Search

Satellite vegetation index (VI) products, such as normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI), have been widely used. However, they are severely contaminated by clouds and other factors and provide false signals of the surface vegetation conditions. In this study, the new global seamless 250 m, eight-day NDVI and EVI products from 2000–2021 were developed from Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance data using a long short-term memory (LSTM) neural network method. High-quality globally representative time series VI samples were constructed to train the model using a combination of the Savitzky-Golay filter (SG), Global LAnd Surface Satellite (GLASS) leaf area index (LAI) fitting and upper envelope methods. To evaluate the proposed method and the 250 m VI products, the MODIS VI product (MOD13Q1) was used for the inter-comparisons using four widely used VI reconstruction methods. Assuming that the MODIS VI data of high quality represents the true values, the root mean square error (RMSE) for NDVI and EVI generated by the LSTM model are 0.0734 and 0.0509, respectively. [ABSTRACT FROM AUTHOR]

Motivated by the lack of long-term global soil moisture products with both high spatial and temporal resolutions, a global 1 km daily spatiotemporally continuous soil moisture product (GLASS SM) was generated from 2000 to 2020 using an ensemble learning model (eXtreme Gradient Boosting – XGBoost). The model was developed by integrating multiple datasets, including albedo, land surface temperature, and leaf area index products from the Global Land Surface Satellite (GLASS) product suite, as well as the European reanalysis (ERA5-Land) soil moisture product, in situ soil moisture dataset from the International Soil Moisture Network (ISMN), and auxiliary datasets (Multi-Error-Removed Improved-Terrain (MERIT) DEM and Global gridded soil information (SoilGrids)). Given the relatively large-scale differences between point-scale in situ measurements and other datasets, the triple collocation (TC) method was adopted to select the representative soil moisture stations and their measurements for creating the training samples. To fully evaluate the model performance, three validation strategies were explored: random, site independent, and year independent. Results showed that although the XGBoost model achieved the highest accuracy on the random test samples, it was clearly a result of model overfitting. Meanwhile, training the model with representative stations selected by the TC method could considerably improve its performance for site- or year-independent test samples. The overall validation accuracy of the model trained using representative stations on the site-independent test samples, which was least likely to be overfitted, was a correlation coefficient (R) of 0.715 and root mean square error (RMSE) of 0.079 m 3 m -3. Moreover, compared to the model developed without station filtering, the validation accuracies of the model trained with representative stations improved significantly for most stations, with the median R and unbiased RMSE (ubRMSE) of the model for each station increasing from 0.64 to 0.74 and decreasing from 0.055 to 0.052 m 3 m -3 , respectively. Further validation of the GLASS SM product across four independent soil moisture networks revealed its ability to capture the temporal dynamics of measured soil moisture (R=0.69–0.89; ubRMSE = 0.033–0.048 m 3 m -3). Lastly, the intercomparison between the GLASS SM product and two global microwave soil moisture datasets – the 1 km Soil Moisture Active Passive/Sentinel-1 L2 Radiometer/Radar soil moisture product and the European Space Agency Climate Change Initiative combined soil moisture product at 0.25 ∘ – indicated that the derived product maintained a more complete spatial coverage and exhibited high spatiotemporal consistency with those two soil moisture products. The annual average GLASS SM dataset from 2000 to 2020 can be freely downloaded from 10.5281/zenodo.7172664 (Zhang et al., 2022a), and the complete product at daily scale is available at http://glass.umd.edu/soil%5fmoisture/ (last access: 12 May 2023). [ABSTRACT FROM AUTHOR]

Inaccurate parameter estimation is a significant source of uncertainty in complex terrestrial biosphere models. Model parameters may have large spatial variability, even within a vegetation type. Model uncertainty from parameters can be significantly reduced by model–data fusion (MDF), which, however, is difficult to implement over a large region with traditional methods due to the high computational cost. This study proposed a hybrid modeling approach that couples a terrestrial biosphere model with a data-driven machine learning method, which is able to consider both satellite information and the physical mechanisms. We developed a two-step framework to estimate the essential parameters of the revised Integrated Biosphere Simulator (IBIS) pixel by pixel using the satellite-derived leaf area index (LAI) and gross primary productivity (GPP) products as "true values." The first step was to estimate the optimal parameters for each sample using a modified adaptive surrogate modeling algorithm (MASM). We applied the Gaussian process regression algorithm (GPR) as a surrogate model to learn the relationship between model parameters and errors. In our second step, we built an extreme gradient boosting (XGBoost) model between the optimized parameters and local environmental variables. The trained XGBoost model was then used to predict optimal parameters spatially across the deciduous forests in the eastern United States. The results showed that the parameters were highly variable spatially and quite different from the default values over forests, and the simulation errors of the GPP and LAI could be markedly reduced with the optimized parameters. The effectiveness of the optimized model in estimating GPP, ecosystem respiration (ER), and net ecosystem exchange (NEE) were also tested through site validation. The optimized model reduced the root mean square error (RMSE) from 7.03 to 6.22 gC m -2 d -1 for GPP, 2.65 to 2.11 gC m -2 d -1 for ER, and 4.45 to 4.38 gC m -2 d -1 for NEE. The mean annual GPP, ER, and NEE of the region from 2000 to 2019 were 5.79, 4.60, and -1.19 Pg yr -1 , respectively. The strategy used in this study requires only a few hundred model runs to calibrate regional parameters and is readily applicable to other complex terrestrial biosphere models with different spatial resolutions. Our study also emphasizes the necessity of pixel-level parameter calibration and the value of remote sensing products for per-pixel parameter optimization. [ABSTRACT FROM AUTHOR]

The medium-resolution spectral imager (MERSI) onboard China’s new generation of polar-orbiting meteorological satellite series FengYun-3 (FY-3) is providing global observations of the Earth’s environment. One of the important characteristics of the MERSI sensor is that it has four bands at 250-m spatial resolution. However, very few high-level land surface and atmospheric products have been generated from MERSI data. This article presents an effective optimization method to estimate the continuous temporal distribution of multiple land surface and atmospheric variables at a 250-m spatial resolution from time series FY-3B MERSI top-of-atmosphere (TOA) observations. The MODIS cloud detection method was applied to the MERSI TOA reflectance to determine the clear observations. Then, a physics-based BRDF correction was adopt to reduce the topographic effects using the slope angle and aspect angle of the slope. Finally, a coupled land surface-atmosphere radiative transfer (RT) model and the shuffled complex evolution (SCE) optimization algorithm were used to estimate a suit of variables, including the leaf area index (LAI), the aerosol optical depth (AOD), the cloud optical thickness (COT), the cloud effective particle radius (CER), the land surface reflectance, the shortwave and visible albedo, the incident shortwave radiation (ISR), the incident photosynthetically active radiation (PAR), the fraction of absorbed PAR (FAPAR), the surface broadband emissivity (BBE), and the TOA shortwave albedo. The method was applied to the data from the 18 SURFRAD, the America FluxNet (AmeriFlux), and the Images and Beijing Normal University net (BNUnet) sites. All the estimated variables were also compared with Moderate Resolution Imaging Spectroradiometer (MODIS) and Global Land Surface Satellite (GLASS) products. The estimated LAI, PAR, FAPAR, shortwave albedo, and ISR were also validated using ground measurements from the selected sites. The results show that our method can simultaneously estimate multiple temporally continuous variables in both clear-and cloudy-sky conditions, with an accuracy comparable to those of MODIS and GLASS products. The estimated LAI, PAR, FAPAR, shortwave albedo, and ISR are consistent with the field measurements, with coefficients of determination ($R^{2}$) of 0.692, 0.783, 0.788, 0.559, and 0.833, and root mean square errors (RMSEs) of 0.427, 56.681 W/m2, 0.092, 0.084, and 115.305 W/m2, respectively. The proposed algorithm can effectively estimate the instantaneous atmospheric and land surface variables directly from satellite observations without cumbersome atmospheric corrections. Moreover, it can potentially be applied to multiple satellite observations. [ABSTRACT FROM AUTHOR]

Since a new generation of geostationary satellite data has incredibly high temporal, spatial, and spectral resolutions, new methodologies are now needed to take advantage of both the temporal and spectral signatures of them for accurate estimation of Earth’s environmental variables. This article describes a novel optimization method to estimate a suite of 11 physically consistent land surface and atmospheric variables under all-sky conditions from the geostationary advanced Himawari imager (AHI) top-of-atmosphere (TOA) observations. This method is based on a coupled soil, snow, vegetation, and atmospheric radiative transfer (RT) model from 0.28 to 14 μm. The inversion algorithm consists of three major steps. First, the “clearest” observations at each moment during a temporal window were determined and then the essential variables that characterize surface RT models, such as leaf area index (LAI), leaf chlorophyll concentration, and soil parameters were estimated. Second, the atmospheric variables, including aerosol optical depth (AOD) under clear-sky conditions, and cloud optical thickness (COT) and cloud effective particle radius (CER) under cloudy-sky conditions, were inverted given surface reflectance calculated by the surface RT models. Finally, the inverted atmospheric and land surface variables were fed into the coupled RT model to calculate the remaining set of variables, including spectral directional reflectance, surface broadband albedo, thermal emissivity, incident shortwave radiation (ISR), photosynthetically active radiation (PAR), fraction of absorbed PAR by green vegetation (FAPAR), and TOA shortwave albedo. The retrieved variables were validated using in-situ measurements from Ozflux network sites and compared with the other existing satellite products. Intercomparisons demonstrate that the AHI-retrieved atmospheric variables (AOD, CER, and COT) and surface variables (surface reflectance, LAI, FAPAR, PAR, and surface emissivity) are well correlated with the corresponding JAXA released AHI, NASA Moderate Resolution Imaging Spectroradiometer (MODIS) and Clouds and the Earth’s Radiant Energy System (CERES), and the Global LAnd Surface Satellite (GLASS) products. Direct validation using in-situ measurements indicates that the retrieved ISR achieves higher accuracy than the CERES ISR product (with R2 values of 0.95 and 0.89, and root-mean-square error (RMSE) of 20.3 and 29.6 Wm2 for the AHI-retrieved and CERES daily ISR, respectively). Validation also shows that the estimated daily surface albedo has an accuracy comparable to the MODIS daily albedo product (RMSE = 0.03). Both the direct validation and product comparisons have demonstrated that this proposed inversion framework works very well for the AHI data. Unlike other algorithms that are usually used for estimating an individual parameter and rely heavily on a separate atmospheric correction, this inversion framework can effectively estimate a group of atmospheric and land surface variables and be easily applied to other similar multispectral geostationary satellite data. A comprehensive sensitivity and validation study is still needed to quantify the uncertainties of the retrieval variables. [ABSTRACT FROM AUTHOR]

Leaf area index (LAI), fraction of absorbed photosynthetically active radiation (FAPAR), and surface broadband albedo are three routinely generated land-surface parameters from satellite observations, which have been widely used in land-surface modeling and environmental monitoring. Currently, most global land products are retrieved separately from individual satellite data. Many issues, such as data gaps, spatial and temporal inconsistencies, and insufficient accuracy under certain conditions resulting from the inadequacies of single-sensor observations, have made the incorporation of multiple sensors a reasonable solution. In this paper, an approach to simultaneous estimation of LAI, broadband albedo, and FAPAR from multiple-satellite sensors is further refined. The method, improved from that proposed in an earlier study using Moderate Resolution Imaging Spectroradiometer (MODIS) data, consists of several steps. First, a coupled dynamic and radiative-transfer model based on MODIS, SPOT/VEGETATION, and Multiangle Imaging SpectroRadiometer data was developed to retrieve LAI values and use them to construct a time-evolving dynamic model. Second, an iteration process with predefined exit criteria was developed to obtain consistent gap-filled LAI estimates. Third, a spectral albedo based on the retrieved LAI values was simulated using a radiative-transfer model and then converted to a broadband albedo using empirical methods. Snow-covered pixels identified by normalized difference snow index thresholds were adjusted to the weighted average of the underlying albedo and the maximum snow albedo. Finally, the FAPAR of green vegetation was calculated as a combination of the albedo at the top of the canopy, the soil albedo, and the transmittance of the PAR down to the background. Validation of retrieved LAI, albedo, and FAPAR values obtained from multiple-satellite data over ten study sites has demonstrated that the proposed method can produce more accurate products than presently distributed global products. [ABSTRACT FROM PUBLISHER]

The Visible Infrared Imaging Radiometer Suite (VIIRS) has observed the Earth since 2011 and will continue for several decades. Unfortunately, few high-level land surface products have been produced and publicly released. The numerical inversion of radiative transfer model (RTM) has long been used for the retrieval of land surface and atmospheric variables from satellite data; however, it is computationally intensive. Some methods (e.g., using look-up tables) are efficient but are most suitable for estimating individual variable. Various machine learning (ML) models have been developed, mostly trained by either ground measurements that may not be adequate to represent different conditions, or RTM simulations that may not be realistic due to model uncertainty and unnecessary combinations of illumination-viewing geometries and variable values. In this study, we replaced the numerical inversion scheme by the ML algorithm that is trained using RTM inversion for estimating multiple variables simultaneously. Our training dataset for the ML models was generated by inverting the land and atmospheric variables from VIIRS top-of-atmosphere (TOA) reflectance data using a coupled land-surface–atmosphere RT model. Two multi-output ML algorithms were explored: backpropagation neural networks and random forest (RF) regression. The best-performing RF model was then applied to estimate seven land and atmospheric variables globally from VIIRS TOA data: the leaf area index (LAI), incident photosynthetically active radiation (PAR), fraction of absorbed photosynthetically active radiation (FAPAR), incident shortwave radiation (ISR), land surface albedo, land surface reflectance, and TOA albedo. The inversion results were validated using ground measurements at 54 sites and showed comparable accuracy to the numerical inversion scheme that is suitable for only small regions. An experiment was also conducted to test this method on a much larger region for detailed evaluation. Finally, seven global estimates in 2013 were experimentally produced. Inter-comparison with the existing satellite products showed that the VIIRS estimates had high consistency with the LAI and FAPAR products of the Global LAnd Surface Satellite (GLASS) products suite. Moreover, the VIIRS officially released the VNP43MA3 albedo and VNP09A1 surface reflectance products, the retrieved estimates also matched them well. It was also shown that simultaneous estimation of all seven variables was more accurate and efficient than estimating individual variable sequentially. • A method to replace the coupled RT model by the machine learning method for estimating multiple variables simultaneously. • The training dataset was generated by reversely simulating using the coupled RT model from TOA reflectance. • Simultaneous estimation of multiple variables is more accurate and efficient than estimating individual variable sequentially • The generated multiple products have high accuracy based on extensive validation. [ABSTRACT FROM AUTHOR]

Leaf area index (LAI) is a terrestrial essential climate variable that is required in a variety of ecosystem and climate models. The Global LAnd Surface Satellite (GLASS) LAI product has been widely used, but its current version (V5) from Moderate Resolution Imaging Spectroradiometer (MODIS) data has several limitations, such as frequent temporal fluctuation, large data gaps, high dependence on the quality of surface reflectance, and low computational efficiency. To address these issues, this paper presents a deep learning model to generate a new version of the LAI product (V6) at 250-m resolution from MODIS data from 2000 onward. Unlike most existing algorithms that estimate one LAI value at one time for each pixel, this model estimates LAI for 2 years simultaneously. Three widely used LAI products (MODIS C6, GLASS V5, and PROBA-V V1) are used to generate global representative time-series LAI training samples using K-means clustering analysis and least difference criteria. We explore four machine learning models, the general regression neural network (GRNN), long short-term memory (LSTM), gated recurrent unit (GRU), and Bidirectional LSTM (Bi-LSTM), and identify Bi-LSTM as the best model for product generation. This new product is directly validated using 79 high-resolution LAI reference maps from three in situ observation networks. The results show that GLASS V6 LAI achieves higher accuracy, with a root mean square (RMSE) of 0.92 at 250 m and 0.86 at 500 m, while the RMSE is 0.98 for PROBA-V at 300 m, 1.08 for GLASS V5, and 0.95 for MODIS C6 both at 500 m. Spatial and temporal consistency analyses also demonstrate that the GLASS V6 LAI product is more spatiotemporally continuous and has higher quality in terms of presenting more realistic temporal LAI dynamics when the surface reflectance is absent for a long period owing to persistent cloud/aerosol contaminations. The results indicate that the new Bi-LSTM deep learning model runs significantly faster than the GLASS V5 algorithm, avoids the reconstruction of surface reflectance data, and is resistant to the noises (cloud and snow contamination) or missing values contained in surface reflectance than other methods, as the Bi-LSTM can effectively extract information across the entire time series of surface reflectance rather than a single time point. To our knowledge, this is the first global time-series LAI product at the 250-m spatial resolution that is freely available to the public (www.geodata.cn and www.glass.umd.edu). • Exploring a deep learning model for global LAI estimation based on temporal and spectral signatures of MODIS reflectance. • Generating the first global 21-year LAI product at 250-m spatial resolution. • The retrieved LAI maintains spatio-temporal consistency even when the high-quality reflectance is absent for a long period. • The learning model is very efficient and requires no smoothing and gap-filling of surface reflectance. [ABSTRACT FROM AUTHOR]

Various satellite sensors have provided a huge amount of observations of Earth's environment at variable spatial and temporal resolutions. Many global coarse-resolution land products have been generated from single-satellite data, but global temporally regular land products at fine spatial resolutions (say 10-30 m) are scarce because of infrequent observations. An ideal inversion framework can estimate multiple global continuous land variables at different spatial resolutions by combining all sources of satellite data. This paper proposes a new framework that can estimate five land variables simultaneously from the top-of-atmosphere (TOA) reflectance acquired by seven satellite sensors based on a multi-scale and multi-depth convolutional neural network (MSDCNN). This framework enables us to estimate temporally regular land variables at as fine as 10 m spatial resolution by transforming information from satellite data at coarser spatial resolutions. The estimated land variables include Leaf Area Index (LAI), Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), shortwave albedo, visible albedo, and spectral reflectance. Seven sensors that acquire data at different spatial resolutions, include Visible Infrared Imaging Radiometer Suite (VIIRS) (750 m), Moderate Resolution Imaging Spectroradiometer (MODIS) (500 m), Fengyun-3 (FY-3) Medium Resolution Spectral Imager (MERSI) (250 m), China-Brazil Earth Resources Satellite 04 (CBERS-04) Wide Field Imager (WFI) (73 m), Landsat 8 Operational Land Imager (OLI) (30 m), Gaofen-1 (GF-1) Wide-Field-of-View (16 m) and Sentinel 2 A/B Multispectral Imager (MSI) (10 m). This framework is mainly composed of four steps. First, a Shuffled Complex Evolution (SCE) optimization method is adopted to estimate these variables from VIIRS TOA reference. Second, a joint-output random forest regression (RF) method is used to link the satellite observations and the estimated values from step 1. Third, the six-type multi-resolution satellite observations are used to downscale the VIIRS TOA reflectance to six different spatial resolutions by using the MSDCNN. Finally, the downscaled six-resolution VIIRS TOA reflectance is fed into the multiple-variable RF model to estimate land variables. The framework was validated, and the results based on high-resolution reference maps from ImagineS network and time series shortwave albedo field values from Surface Radiation (SURFRAD) and Integrated Carbon Observation System (ICOS) network show that the retrieved variables had high validation accuracy, with root mean square error (RMSE) ranges of 0.361–0.489 (LAI), 0.023–0.120 (FAPAR), 0.013–0.026 (snow-free shortwave albedo), respectively. Comparison results of the retrieved multi-scale variables with the Sentinel 2 A/B (10 m), Landsat 8 (30 m), Global LAnd Surface Satellite (GLASS, 250 m, 500 m), MODIS (500 m), and VIIRS (750 m) products show that their values were close, with RMSE ranges of 0.107–0.273 (LAI), 0.015–0.027 (FAPAR), 0.003–0.007 (shortwave albedo), 0.001–0.007 (visible albedo), and 0.003–0.025 (spectral reflectance), respectively. The results of the direct validation as well as the product intercomparison show that this novel framework has the potential to be used in estimating global land variables at various spatial scales using a variety of satellite data sources. • Exploring a deep learning method for seven-resolution land surface variable estimation. • The finest resolution of estimated variable is 10 m and the temporal continuity is improved. • Estimated variables at seven resolution have consistent spatial distribution. • The estimated multi-resolution variables have high accuracy based on extensive validation. [ABSTRACT FROM AUTHOR]