Geomatics 10, 65–75. doi:10.1007/s12518-018-0206-6

CrossRef Full Text | Google Scholar

Shaik, R. U., Periasamy, S., and Zeng, W. (2023). Potential assessment of PRISMA hyperspectral imagery for remote sensing applications. Remote Sens. (Basel) 15, 1378. doi:10.3390/rs15051378

CrossRef Full Text | Google Scholar

Shi, Y., Gamba, P., Qu, J., and Li, Y. (2022). “Bilinear sparse target detection for asbestos identification in hyperspectral PRISMA data,” in IGARSS 2022 - 2022 IEEE international geoscience and remote sensing symposium, 1959–1962. doi:10.1109/IGARSS46834.2022.9884734

CrossRef Full Text | Google Scholar

Slade, G., Fawcett, D., Cunliffe, A. M., Brazier, R. E., Nyaupane, K., Mauritz, M., et al. (2023). Optical reflectance across spatial scales-an intercomparison of transect-based hyperspectral, drone, and satellite reflectance data for dry season rangeland. Drone Syst. Appl. 11, 1–20. doi:10.1139/dsa-2023-0003

CrossRef Full Text | Google Scholar

Snehmani, , Gore, A., Ganju, A., Kumar, S., Srivastava, P. K., and R P, H. R. (2017). A comparative analysis of pansharpening techniques on QuickBird and WorldView-3 images. Geocarto Int. 32, 1268–1284. doi:10.1080/10106049.2016.1206627

CrossRef Full Text | Google Scholar

Song, C., Woodcock, C. E., Seto, K. C., Lenney, M. P., and Macomber, S. A. (2001). Classification and change detection using Landsat TM data: when and how to correct atmospheric effects? Remote Sens. Environ. 75, 230–244. doi:10.1016/S0034-4257(00)00169-3

CrossRef Full Text | Google Scholar

Storch, T., Honold, H. P., Chabrillat, S., Habermeyer, M., Tucker, P., Brell, M., et al. (2023). The EnMAP imaging spectroscopy mission towards operations. Remote Sens. Environ. 294, 113632. doi:10.1016/j.rse.2023.113632

CrossRef Full Text | Google Scholar

Taggio, N., Aiello, A., Ceriola, G., Kremezi, M., Kristollari, V., Kolokoussis, P., et al. (2022). A combination of machine learning algorithms for marine plastic litter detection exploiting hyperspectral PRISMA data. Remote Sens. (Basel) 14, 3606. doi:10.3390/rs14153606

CrossRef Full Text | Google Scholar

Toosi, A., Javan, F. D., Samadzadegan, F., and Mehravar, S. (2020). Object-based spectral quality assessment of high-resolution pan-sharpened satellite imageries: new combined fusion strategy to increase the spectral quality. Arabian J. Geosciences 13, 499. doi:10.1007/s12517-020-05523-3

CrossRef Full Text | Google Scholar

Vangi, E., D’amico, G., Francini, S., Giannetti, F., Lasserre, B., Marchetti, M., et al. (2021). The new hyperspectral satellite prisma: imagery for forest types discrimination. Sensors Switz. 21, 1–19. doi:10.3390/s21041182

CrossRef Full Text | Google Scholar

Vavassori, A., Oxoli, D., Venuti, G., Brovelli, M. A., Ali Mohamed, A. B. E., Moazzam, A., et al. (2024). “PRISMA hyperspectral satellite imagery application to local climate zones mapping”. Göttingen, Germany: International Society for Photogrammetry and Remote Sensing, 643–648. doi:10.5194/isprs-archives-XLVIII-1-2024-643-2024

CrossRef Full Text | Google Scholar

Vermote, E., Justice, C., Claverie, M., and Franch, B. (2016). Preliminary analysis of the performance of the Landsat 8/OLI land surface reflectance product. Remote Sens. Environ. 185, 46–56. doi:10.1016/j.rse.2016.04.008

PubMed Abstract | CrossRef Full Text | Google Scholar

Vivone, G., Alparone, L., Chanussot, J., Dalla Mura, M., Garzelli, A., Licciardi, G. A., et al. (2015). A critical comparison among pansharpening algorithms. IEEE Trans. Geoscience Remote Sens. 53, 2565–2586. doi:10.1109/TGRS.2014.2361734

CrossRef Full Text | Google Scholar

Vivone, G., Dalla Mura, M., Garzelli, A., Restaino, R., Scarpa, G., Ulfarsson, M. O., et al. (2021). A new benchmark based on recent advances in multispectral pansharpening: revisiting pansharpening with classical and emerging pansharpening methods. IEEE Geosci. Remote Sens. Mag. 9, 53–81. doi:10.1109/MGRS.2020.3019315

CrossRef Full Text | Google Scholar

Xu, Q., Zhang, Y., and Li, B. (2014). Recent advances in pansharpening and key problems in applications. Int. J. Image Data Fusion 5, 175–195. doi:10.1080/19479832.2014.889227

CrossRef Full Text | Google Scholar

Xue, J., and Su, B. (2017). Significant remote sensing vegetation indices: a review of developments and applications. J. Sens. 2017, 2017. doi:10.1155/2017/1353691

CrossRef Full Text | Google Scholar

Yilmaz, V., Serifoglu Yilmaz, C., Güngör, O., and Shan, J. (2020). A genetic algorithm solution to the gram-schmidt image fusion. Int. J. Remote Sens. 41, 1458–1485. doi:10.1080/01431161.2019.1667553

CrossRef Full Text | Google Scholar

Yilmaz, C. S., Yilmaz, V., and Gungor, O. (2022). A theoretical and practical survey of image fusion methods for multispectral pansharpening. Inf. Fusion 79, 1–43. doi:10.1016/j.inffus.2021.10.001

CrossRef Full Text | Google Scholar

Zeng, Y., Hao, D., Huete, A., Dechant, B., Berry, J., Chen, J. M., et al. (2022). Optical vegetation indices for monitoring terrestrial ecosystems globally. Nat. Rev. Earth Environ. 3, 477–493. doi:10.1038/s43017-022-00298-5

CrossRef Full Text | Google Scholar

Zhu, Z., and Woodcock, C. E. (2014). Continuous change detection and classification of land cover using all available Landsat data. Remote Sens. Environ. 144, 152–171. doi:10.1016/j.rse.2014.01.011

CrossRef Full Text | Google Scholar

Ziaja, M., Kowaleczko, P., Nalepa, J., Kostrzewa, D., Latini, D., De Santis, D., et al. (2023). “Hyperspectral image pansharpening: the prisma case study,” in IGARSS 2023 - 2023 IEEE international geoscience and remote sensing symposium, 1633–1636. doi:10.1109/IGARSS52108.2023.10282612

CrossRef Full Text | Google Scholar