The rapid expansion of materials databases offers unprecedented opportunities for accelerating materials discovery via machine learning. However, the widespread assumption that larger datasets inherently produce better models does not hold in practice. We propose FUSION (Fusing Uncertainty with Structural Information for Optimal Neural training), an offline dataset pruning strategy that synergistically combines uncertainty quantification with crystallographic structure analysis via geometric fingerprinting, framing dataset pruning as a discrete optimization problem. Through evaluation across 3 benchmark datasets, FUSION consistently outperforms baselines, including random pruning, uncertainty sampling, weighting factor pruning, diversity sampling, and active learning. It demonstrates robust transferability across 11 diverse architectures, outperforming random pruning by 1.91–13.65% across different datasets, with an average improvement of 6.36%. Moreover, our analysis suggests that different models exhibit varying robustness characteristics when faced with pruned training data, highlighting the importance of model selection tailored to dataset composition. We identify optimal pruning points where removing just 0–8% of training data improves model performance, yielding gains up to 12.67% in specific model–dataset combinations. These results establish a new paradigm for materials informatics that prioritizes data quality over quantity, offering a pathway toward more efficient and sustainable machine learning workflows in computational materials science.

In this paper, we propose DGCL, a dual-graph neural networks (GNNs)-based contrastive learning (CL) integrated with mixed molecular fingerprints (MFPs) for molecular property prediction. The DGCL-MFP method contains two stages. In the first pretraining stage, we utilize two different GNNs as encoders to construct CL, rather than using the method of generating enhanced graphs as before. Precisely, DGCL aggregates and enhances features of the same molecule by the Graph Isomorphism Network and the Graph Attention Network, with representations extracted from the same molecule serving as positive samples, and others marked as negative ones. In the downstream tasks training stage, features extracted from the two above pretrained graph networks and the meticulously selected MFPs are concated together to predict molecular properties. Our experiments show that DGCL enhances the performance of existing GNNs by achieving or surpassing the state-of-the-art self-supervised learning models on multiple benchmark datasets. Specifically, DGCL increases the average performance of classification tasks by 3.73$\%$ and improves the performance of regression task Lipo by 0.126. Through ablation studies, we validate the impact of network fusion strategies and MFPs on model performance. In addition, DGCL's predictive performance is further enhanced by weighting different molecular features based on the Extended Connectivity Fingerprint. The code and datasets of DGCL will be made publicly available.

We provide an overview of the progress on producing accurate global ocean color products from the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (SNPP) and NOAA-20 satellites. VIIRS global ocean color products include normalized water-leaving radiance spectra nLw(l) at VIIRS five spectral bands, chlorophyll-a (Chl-a) concentration, water diffuse attenuation coefficients at 490 nm, Kd(490), and at the domain of photosynthetically available radiation (PAR), Kd(PAR). However, VIIRS-derived daily ocean color images on either SNPP or NOAA-20 have some limitations in ocean coverage due to its swath width, high sensor-zenith angle, sun glint, and cloud, etc. Merging VIIRS ocean color products derived from the SNPP and NOAA-20 significantly increases the spatial coverage of daily data images. The two VIIRS sensors on SNPP and NOAA-20 have similar sensor characteristics, and ocean color products are derived using the same Multi-Sensor Level-1 to Level2 (MSL12) ocean color data processing system. Therefore, the merged VIIRS ocean color data from the two sensors have high data quality with consistent statistical property and accuracy globally. We describe an approach to remove data gaps of missing pixels from the SNPP and NOAA-20 merged global ocean color data. Some applications using the new complete global data coverage of daily ocean color product are also presented and discussed.

In this paper, we provide an overview of the progress on producing accurate global ocean color products from the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (SNPP) and NOAA-20 satellites. SNPP and NOAA-20 were launched on October 28, 2011 and November 18, 2017, respectively. VIIRS global ocean color products include normalized water-leaving radiance spectra nLw(λ) at VIIRS five spectral bands, chlorophyll-a (Chl-a) concentration, water diffuse attenuation coefficients at the wavelength of 490 nm, Kd(490), and at the domain of photosynthetically available radiation (PAR), Kd (PAR). In addition, new products of nLw(λ) at VIIRS I-band (638 nm for SNPP and 642 nm for NOAA-20) and quality assurance (QA) score are now included. VIIRS global ocean color products are being routinely produced using the Multi-Sensor Level-l to Level-2 (MSL12) ocean color data processing system. Specifically, we describe our effort for the improvements of MSL12, particularly over coastal and inland waters, as well as some evaluations with in situ data from the Marine Optical Buoy (MOBY). Furthermore, we provide VIIRS ocean color data from both SNPP and NOAA-20, as well as from merged global Chl-a data from two VIIRS sensors, showing significantly improved data coverage. Some examples from the recently developed data gap-filling technique for VIIRS-SNPP are also presented and discussed.

In this paper, we provide an overview of the progress on producing accurate and consistent ocean color products from the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (SNPP). To meet requirements from broad users (e.g., operational, research, modeling, etc.), we have proposed and now been routinely producing two VIIRS ocean color data streams, i.e., the near-real-time (NRT) and delayed science quality ocean color product data. The implementation details for the two data streams will be discussed. In addition, with significantly improved satellite data processing algorithms, as well as considerably improved sensor on-orbit radiometer calibration using both solar and lunar methods, VIIRS mission-long ocean color data have been successfully reprocessed using the Multi-Sensor Level-1 to Level-2 (MSL12) ocean color data processing system. VIIRS ocean color data have been significantly improved over the global open ocean, as well as turbid coastal and inland waters. In particular, VIIRS-derived water property data over global high altitude lakes (e.g., Lake Victoria in South Africa) are now very much improved, providing a significant progress for satellite remote sensing of inland water properties. Some detailed evaluation of VIIRS ocean color products, as well as sensor performance, will be discussed. Results show that VIIRS can provide high-quality global ocean color products in support of the science researches and operational applications.

In this paper, we provide overview of the progress on the evaluations of the Visible Infrared Imaging Radiometer Suite (VIIRS) ocean color products with long-term time series and more in situ data for matchup analysis. Specifically, VIIRS ocean color products include normalized water-leaving radiance spectra nLw(λ) at VIIRS five spectral bands, chlorophyll-a concentration (Chl-a), water diffuse attenuation coefficients at the wavelength of 490 nm, Kd(490), and at the domain of photosynthetically available radiation (PAR), Kd(PAR). VIIRS ocean color products derived from the NOAA Multi-Sensor Level-1 to Level-2 (MSL12) ocean color data processing system are evaluated and compared routinely with in situ data from the Marine Optical Buoy (MOBY) and several AERONET-OC sites. In order to meet requirements from all users, we propose to routinely produce two ocean color data streams, i.e., the near-real-time (NRT) data and delayed science quality data. The NRT ocean color data stream has the advantage of quick data turn around with data latency −12–24 hours, while the science quality data stream has high consistence (with the mission-long data reprocessing) and accuracy of ocean color products. In addition, we have significantly improved on-orbit sensor calibration by combining the lunar calibration into the current solar calibration method. Our results show that VIIRS is now providing high-quality global ocean color products in support of the scientific research and operational applications.

In this paper, we provide evaluations and assessments of the Visible Infrared Imaging Radiometer Suite (VIIRS) ocean color products, including normalized water-leaving radiance spectra nLw(λ) at VIIRS five spectral bands, chlorophyll-a concentration (Chl-a), water diffuse attenuation coefficients at the wavelength of 490 nm, Kd(490), and at the domain of photosynthetically available radiation (PAR), Kd(PAR). Specifically, VIIRS ocean color products derived from the NOAA Multi-Sensor Level-1 to Level-2 (MSL12) ocean color data processing system are evaluated and compared with in situ data from the Marine Optical Buoy (MOBY) and measurements from the Moderate Resolution Imaging Spectroradiometer (MODIS). In general, VIIRS ocean color products are matched well with MOBY in situ measurements, and are also consistent with those from MODIS-Aqua. Ocean color products were found to be highly sensitive to some operational sensor calibration issues. We have improved sensor calibration by combining the lunar calibration into the current calibration method. Here, the ocean color products based on the new sensor calibration are evaluated. Our results show that VIIRS is capable of providing high-quality global ocean color products in support of the scientific research and operational applications.

The Visible Infrared Imaging Radiometer Suite (VIIRS) Raw Data Records (or Level-0 data) are processed using the current standard Algorithm Development Library (ADL) to produce Sensor Data Records (SDR; or Level-1B data). The ocean color Environmental Data Records (EDR), one of the most important product sets derived from VIIRS, are processed from the SDR of the visible and near-infrared moderate resolution (M) bands. As the ocean color EDR are highly sensitive to the quality of the SDR, the bands from which the EDR data arise must be accurately calibrated. These bands are calibrated on-orbit using the onboard Solar Diffuser, and the derived calibration coefficients are called F-factors. The F-factors used in the forward operational process may have large uncertainty due to various reasons, and thus, to obtain high-quality ocean color EDR, the SDR needs to be regularly reprocessed with improved F-factors. The SDR reprocessing, however, requires tremendous computational power and storage space, which is about 27 TB for one year of ocean-color-related SDR data. In this letter, we present an efficient and robust method for reduction of the computational demand and storage requirement. The method is developed based on the linear relationship between the SDR radiance/reflectance and the F-factors. With this linear relationship, the new SDR radiance/reflectance can be calculated from the original SDR radiance/reflectance and the ratio of the updated and the original F-factors at approximately 100th or less of the original central processing unit requirement. The produced SDR with this new approach fully agrees with those generated using the ADL package. This new approach can also be implemented to directly update the SDR in the EDR data processing, which eliminates the hassle of a huge data storage requirement as well as that of intensive computational demand. This approach may also be applied to other remote sensors for data reprocessing from raw instrument data to science data.