To refine the assumption that the contrast threshold of the human eye (Crt) remains constant in general visibility theory, we integrated an artificial intelligence-based approach with a mechanistic model (AI-SD) to derive the equivalent contrast threshold at 488 nm (C488) from remote sensing reflectance on a pixel-by-pixel basis. The derived C488 was then applied to estimate Secchi depth (Zsd). We trained and validated the AI-SD model using an extensive field-measured dataset (N = 1 577) encompassing oceanic, coastal, and inland waters and compared its performance with a traditional mechanistic model. Our findings indicate that C488 theoretically ranges from 1.85 × 10−5 sr−1 to 0.138 sr−1 and improves the accuracy of Zsd estimates from field-measured or satellite-derived remote sensing reflectance (Rrs) by over 10

Terrestrial water storage anomalies (TWSA) derived from the Gravity Recovery and Climate Experiment (GRACE) and its Follow-On mission (GRACE-FO) provide unique constraints on large-scale water mass redistribution, but their coarse effective spatial resolution limits regional applications across hydroclimatically heterogeneous domains such as the conterminous United States (CONUS). In this study, we developed a Vision Transformer (ViT)-based framework to produce GRACE-constrained, predictor-informed monthly TWSA reconstructions over CONUS by combining GRACE/GRACE-FO observations with multiple hydroclimatic predictors, including precipitation, evapotranspiration, runoff, snow-related variables, soil moisture, canopy water, and land surface temperature. To maintain consistency with satellite gravimetry, hydroclimatic predictors were harmonized to the effective GRACE spatial scale before model training. Model evaluation was conducted using a chronological temporal partition, with early-period samples used for training and validation and later-period samples reserved for independent testing. The ViT reconstruction preserved the large-scale spatial organization of GRACE while introducing predictor-informed subregional spatial heterogeneity within GRACE-consistent constraints. Benchmark comparisons with RF, CNN, and LSTM baselines further indicate that ViT provides a more balanced reconstruction in terms of regional error, temporal skill, and seasonal spatial coherence. Regional analyses reveal hydroclimate-dependent performance, with stronger agreement in the Mississippi and Colorado River basins and more conservative estimates in the Central Valley and High Plains. SHAP analysis suggests that evapotranspiration, snow, and soil moisture are the most influential predictors used by the trained model, although their relative importance varies regionally. Our findings demonstrate the potential of transformer-based data fusion to improve the regional spatial usability of GRACE/GRACE-FO TWSA products and provide a scalable pathway for regional water resource assessment.

Objective To address the challenges of low temporal validity and stability in conventional calibration methods for Chinese land observation satellites, this study proposes a comprehensive in-orbit radiometric calibration framework for wide-swath cameras onboard these satellites, leveraging a multi-site network to resolve issues of infrequent calibration and inconsistent results. The research objectives include: 1) developing a multi-site, high-frequency calibration capability by integrating synchronized measurements from automated calibration sites and semi-automated supplementary sites, thereby overcoming the limitations of traditional single-site approaches. 2) Enhancing atmospheric correction accuracy through the fusion of the site network's automated ground observations (e.g., spectrometers, sun-photometers) with multi-source data and radiative transfer modeling (MODTRAN), which reduces errors in atmospheric parameter estimation and manual operational uncertainties, achieving a radiometric calibration uncertainty of <= 6 degrees o. 3) Quantifying long-term radiometric stability of satellites such as Gaofen-6 and Gaofen-1 using time-series calibration data to determine annual variation rates and investigate how payload geometry (e.g., viewing angles, orbital configurations) mechanistically influences calibration outcomes. 4) Validating the method's universality by assessing spatiotemporal consistency and international compatibility through cross-calibration with diverse sites (Dunhuang, Baotou, La Crau) and sensors (GF series, Sentinel-2/MSI). The framework prioritizes data harmonization, uncertainty propagation modeling, and bidirectional reflectance distribution function (BRDF) corrections in automated systems. These innovations establish a high-frequency radiometric calibration system to support long-term performance monitoring and ensure data quality for China's optical satellites, including the Gaofen and Ziyuan series, while aligning with global Earth observation standards. Methods Leveraging standardized multi-source observational data from automated calibration site networks and semi-automated measurement sites, this study performed comprehensive in-orbit radiometric calibration for wide-swath cameras onboard land observation satellites. Five representative radiometric calibration sites were selected, automated sites included Baotou Sand (BSCN, China), Golmud Sand (GSCN, China), La Crau (LCFR, France), and Railroad Valley Playa (RVUS, USA), equipped with automated instruments such as hyperspectral radiometers (380-1080 nm), sun-photometers, and meteorological stations, enabling continuous 2-minute interval observations. Dunhuang Semi-automated sites (DHCN, China), where portable spectrometers measured surface reflectance and CE318 sun-photometers acquired atmospheric parameters synchronously. First, ground radiance from hyperspectral radiometers was combined with MODTRAN-simulated direct solar and diffuse sky radiation to derive surface reflectance. Second, top-of-atmosphere (TOA) radiance simulation: aerosol optical depth (AOD) and water vapor content (WVC) were input into MODTRAN to simulate TOA radiance, which was then convolved with the sensor's spectral response function to generate channel-specific irradiance. Finally, calibration coefficient derivation: satellite digital number (DN) values were linearly regressed against channel-equivalent radiance, accounting for the radiometric response characteristics of wide-swath land observation cameras. Errors in calibration results were analyzed using uncertainty propagation theory. 1) Surface reflectance uncertainty: 4 %-5 % for automated sites (instrument calibration and atmospheric errors) and 1.5% for semi-automated sites (operational errors). 2) TOA simulation uncertainty: contributions from aerosol type (2 % ), MODTRAN model error (2 % ), and solar irradiance (1.5 % ). 3) BRDF correction: angular reflectance biases were calculated using MCD43A1 BRDF products. To further analyze the uncertainty of the calibration results, a cross-calibration test method was used to verify and analyze the comprehensive calibration results. Internationally calibrated Sentinel-2/MSI data served as reference to calculate relative apparent reflectance of Chinese sensors via spatiotemporal-spectral angle matching, enabling cross-sensor validation. Spatiotemporal constraints, Data pairs with <1-hour overpass time difference and <10 degrees observation angle difference were selected to evaluate radiometric discrepancies. time-series radiometric performance evaluation of wide-swath cameras, multi-site time-series calibration coefficients were analyzed to quantify year-on-year radiometric drift. Standard deviations of calibration coefficients were computed to evaluate payload stability over mission durations. Results and Discussions The proposed multi-site calibration framework significantly enhanced the radiometric calibration frequency and accuracy for Chinese land observation satellites. For the GF-6/WFV camera, annual calibration frequency reached 30 times- 5-30 times higher than traditional single-site methods-while GF-1 series wide-swath cameras achieved 7 -9 calibrations annually. Automated calibration uncertainties ranged from 5.26 degrees o (GF-6) to 9.81 degrees o (GF-1/WFV4), whereas semi-automated methods reduced uncertainty to 1.5 % by minimizing atmospheric retrieval errors. Payload stability analysis revealed an annual variation rate of 1.818 % for GF-1/WFV1's blue band and-2.393% for GF-6's near-infrared band, demonstrating robust long-term performance. However, large viewing angles (e.g., GF-1/WFV1, view zenith angle (VZA) is 26.09 degrees) introduced BRDF-related uncertainties up to 9.41 degrees o, underscoring the need for angular correction models. Cross-validation with Sentinel-2/MSI confirmed a band-average difference <6 degrees o, but discrepancies exceeding 10% at VZA>15 degrees highlighted the critical role of BRDF adjustments. Automated RadCalNet sites (e.g., RVUS, BSCN) improved standardization through cloud-filtered data (AOD<0.3) and remote operations, reducing manual intervention. The inclusion of GSCN further leveraged plateau-specific low-aerosol conditions to refine accuracy. However, limitations persisted: fixed-location automated sites lacked spatial representativity, while semi-automated sites faced frequency constraints. Additionally, MODTRAN simulations under complex atmospheres (e.g., sandstorms) required further validation. Conclusions This study established a high-frequency, multi-source radiometric calibration system integrating automated and semi-automated methods, yielding three key outcomes: 1) Technical advancements: GF-6/WFV achieved 30 annual calibrations, transitioning China's satellite calibration from "annual" to "monthly" frequencies. A BRDF correction evaluation method quantified large-angle uncertainties, enabling geometric normalization strategies. 2) Operational impact: enhanced radiometric consistency across >40 satellites (Gaofen, Ziyuan series) improved agricultural and environmental monitoring reliability. GSCN's integration into RadCalNet elevated China's role in global calibration networks. 3) Future directions: dynamic BRDF correction models and polar calibration sites (leveraging ice/snow reflectance) will expand global network robustness. AI-driven real-time uncertainty estimation could streamline automated calibration workflows. This framework operationalizes high-frequency calibration for Chinese optical satellites, advancing the standardization of global remote sensing data and fostering interoperability in Earth observation.

To assess the ability of a satellite instrument to detect submerged targets, we constructed a semi-analytical relationship to link target reflectance and the contrast threshold of the satellite instrument to visibility ranges. Using numerical simulation, we found that the contrast threshold of the satellite instrument was equal to 50 % of the residual error contained in satellite R-rs data. We evaluated our model using known sea depths of optically shallow water and found that the model produced similar to 16 % uncertainty in retrieving the visibility range around the edge of the optically shallow water. By comparison, the contrast threshold of the human eye was more than 20 times larger than the satellite instrument contrast threshold. In addition, using a Secchi disk submerged in the shallow water, we found that the Secchi disk was invisible to the human eye when the disk was still visible to a high-quality camera handheld or mounted on an unmanned aerial vehicle. Moreover, when the image data quality was as well as MODIS instrument, we found that the maximum instrument visibility range reached 130 m in theory, which was approximately four times larger than the maximum reached by the human eye. Our findings suggest that high-quality cameras such as satellite instruments are more effective than the human eye for detecting underwater targets.

Red-green-blue (RGB) images (or videos) captured by consumer-level uncrewedaerial vehicle (UAV) cameras are widely used in high-resolution remote observations. However, digital number (DN) values of these RGB images usually have a nonlinear relationship with the incident radiance, which reduces the accuracy of quantitative remote sensing of macroalgae. To solve this problem, we proposed an improved processing procedure for UAV RGB images (or videos) based on camera response functions (CRFs). The CRF was utilized to convert the DN values into energy values (E values), which demonstrate a linear relationship with the incident radiance. When the DN values were replaced by their corresponding E values to calculate the reflectance of green macroalgae under different illumination intensities, the errors in reflectance were reduced by similar to 21%; for the corresponding green macroalgae indices, such as the red-green band virtual baseline floating green algae height (RG-FAH), the E-value-based RG-FAH demonstrates more resistance to the impacts of sun glints; and the E values were further applied to estimate the coverage portion of macroalgae (POM, %) in RGB videos; the illumination-induced deviations of the POM were effectively reduced by up to 33.06%, showing an advantage in quantitative estimation of macroalgae biomass. The results of applications to UAV RGB images show that the E values have significant suitability in estimating POM across diverse green macroalgae species and various algae indices, suggesting promising potentials of the proposed processing procedure with E-based photo and/or video RGB images in monitoring aquatic plants and environment.

Over the last three decades, ocean color satellites have recorded continuous change in phytoplankton concentration, and in doing so, the satellites have helped marine remote sensing to become an emerging discipline. However, ocean color satellite has provided the state of global phytoplankton only around midday whereas phytoplankton have significant diurnal variation characteristics over an entire day-night cycle, which would lead to large estimation uncertainty in downstream products. A large global field observations show that the diurnal variation curves can be constructed as the sum of an exponential and a polynomial function. After correction with our diurnal variation curves, the intermission difference between MODISA and MODIST phytoplankton product could be decreased from 7.57% to 6%, which is critical to monitor the long-term changing trend with multi-mission data. When the satellite-recorded chlorophyll-a concentration is simply equal to a daily average value, it results in overestimating uncertainties of 16% and 54% for the northern and southern hemispheric oceans, respectively. Moreover, as a typical example of downstream product of the chlorophyll-a, the coverage of oligotrophic waters can oscillate 76% from day to night simulated from satellite images with our diurnal variation curves, further confirming the importance of global synchronous observation. The geostationary satellite might be an efficient regional tool for diurnal observation, but it must account for its limitations under the weak light conditions of early morning and late afternoon. In the real world, the effects of diurnal variations in ocean properties reach far beyond chlorophyll-a in the global oceans. We anticipate these quantitatively findings to be a starting point for quantitatively re-examining the proper application of ocean color products in scientific communication.

The role of land surface temperature (LST) is of the utmost importance in multiple academic disciplines, such as climatology, hydrology, ecology, and meteorology. To date, many methods have been proposed to estimate LST from satellite thermal infrared data. The single-channel (SC) algorithm can provide an accurate result in retrieving LST based on prior knowledge of known land surface emissivity (LSE). The SC algorithm is extensively employed for retrieving LST from Landsat series data due to its simplicity and its reliance on just one thermal infrared channel. The Thermal Infrared Sensor (IRS) on the Chinese ZY1-02E satellite is a pivotal instrument employed for gathering thermal infrared (TIR) data of land surfaces. The objective of this research is to evaluate the feasibility of a single-channel approach based on water vapor scaling (WVS) for deriving LST from ZY1-02E IRS data because of its wide spectrum range, i.e., 7~12 μm, which is affected strongly by both atmospheric water vapor and ozone. Three study areas, namely the Baotou, Heihe River Basin, and Yantai Sea sites, were selected as validation sites to evaluate the LST inversion accuracy. This evaluation was also conducted via cross-comparison between the retrieved LST and MODIS LST products. The results revealed that the WVS-based method exhibited an average bias of 0.63 K and an RMSE of 1.62 K compared to the in situ LSTs. The WVS-based method demonstrated reasonable accuracy through cross-validation with the MODIS LST product, with an average bias of 0.77 K and an RMSE of 2.0 K. These findings indicate that the WVS-based method is effective in estimating LST from ZY1-02E IRS data.

To enhance the long-term monitoring ability of multi-sourced remote sensing data, we needed to minimize the inter-mission biases between the data retrieved from different satellites. We evaluated three existing diffuse attenuation coefficients at 490 nm (K-d(490)) using nine independent datasets collected from the global oceans. The results indicate that the updated neural network-based four-band K-d(490) retrieval model (updated-NFKM) decreased the uncertainty by > 4% from the NASA official K-d(490) retrieval model (NOKM) and the inherent optical properties-based K-d(490) retrieval model (IOPK). Specifically, matchup analysis showed that the updated-NFKM model produced < 40% uncertainty in deriving K-d(490) from MODISA and SeaWiFS data, and the uncertainty of SeaWiFS-predicted K-d(490) was similar to 5% lower than the MODISA-predicted K-d(490). Using the updated-NFKM model, >80% of the global ocean had an uncertainty for K-d(490) estimates that were lower than 30%, while the model performed much better for the Western Pacific, Arctic Ocean, and Northern Atlantic compared to the Eastern Pacific, South Ocean, and Southern Atlantic. To enable a naturally smooth transition from SeaWiFS to MODISA-observed K-d(490) products, it was critical to cross-calibrate the inter-mission difference. The results show that using the cross-calibration models proposed in this study, the MODISA-predicted K-d(490) accounted for > 86% of the variations of SeaWiFS-predicted K-d(490), even though the empirical coefficients of the cross-calibration model had to be adjusted according to the time scales of the composite remote sensing data. Finally, we used the cross-calibration model to minimize the time series of SeaWiFS and MODISA-predicted K-d(490) data. Our results indicate that the K-d(490) values gradually increased in the low-latitude regions during past two decades.

Remote sensing reflectance (Rrs) is an essential parameter in ocean color remote sensing and a fundamental input for the estimation of ocean color elements. Predicting Rrs has the potential to enable simultaneous prediction of multiple marine environmental parameters, facilitating multi-perspective analysis of marine environmental changes. This paper proposes a spatiotemporal attention-augmented ConvLSTM-based model for ocean Rrs prediction. The developed model can predict Rrs for up to seven days by simultaneously learning spatiotemporal features from time series Rrs and auxiliary environmental variables. According to the experiments, the proposed model achieves optimal performances on Rrs predictions at 443, 488, and 555 nm, with Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) for the first four prediction days less than 5.6*10-4 sr-1 and 8.6 %, respectively, which are better than the performance of the convolutional neural network (CNN), the LSTM, CNN-LSTM, and the ConvLSTM. The spatial and temporal variations of Rrs are also compared to evaluate the effectiveness of the model, presenting a consistent spatiotemporal pattern between predicted and observed Rrs. We also found that integrating sea surface temperature (SST), photosynthetically available radiation (PAR), and aerosol optical thickness at 869 nm (AOT869) into the model can improve the prediction accuracy in various degrees. This work suggests the proposed deep learning model can predict Rrs for 7 days with a convincing performance, providing critical data and technical support for ocean-related applications, such as algae bloom monitoring.

In this study, a neural network-based Secchi depth retrieval model (NNSD) was developed using observations obtained during 2003-2012 in the Bohai, Yellow, and China East Seas (Eastern China Coastal Seas, ECCS). Based on the results of the analyses, the NNSD model produced less than 25% uncertainty in quantifying the Secchi depth in the optically complex ECCS. The slope of the linear relationship between the field-measured and NNSD model-derived Secchi depth varied from 0.98 to 1.08 among the datasets. However, the corresponding determination coefficients were not any lower than 0.91. To determine the effectiveness of the NNSD model in deriving the Secchi depth in the ECCS, the performances of the three existing models were also evaluated, and the results are presented in this study. By comparison, using the NNSD model decreased the uncertainty by 26% when compared with the three existing models in deriving the Secchi depth in the ECCS. Finally, the NNSD model was further applied to the MODIS data over the ECCS to briefly illustrate its applicability to general oceanographic studies. The NNSD model could produce 24.62% MRE and 0.14 RMSE values in deriving the Secchi depth from the ECCS, which was > 26% MRE and > 0.07 RMSE values better than all the BGSD, KDSD and SASD models. In regards to the spatial distribution pattern of the Secchi depth, the eastern and northern sections of the ECCS were much higher than the western and southern areas. Additionally, the temporal distribution mode indicated that the winter season was higher than the summer season. These spatiotemporal characteristics were caused primarily by the regional climate, river discharges, and strong tidal currents and winds, among other factors.

Remote sensing reflectance (R (rs)) and inherent optical properties (IOPs) conversions are fundamental in accurate satellite measurements to guarantee the semi-analytical retrieval quality of IOPs and the biogeochemical products. Traditionally, the R (rs)-IOPs conversions are determined by a quadratic polynomial function with two model parameters (G (x = 0,1)). However, G (x) values vary in time and location, which are attributed to the spatial and temporal variability inherent to illumination conditions, sea surface properties, and meteorological states. To improve the performance of three classical existing models used for R (rs)-IOPs conversions, we designed two novel neural network models (NNG (x = 1,2)) to quantitatively calculate G (x) from the R (rs) spectrum pixel by pixel without requirement of any auxiliary illumination and meteorological data, and then proposed for R (rs)-IOPs conversions. We evaluated these approaches with numerical simulations and field measurements, and the results show that the NNG (x) models are more effective in semi-analytically converting R (rs) into IOPs than the three existing models. Furthermore, we applied the NNG (x) models to satellite images to understand the downstream influence of the G (x) values on IOPs estimates for the global oceans. We further confirm that the G (x) values dramatically change for the global ocean, which is especially true for very oligotrophic gyres, coastal waters, and high latitude oceans. When we use a constant G (x) for the R (rs)-IOPs conversions, it leads to substantial uncertainty of up to 30% in the IOPs retrievals for China's coastal regions. Our results suggest that it is possible to improve the data quality of IOPs for the global oceans by providing accurate pixel-level G (x) values using NNG (x) models.

To analytically or semi-analytically convert satellite-measured remote sensing reflectance (R (rs)) to widely used biogeochemical variables from the global oceans, it is common to first determine the total absorption coefficient from satellite R (rs) data and then separate the total absorption coefficient into the absorption coefficients of phytoplankton (a (ph) ) and non-phytoplankton (a (dg) ) pigments. While still considering the spectral characteristics of a (ph) and a (dg) , we developed an operational model to improve the accuracy of estimating a (ph) and a (dg) products from satellite R (rs) data. Our results show that our model effectively separates a (ph) and a (dg) from the total absorption coefficient spectrum, which decreases the uncertainty by 6.81% to 17.52% compared to a widely used quasi-analytical algorithm. Moreover, after applying the new model to SeaWiFS satellite images, we present the spatial and temporal variations of a (ph) (443) and a (dg) (443) for the global oceans from 1997 to 2010, further confirming that coastal zones exhibited higher a (ph) (443) and a (dg) (443) than the open oceans during that time. In addition, a (ph) (443) and a (dg) (443) declined in the tropical Indian Ocean Gyre over those 13 years. Our results improve our knowledge of the spatiotemporal variations of optical properties of the global oceans, and also demonstrate the potential of our model in separating the total absorption.

Algal cell abundance weakly depends on inherent optical properties and chlorophyll-a concentration in the Bohai Sea, so it is very hard to derive algal cell abundance (ACA) from ocean color data using a simple bio-optical model. To obtain ACA for biological communication at large scale, a neural network model has been developed and then applied for investigating the changing monthly trend of ACA, intracellular chlorophyll-a concentration, and cell size in the Bohai Sea using MODIS data from 2002 to 2015. The results showed that the neural network model could provide an accurate log-transformed value of algal cell abundance (LACA) from ocean color images whose retrieval uncertainty did not exceed 9%. Furthermore, when the model was applied to map the monthly mean LACA and then further convert it to cell size in the Bohai Sea, the results showed that the satellite-derived monthly mean cell size varied from 4.81 to 15.29 μm. The decreasing monthly mean algal cell abundance and increasing monthly mean chlorophyll-a concentration imply that the monthly mean intracellular chlorophyll-a concentration from 2002 to 2015 increased, which indicates that the waters in the Bohai Sea became more eutrophic over those 14 years. Moreover, due to seasonal variations in vertical mixing or other physical forcing factors, the ACA and cell size exhibited significant seasonal variations. Although further tests are required to validate the model’s robustness, these preliminary results indicate that the neural network model is an encouraging approach to exploiting more novel biological parameters such as the LACA from ocean color satellites for oceanic communication.

The residual error was a critical indicator to measure the data quality of ocean color products, which allows a user to decide the valuable envisioned application of these data. To effectively remove the residual errors from satellite remote sensing reflectance (Rrs) using the inherent optical data processing system (IDAS), we expressed the residual error spectrum as an exponential plus linear function, and then we developed neural network models to derive the corresponding spectral slope coefficients from satellite Rrs data. Coupled with the neural network models-based spectral relationship, the IDAS algorithm (IDASnn) was more effective than an invariant spectral relationship-based IDAS algorithm (IDAScw) in reducing the effects of residual errors in Rrs on IOPs retrieval for our synthetic, field, and Chinese Ocean Color and Temperature Scanner (COCTS) data. Particularly, due to the improved spectral relationship of the residual errors, the IDASnn algorithm provided more accurate and smoother spatiotemporal ocean color product than the IDAScw algorithm for the open ocean. Furthermore, we could monitor the data quality with the IDASnn algorithm, suggesting that the residual error was exceptionally large for COCTS images with low effective coverage. The product effective coverage should be rigorously controlled, or the residual error should be accurately corrected before temporal and spatial analysis of the COCTS data. Our results suggest that an accurate spectral relationship of residual errors is critical to determine how well the IDAS algorithm corrects for residual error.

Remotely sensed hyperspectral data can support more effective water quality monitoring. Nevertheless, the variability and complexity of urban river water make it hard to retrieve comprehensive water quality characteristics directly, so that most current water quality assessments rely on semiempirical, semianalytical, or bio-optical approaches. In this study, we carried out simultaneous in situ hyperspectral data and water quality measurements. We used the 382 hyperspectral data from urban rivers of Zhongshan City in the Pearl River Delta to test how well the random forest (RF) and one-dimensional convolutional neural networks (1D-CNNs) algorithms retrieved the newly established water quality index (WQI). The RF algorithm also identified essential wavelengths for retrieving the WQI. Our results demonstrate that the RF and 1D-CNN algorithms performed well in WQI estimations. The 1D-CNN model performed significantly better than the RF model, especially on high WQI samples. Both models were insensitive to smoothing of the hyperspectral data, showing that the noise of the original hyperspectral reflectance data has a limited impact on the algorithms. In addition, when we used the essential wavelength data (mainly located between 580–590 nm and near 722, 751, 821, and 830 nm) as input data, we achieved better retrieval results. The 1D-CNN model performed the best with an $R^{2}$ of 0.87, RMSE of 0.574, and RPIQ of 3.082 when we used the top tenth percentile of the essential wavelength data. This study demonstrates the potential of the 1D-CNN algorithm for hyperspectral data analysis to retrieve comprehensive water quality.

Knowing how much measurement noise is in a signal is critical for evaluating the overall performance of a satellite observation. We developed a triple collocation observation (TCO) algorithm for estimating measurement noise by collocation comparing the local deviations of three satellite data sets. When we evaluated our algorithm with a synthetic data set, the results showed that the algorithm effectively derived measurement noise from satellite signals despite the many intermission signal differences among the satellites. The TCO algorithm produced <6.66% uncertainty in the measurement noise estimates that we derived from the synthetic data set. In addition, to maximally isolate measurement noise from ocean color images, we developed a set of data quality control criteria to apply when identifying synchronous pixel pairs. Using images from the Medium Resolution Spectral Imager II (MERSI II), the Visible Infrared Imaging Radiometer Suite (VIIRS), and the Moderate Resolution Imaging Spectroradiometer (MODIS) instruments, we applied our data quality control criteria and found that the TCO algorithm produced measurement noise consistent with the measured prelaunch or specifications for VIIRS and MERSI II instrument noise. However, the TCO measurement noise was significantly lower than the spaced MODIS noise because MODIS’s extended service time likely produced instrument degradation. Overall, MODIS performed better than MERSI II but worse than VIIRS. Furthermore, we found that the residual error in remote sensing reflectance exponentially decreased as the measurement signal-to-noise ratio (MSNR) increased. Because of this exponential relationship, the MSNR should not be lower than 181 to achieve the <5% uncertainty goal of remote sensing reflectance at 443 nm that NASA proposed. Our results suggest that the TCO algorithm is an effective approach for comprehensively estimating and comparing instrument performance.

Comparing inter-mission space instrument performance is crucial to accurate satellite measurements which guarantees the quality of ocean color products. However, comparing inter-mission instrument performance is limited by strong dispersion, which clearly originates from the spatial and temporal variability inside the oceanic sampling sites. We designed a novel comprehensive score metric (CSM) to quantitatively identify the candidate pseudo-invariant calibration sites (PICS) for inter-mission comparisons. We calculated the CSM from a year of ocean color and meteorological products from 2018 using a pixel-by-pixel method with a simple average of a temporal meteorological metric, a spatial aerosol metric, a temporal optical metric, a spatial optical metric, a data quality metric, a spectral shape metric, and a directional homogeneity metric. When we filtered the data with the threshold CSM > 0.6, we found that atmospheric and oceanic conditions from two smooth belts in the low latitude open ocean were more clear, stable, and homogeneous than other regions, and we suggested these smooth belts as candidate PICS. With image data from the Visible Infrared Imaging Radiometer (VIIRS) and Medium Resolution Spectral Imager II (MERSI II), we found that our candidate PICS were more effective than regions with low CSM in providing stable synchronous data for inter-mission comparisons. The MERSI II instrument, however, experienced significant degradation in radiance measurements from January to April in 2021, while the VIIRS instrument performed well. These results suggested that the CSM values were “experimental” but, under restrictive conditions, were sufficient for an inter-mission comparison and calibration application.

Climate change and human activities have been heavily affecting oceanic and inland waters, and it is critical to have a comprehensive understanding of the aquatic optical properties of lakes. Since many key watercolor parameters of Qinghai Lake are not yet available, this paper aims to study the spatial and temporal variations of the water clarity (i.e., Secchi-disk depth, ZSD) and suspended particulate matter concentration (CSPM) in Qinghai Lake from 2001 to 2020 using MODIS images. First, the four atmospheric correction models, including the NIR–SWIR, MUMM, POLYMER, and C2RCC were tested. The NIR–SWIR with decent accuracy in all bands was chosen for the experiment. Then, four existing models for ZSD and six models for CSPM were evaluated. Two semi-analytical models proposed by Lee (2015) and Jiang (2021) were selected for ZSD (R2 = 0.74) and CSPM (R2 = 0.73), respectively. Finally, the distribution and variation of the ZSD and CSPM were derived over the past 20 years. Overall, the water of Qinghai Lake is quite clear: the monthly mean ZSD is 5.34 ± 1.33 m, and CSPM is 2.05 ± 1.22 mg/L. Further analytical results reveal that the ZSD and CSPM are highly correlated, and the relationship can be formulated with ZSD=8.072e−0.212CSPM (R2 = 0.65). Moreover, turbid water mainly exists along the edge of Qinghai Lake, especially on the northwestern and northeastern shores. The variation in the lakeshore exhibits some irregularity, while the main area of the lake experiences mild water quality deterioration. Statistically, 81.67% of the total area is dominated by constantly increased CSPM, and the area with decreased CSPM occupies 4.56%. There has been distinct seasonal water quality deterioration in the non-frozen period (from May to October). The water quality broadly deteriorated from 2001 to 2008. The year 2008 witnessed a sudden distinct improvement, and after that, the water quality experienced an extremely inconspicuous degradation. This study can fill the gap regarding the long-time monitoring of water clarity and total suspended matter in Qinghai Lake and is expected to provide a scientific reference for the protection and management of the lake.

In this study, we combined ground-based hyperspectral data, unmanned aerial vehicles (UAVs) remotely sensed hyperspectral images, and 1D-CNN algorithms to quantitatively characterize and estimate the Chemical Oxygen Demand (COD) of estuarine urban rivers. The spectral response mechanism of COD is imprecise due to its complex composition; however, we found that hyperspectral remote sensing data could be used for COD monitoring because of the data's rich spectral information. The potential of hyperspectral sensors installed on UAVs to estimate and map the COD of urban rivers has not been thoroughly explored. We used in situ above-water hyperspectral data from 498 sites and synchronous water samples in band ratio, SVM, and 1D-CNN algorithms to build retrieval models. We found that the 1D-CNN model performed the best with an R2 of 0.78 and an RMSE of 5.22 when using the original reflectance data as input. The 1D-CNN model may also have a better ability to identify water samples with abnormally high concentrations. Our results revealed that transferring the ground-based derived 1D-CNN retrieval model for COD to the high-resolution hyperspectral images is a reliable method for determining COD from the images. We concluded that UAV remotely sensed hyperspectral images are valuable for COD concentration monitoring and mapping, critical to urban water quality management decision-making.