Harmonized Planet Scope and Sentinel-2 Data for Agricultural Monitoring in Gimje Paddy Area

Abstract

This study presents a prototype harmonized surface reflectance dataset generated by integrating PlanetScope (SuperDove) and Sentinel-2 (MSI) imagery following the NASA-ESA Harmonized Landsat-Sentinel-2 (HLS) framework. Atmospheric correction of PlanetScope imagery using the Land Surface Reflectance Code (LaSRC) algorithm effectively reduced scene-to-scene variability and improved aerosol retrieval, resulting in more stable surface reflectance. PlanetScope reflectance was then normalized to the HLS standard, yielding substantial improvements in radiometric consistency, with regression slopes approaching one and coefficients of determination (R²) increasing to 0.91-0.93 for the Red and Near-Infrared (NIR) bands. Time-series evaluation using Normalized Difference Vegetation Index (NDVI) demonstrated that the harmonized PlanetScope NDVI closely reproduced the phenological trajectory observed in HLS, confirming the potential of high-resolution, high-frequency imagery for agricultural monitoring. The findings provide an important technical foundation for applying similar harmonization strategies to the upcoming Korean agricultural satellite, the Compact Advanced Satellite 500-4 (CAS500-4). Further research is required to expand validation across seasons, crop types, and diverse observation conditions. In addition, the prototype results suggest that the integration of satellite data can substantially enhance temporal continuity during cloud-affected seasons. These outcomes highlight the need to expand validation across broader spatial domains and crop systems to further generalize the harmonization framework.

1. Introduction

Remote sensing has become a critical tool for agricultural monitoring, enabling non-destructive and scalable observation of crop growth, soil conditions, and environmental changes under increasing climate and food-security pressures (Food and Agriculture Organization of the United Nations, 2017; Hong et al., 2012). In the Republic of Korea, remote-sensing applications have evolved from early Landsat-based rice mapping to multi-sensor approaches utilizing Sentinel-1/2, commercial multispectral satellites, Unmanned Aerial Vehicles (UAVs), and ground spectrometers. These developments have supported crop-area estimation, growth monitoring, yield prediction, and agricultural environmental assessments across major cropping regions (Lee et al., 2017; Na et al., 2017a; 2017b; Ryu et al., 2021). Building on this progress, the Rural Development Administration (RDA) established the National Agricultural Satellite Center and is preparing to launch the Compact Advanced Satellite 500-4 (CAS500-4) in 2026 (Hong et al., 2024).

CAS500-4, a mid-sized Earth-observation satellite weighing approximately 500 kg. CAS500-4 is the first Korean satellite equipped with a red-edge band positioned between the Red and Near-Infrared (NIR) wavelengths for applications in forest and agricultural vegetation monitoring (Hong et al., 2024). With a one-day orbital revisit cycle and a tilt-imaging capability, CAS500-4 is expected to provide 5-m multispectral imagery covering the entire Korean Peninsula every three days. These characteristics are expected to significantly strengthen national capabilities for monitoring crop growth, crop damage, land-use change, and agricultural environmental variability.

However, relying solely on a single satellite sensor still presents considerable limitations for agricultural monitoring. Persistent cloud cover during the monsoon season, differences in spatial and spectral resolutions across sensors, variations in observation geometry, and radiometric inconsistencies between platforms reduce the accuracy and continuity of time-series analyses (Ahn et al., 2018a; 2018b; 2020). In multisensor agricultural monitoring, differences in spectral response functions, radiometric calibration, and sun–sensor geometry introduce systematic inconsistencies in surface reflectance, even when satellites observe the same target. Moreover, variations in spatial resolution among satellite sensors create mixed-pixel effects and inconsistencies in spatial detail, limiting direct comparison across time. Combined with cloud-induced temporal gaps, single-sensor time series often suffer from discontinuities and reduced reliability. As noted by Claverie et al. (2018) in the development of Harmonized Landsat–Sentinel-2 (HLS), these factors distort phenological trajectories and hinder long-term monitoring. Therefore, harmonized surface reflectance, which standardizes multisensor data to a common radiometric, spectral, and geometric framework, is essential for producing consistent and scientifically robust multisensor time-series datasets.

To address these limitations, multi-satellite data fusion and reflectance harmonization have become essential. Combining PlanetScope’s high temporal resolution with Sentinel-2’s radiometric stability can improve temporal density while minimizing inter-sensor differences. As shown in the HLS dataset, harmonized reflectance provides significant advantages for growth-stage detection, long-term time-series analysis, and agricultural productivity assessments.

The objective of this study is to develop a prototype Planet Scope–Sentinel-2 harmonized reflectance framework to apply harmonization strategies to the upcoming CAS500-4. This study aims to: (1) present a multisensor harmonization approach that addresses the limitations of single-sensor agricultural monitoring; (2) evaluate whether the harmonized PlanetScope–Sentinel-2 reflectance achieves radiometric and spectral consistency comparable to HLS; and (3) assess how the resulting harmonized dataset improves the continuity and reliability of agricultural time-series monitoring.

2. Materials and Methods

2.1. Study Area

The study area, Gimje, is one of the largest rice-producing regions in Korea, located in the central part of the Honam Plain in the southwest. According to the reference document, Gimje covers 545.87 km², of which approximately 70% is farmland, followed by forest (16%), residential and facility areas (7%), grassland (5%), and bare land (2%). The western area consists of wide alluvial plains, while the eastern boundary transitions into low hills connected to Moak Mountain.

Fig. 1. Study area for satellite image harmonization in the Gimje paddy rice region.

Farm Map analysis shows that several towns—such as Jinbong, Buryang, Gongdeok, and Gwanghwal—each contain more than 1,000 ha of paddy fields, confirming Gimje as a representative large-scale rice cultivation zone. The cropping system is mainly rice monoculture or double-cropping sequences (e.g., rice–barley, rice–forage crops), with transplanting typically occurring from late May to mid-June following a clear flooding and puddling period. These distinctive irrigation schedules and phenological patterns make Gimje an ideal test site for evaluating multi-sensor harmonized reflectance and time-series monitoring.

2.2. Satellite Data

In this study, PlanetScope (SuperDove) and Sentinel-2 (MSI) optical imagery were jointly used to generate harmonized surface reflectance for paddy rice fields in Gimje. PlanetScope is a constellation of more than 130 CubeSats and provides daily revisit capability and 3-m spatial resolution, making it suitable for capturing rapid agricultural dynamics such as flooding, transplanting, and early growth. The SuperDove sensor offers eight spectral bands ranging from Coastal Blue to NIR. In this study, PlanetScope imagery was atmospherically corrected using the Landsat Surface Reflectance Code (LaSRC) algorithm to derive Surface Reflectance. Although PlanetScope provides high spatial and temporal detail, sensor-to-sensor variability and long-term radiometric stability issues necessitate cross-sensor adjustment when generating harmonized reflectance products.

Sentinel-2, operated by the European Space Agency (ESA), consists of the S2A and S2B satellites and provides a 5-day revisit cycle, 10-m spatial resolution for the Blue–NIR bands, and 13 spectral bands with a wide 290-km swath. Its stable radiometric calibration makes it well-suited for long-term time-series monitoring. In this study, we used Level-2A Surface reflectance processed with ESA’s standard atmospheric correction algorithm, Sen2Cor. Although Sentinel-2 has coarser spatial resolution relative to PlanetScope, its spectral and radiometric stability make it an appropriate reference sensor for harmonization.

Because the two sensors exhibit complementary characteristics, harmonization was performed using the common spectral domain (RGB–NIR) following the concept of the HLS framework. The integration of PlanetScope’s high temporal–spatial resolution and Sentinel-2 radiometric stability allows both improved temporal density and reduced cross-sensor reflectance differences. This approach enhances time-series consistency and monitoring continuity, which is particularly advantageous for large contiguous paddy field regions such as Gimje.

Table 1. Summary of PlanetScope and Sentinel-2 sensor characteristics

2.3. Processing Chain

This study adopted a harmonization approach analogous to the HLS framework to generate consistent surface reflectance from PlanetScope and Sentinel-2 imagery over the Gimje paddy area (Claverie et al., 2018). The harmonization was conducted using the common spectral domain (Blue, Green, Red, and NIR) shared by both sensors to minimize differences arising from unequal spectral response functions.

PlanetScope SuperDove imagery, originally provided as Top-of-Atmosphere (TOA) reflectance, was atmospherically corrected using the LaSRC algorithm to obtain surface reflectance (Vermote et al., 2016). LaSRC applies radiative transfer modeling, aerosol retrieval, adjacency correction, and water vapor adjustment, thereby reducing atmospheric noise and enabling cross-sensor comparison (Vermote et al., 2016). This step is essential because PlanetScope exhibits sensor-to-sensor variability and lacks a standardized surface reflectance product.

Sentinel-2 imagery was obtained directly as Level-2A Surface reflectance, processed using ESA’s standard Sen2Cor algorithm. Since Sentinel-2 scenes were acquired near-nadir over the study area, Bidirectional Reflectance Distribution Function (BRDF) normalization was not applied; the viewing geometry effects were assumed minimal for the scope of this study. To ensure compatibility between the two sensors, reflectance was harmonized only for the RGB and NIR bands, following the approach used in Claverie et al. (2018) for the HLS product. Sentinel-2 served as the reference sensor due to its stable radiometric calibration. PlanetScope reflectance was adjusted through linear regression models based on temporally matched and cloud-free pixel pairs. Outliers caused by cloud shadows, saturated pixels, or flooded paddy fields were removed using percentile filtering and Quality Assurance (QA) masking.

Sentinel-2 images (10 m resolution) were resampled to 3 m using bicubic interpolation to match the PlanetScope grid structure, enabling pixel-wise comparison. Cloud and haze were removed using both Sen2Cor QA layers and the s2cloudless mask to ensure consistent temporal compositing. Only scenes with valid overlap and minimal cloud contamination were used for harmonization. After atmospheric correction, spectral alignment, and spatial normalization, harmonized reflectance was generated through cross-sensor adjustment and time-series smoothing. This process reduces differences caused by spectral response mismatch, radiometric bias, and temporal sampling disparities. The final harmonized dataset enhances temporal density (due to PlanetScope) while preserving radiometric stability (due to Sentinel-2), which is particularly advantageous for monitoring dynamic paddy rice phenology.

In this study, percentile filtering, QA-based cloud/shadow masking, and time-series smoothing were not applied, as we used only clear-sky, atmospherically corrected images with minimal cloud contamination. This decision was made to preserve the native radiometric characteristics of the two sensors and to avoid unnecessary preprocessing steps during harmonization. However, for operational applications and large-scale time-series analysis, incorporating percentile-based outlier detection, QA-driven masking, and temporal smoothing will be important. These procedures will be included in future work to enhance robustness under diverse atmospheric and seasonal conditions.

3. Results and Discussion

3.1. Atmospheric Correction Results

LaSRC was applied to PlanetScope SuperDove imagery to derive Surface reflectance. Compared to the original TOA reflectance, the Surface reflectance data showed a substantial reduction in atmospheric scattering and absorption effects across all spectral bands. In the Blue and Red regions, the characteristic overestimation caused by Rayleigh scattering and aerosols was significantly mitigated, while the NIR band exhibited a narrower reflectance range due to improved water vapor correction (Fig. 2).

Fig. 2. PlanetScope TOA and surface reflectance before and after LaSRC atmospheric correction.

Following atmospheric correction, scene-to-scene variability—commonly observed in high-resolution PlanetScope data due to sensor-to-sensor differences—was markedly reduced. This stabilization of spectral responses is critical for achieving a consistent radiometric relationship with the reference HLS dataset during harmonization.

In this study, the Accuracy–Precision–Uncertainty (APU) metric was used to quantitatively evaluate the performance of atmospheric correction. APU represents the accuracy, precision, and the combined uncertainty of the corrected reflectance, where lower values indicate more stable and reliable reflectance correction. According to the APU results for the Red and NIR bands shown in Fig. 3, the differences between LaSRC-corrected PlanetScope reflectance and the HLS reference were substantially reduced, resulting in noticeably lower APU values in both bands. This indicates that atmospheric correction effectively reduced scene-to-scene variability as well as radiometric noise and bias.

Fig. 3. Comparison of surface reflectance between LaSRC-corrected PlanetScope and HLS:scatter plot (left) and APU (right).

3.2. Aerosol Optical Depth Estimation Using LaSRC

The LaSRC algorithm retrieves Aerosol Optical Depth (AOD) as part of its atmospheric correction procedure, and AOD accuracy is a key determinant of the quality of surface reflectance. In this study, AOD values estimated from PlanetScope-LaSRC exhibited excellent agreement with independent AERONET measurements, showing a near-unity linear relationship (slope ≈ 1.03, intercept 0.002). This confirms that LaSRC provides stable aerosol retrieval even for 3-m high-resolution PlanetScope imagery (Fig. 4).

Fig. 4. Comparison of AOD derived from LaSRC-based PlanetScope imagery with AERONET measurements.

Accurate AOD retrieval serves as an essential input for correcting atmospheric scattering and absorption, thereby improving the stability of Surface reflectance. In particular, for spectral bands such as Red and NIR—which are highly influential for vegetation index estimation—precise AOD values reduce radiometric biases introduced during atmospheric correction and enhance the baseline accuracy required for subsequent harmonization with HLS. Overall, the strong performance of LaSRC in both AOD estimation and Surface reflectance correction indicates that PlanetScope data can be reliably standardized to match the radiometric characteristics of HLS.

3.3. Harmonized Reflectance Results

Harmonized surface reflectance was generated by adjusting PlanetScope Surface reflectance to match the radiometric characteristics of the NASA–ESA HLS dataset, which served as the reference product for all comparisons in this study. Prior to harmonization, PlanetScope reflectance showed noticeable discrepancies relative to HLS, including slopes less than unity, non-zero offsets, and band-dependent deviations that affected both the magnitude and temporal consistency of the reflectance signal. These differences were most pronounced in the Red and NIR bands, where variations in sensor calibration, spectral response functions, and atmospheric effects contributed to scene-to-scene inconsistencies.

After applying the harmonization procedure, the corrected PlanetScope reflectance exhibited a substantially improved radiometric relationship with HLS. The slope and intercept of the regression between PlanetScope and HLS converged toward 1 and 0, respectively, indicating that PlanetScope reflectance values were successfully aligned with the HLS standard. The coefficient of determination increased to approximately 0.91 in the Red band and 0.93 in the NIR band, demonstrating a high level of radiometric consistency. The reduction of systematic bias and noise suggests that the harmonization effectively compensated for the spectral response differences between the two sensors, producing reflectance values that are directly comparable to HLS (Fig. 5).

Fig. 5. Comparison of PlanetScope surface reflectance before and after harmonization with HLS data.

The improvement was further evident in the time-series analysis. Prior to harmonization, PlanetScope Normalized Difference Vegetation Index (NDVI) exhibited noticeable temporal discontinuities owing to residual atmospheric effects and sensor-specific variability. In contrast, the harmonized PlanetScope NDVI closely reproduced the phenological trajectory observed in the HLS NDVI series, including the distinct transitions associated with flooding, transplanting, early growth, and the peak growth period of rice paddies. The harmonized series successfully captured the typical double-logistic growth curve, and the temporal patterns aligned closely with HLS despite differences in spatial resolution and revisit frequency (Fig. 6).

Fig. 6. NDVI time-series comparison of PlanetScope before and after harmonization with HLS.

These results demonstrate that the harmonized PlanetScope reflectance dataset inherits the strengths of both sensors: the radiometric stability and standardization of HLS, and the high temporal density and spatial resolution of PlanetScope. The improved agreement with HLS confirms that the harmonization method is robust and that PlanetScope can be effectively integrated into HLS-like time-series analyses for agricultural monitoring. The ability to maintain HLS-level radiometric performance while achieving daily observation frequency highlights the potential of the harmonized dataset for detecting rapid phenological changes in paddy rice systems, particularly in regions such as Gimje, where growth dynamics are highly time-sensitive.

4. Conclusions

This study demonstrated a prototype harmonized surface reflectance dataset by integrating PlanetScope (SuperDove) and Sentinel-2 (MSI) imagery following the NASA–ESA HLS framework. Atmospheric correction using LaSRC effectively reduced scene-to-scene variability in PlanetScope data, and accurate AOD retrieval contributed to improved Surface stability. The normalization to the HLS standard produced strong radiometric and spectral consistency, and the harmonized PlanetScope NDVI successfully reproduced the phenological trajectory observed in HLS. These results highlight the complementary benefits of PlanetScope’s high spatial–temporal resolution and the radiometric stability of HLS.

However, the study is limited by its focus on a single agricultural area and specific observation period, indicating the need for broader validation across diverse crops, seasons, and environmental conditions. In addition, the absence of BRDF normalization and Spectral Band Adjustment Factors (SBAF) may leave residual inter-sensor differences that require further investigation. Future research should expand spatiotemporal validation, incorporate BRDF and spectral adjustment procedures, and evaluate the harmonization framework under variable atmospheric and illumination conditions.

Importantly, the harmonization procedure developed in this study provides a technical foundation for application to the upcoming Korean agricultural satellite, CAS500-4, thereby enabling the establishment of an operational multisensor fusion system tailored for national agricultural monitoring. Moreover, harmonized high-resolution time-series data can support integration with crop growth models, disaster monitoring systems, and policy decision-making frameworks, thereby strengthening quantitative agricultural observation capabilities. In the longer term, standardizing harmonization algorithms across CAS500-4, commercial satellites, and public missions will be essential for expanding this approach into a nationwide agricultural and environmental monitoring system.

Author Contributions

Conceptualization: Ahn HY, Ryu JH; Data curation: Kim JH, Kim JH; Methodology, Formal analysis, Validation: Ahn HY, Jeon DS; Project administration: Ahn HY; Funding acquisition, Supervision: Lee KD.; Writing–originaldraft: Ahn HY; Writing–review & editing: All authors.

Conflicts of Interest

No potential conflict of interest relevant to this article was reported.

Funding

This research was funded by the Rural Development Administration (Grant No. PJ017319).

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.