Frequent surface reflectance measurements at fine spatial resolutions are crucial for land surface monitoring. The NASA Harmonized Landsat and Sentinel-2 (HLS) project addresses this need by combining data from Landsat and Sentinel-2 missions to provide 30-m surface reflectance products with a global average revisit frequency of 2-3 days. This study evaluates the consistency of the HLS Version 2 (V2.0) product against the established Moderate Resolution Imaging Spectroradiometer (MODIS) Nadir Bidirectional Reflectance Distribution Function-Adjusted Reflectance (NBAR) dataset. We found a high degree of consistency between HLS and MODIS NBAR, with strong positive correlations across all six common spectral bands (R2 values: 0.939–0.986). Globally, mean absolute differences (MADs) were generally low, typically under 0.03 for visible bands and below 0.05 for NIR and SWIR bands. However, we observed band-dependent discrepancies, with higher discrepancies in the blue and SWIR1 bands. Spatially, substantial differences were found in polar regions, largely due to isolated snow, ice, and water which are fractional and hard to detect within the coarser MODIS pixels but are filtered out in the HLS data. This finding suggests a potential explanation for past disagreements in vegetation trend analyses involving the use of satellite data of different spatial resolutions, as coarser-resolution datasets may mischaracterize greening trends in areas with partial snow and ice cover. These insights are vital for helping users interpret HLS data more effectively and for refining multi-sensor data fusion methods. Our research also provides a valuable baseline for ongoing quality assessment of the HLS product for applications such as vegetation monitoring, land management, and disaster response.
Annual land cover monitoring using medium-resolution satellite imagery represents a transition from periodic mapping campaigns to continuous observation systems. This transition introduces accuracy assessment challenges not encountered in one-time mapping efforts, particularly when the initial monitoring period is extended forward in time. These challenges are not addressed in existing guidance, which remains largely framed around the evaluation of maps of a single date or maps showing changes between two dates. We present a set of methodological adaptations to adjust to this evolving paradigm, using as a case study the U.S. Geological Survey (USGS) Land Change Monitoring, Assessment, and Projection (LCMAP) initiative, which generated annual land cover and land cover change products across the historical Landsat archive. Assessing the accuracy of these products was an essential component of LCMAP and resulted in an extensive reference sample dataset that may be used for accuracy estimation, area estimation, or support in the development of new algorithms or models. We review the LCMAP methodology and highlight challenges encountered and lessons learned that are transferable to other long-term, temporally dense monitoring programs. This framework demonstrates that statistically rigorous assessment can scale to operational annual land cover production, while the experiences of the project clarify where existing good practice guidance should be extended to support the needs of ongoing, continuous land cover monitoring.
