LAI Products Based on MODIS Data
The GLASS LAI product ((Version 6) from MODIS surface reflectance data is developed using the bidirectional long short-term memory (Bi-LSTM) model. The model takes advantage of the existing global LAI products and the temporal and spectral information of MODIS surface reflectance effectively. Validation against field references shows that the GLASS V6 LAI product at both 250-m and 500-m resolutions is more accurate than other products. Besides, GLASS V6 LAI maintains spatio-temporal consistency even when the high-quality reflectance is absent for a long period. Besides, it captures the vegetation growing cycles and abrupt surface changes well. The GLASS 250-m LAI is the first global 21-year LAI product at 250-m spatial resolution. The algorithm for the Leaf Area Index (LAI) product is designed to estimate the fraction of absorbed photosynthetically active radiation (FAPAR) using a combination of satellite data inputs and deep learning models. The LAI product is generated from Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance data and other sources as part of the Global Land Surface Satellite (GLASS) product suite. A significant component of this algorithm is the use of a bidirectional long short-term memory (Bi-LSTM) deep learning model, which utilizes time-series training samples derived from three existing global FAPAR products—MODIS Collection 6, GLASS V5, and PROBA-V V1. This model enhances the spatial and temporal continuity of the FAPAR estimates. The improved FAPAR product offers detailed, high-resolution (250 m) vegetation data that is crucial for ecological monitoring and carbon cycle modeling (Ma et al.,2022). The 2022-2023 LAI product will be provided upon request.
LAI Products Based on AVHRR Data
The LAI algorithm for AVHRR data employs General Regression Neural Networks (GRNNs) to derive LAI from time-series reflectance data. This method utilizes preprocessed reflectance from the red and near-infrared bands over a year, training the GRNNs with smoothed LAI values from existing products (MOD15 and CYCLOPES). The algorithm alternates between two GRNN groups to ensure temporal continuity, generating LAI estimates which are combined through a weighted linear approach, thereby producing high-quality, continuous global LAI profiles (Xiao et al.,2022).
References：
[1] Ma, H., & Liang, S. (2022). Development of the GLASS 250-m leaf area index product (version 6) from MODIS data using the bidirectional LSTM deep learning model. Remote Sensing of Environment. [Download]
[2] Xiao, Z., Liang, S., Wang, J., Xiang, Y., Zhao, X., & Song, J. (2016). Long-time-series global land surface satellite leaf area index product derived from MODIS and AVHRR surface reflectance. IEEE Transactions on Geoscience and Remote Sensing, 54(9), 5301-5318. [Download]
Albedo Products Based on MODIS and AVHRR Data
The GLASS Broadband Albedo products include three spectral ranges: total shortwave, visible and near-IR under actual atmospheric conditions (so-called blue-sky albedos). The GLASS Broadband Albedo product is based on the integration of two algorithms through a temporal filter scheme (Liu et al. 2013a; Liu et al. 2013b; Qu et al. 2014). One algorithm is based on the surface reflectance that is converted from the top-of-atmosphere (TOA) radiance through atmospheric correction, and another algorithm is based on the estimated surface albedo directly from TOA observations without atmospheric correction. These algorithms are robust and have been applied to many other satellite observations. Recent efforts have extended to include sea ice (Qu et al. 2016) and ocean water(Feng et al. 2016). The GLASS Broadband Albedo product algorithm employs a methodology that combines direct inversion, primary product merging, and downscaling fusion techniques. Initially, albedo values are derived from Moderate Resolution Imaging Spectroradiometer (MODIS) data, which provides textural details at a spatial resolution of 250 meters. These initial values are then refined by merging them with the GLASS albedo product, which has a spatial resolution of 1 km. This merging process involves a statistical temporal filtering technique to maintain spatial and temporal continuity. To enhance the spatial resolution of the albedo product to 250 meters, the algorithm fuses detailed texture information from MODIS data with broader-scale albedo estimates from the GLASS product. This fusion process is designed to retain the high quality and accuracy of the GLASS product while providing detailed resolution necessary for local environmental and climate studies (Lu et al., 2020).
From 2001 to 2020, the GLASS albedo products were generated using MODIS and AVHRR observation data, available at spatial resolutions of 0.05°, 500 m, and 250 m, with a temporal resolution of 4 days. From 1981 to 2000, the GLASS albedo products were generated using AVHRR observation data. The 0.05° and 500 m products for all periods include black-sky albedo and white-sky albedo for three broad bands: visible, near-infrared, and shortwave. To save storage resources, the 250 m resolution value-added product only provides clear-sky shortwave albedo, which is a combination of black-sky and white-sky albedo. Under clear sky conditions, the physical meaning of this parameter is consistent with conventional albedo surface measurements.
References：
[1] Liu, N., Liu, Q., Wang, L., Liang, S., Wen, J., Qu, Y., & Liu, S. (2013a). A statistics-based temporal filter algorithm to map spatiotemporally continuous shortwave albedo from MODIS data. Hydrology and Earth System Sciences, 17, 2121-2129, doi:2110.5194/hess-2117-2121-2013. [Download]
[2] Liu, Q., Wang, L., Qu, Y., Liu, N., Liu, S., Tang, H., & Liang, S. (2013b). Priminary Evaluation of the Long-Term GLASS Albedo Product. International Journal of Digital Earth, 6, 69-95,doi:10.1080/17538947.17532013.17804601. [Download]
[3] Qu, Y., Liu, Q., Liang, S., Wang, L., Liu, N., & Liu, S. (2014). Direct-estimation algorithm for mapping daily land-surface broadband albedo from MODIS data. IEEE Transactions on Geoscience and Remote Sensing, 52, 907-919. [Download]
[4] Qu, Y., Liang, S., Liu, Q., Li, X., Feng, Y., & Liu, S. (2016). Estimating Arctic sea-ice shortwave albedo from MODIS data. Remote Sensing of Environment, 186, 32-46. [Download]
[5] Feng, Y., Liu, Q., Qu, Y., & Liang, S. (2016). Estimation of the Ocean Water Albedo From Remote Sensing and Meteorological Reanalysis Data. IEEE Transactions on Geoscience and Remote Sensing, 54, 850-868. [Download]
[6] Lu Y R, Liu Q, Li X, Li X H, Liu L, Xiao S and Sun M Y. 2020. An algorithm for producing 250 m global albedo product and validation. Journal of Geo-Information Science, 22(2): 328-335 (陆彦 蓉, 刘强, 李霞, 李秀红, 刘璐, 肖洒, 孙美莹 . 2020. 全球 250 m 反照率产品算法及验证 . 地球信息科学学报, 22(2): 328-335) [DOI: 10.12082/dqxxkx.2020.190184]. [Download]
The GLASS BBE products are based on two different algorithms. The surface is classified into several types: water, ice/snow, bare soil, vegetation, and transition zones, with BBE estimated separately for each type.
BBE Products Based on MODIS Data
In the GLASS MODIS BBE product algorithm, the BBE for bare soil is estimated based on the linear relationship between BBE calculated from ASTER spectral emissivity and MODIS seven-band shortwave albedo (Cheng and Liang, 2014). The physical basis of this relationship is further validated through radiative transfer simulations (Cheng et al., 2018). The BBE for vegetation is obtained by interpolating lookup tables constructed from canopy radiative transfer models, with input parameters including leaf BBE, soil BBE, and vegetation LAI (Cheng et al., 2016). The BBE for transition zones is the average of the BBE for bare soil and vegetation. The BBE for ice/snow and water bodies is set as a constant, derived from a combination of radiative transfer simulation results and spectral library/field measurements (Cheng et al., 2010).
BBE Products Based on AVHRR Data
The GLASS AVHRR BBE product is generated using a similar algorithm, with the key difference being the use of AVHRR surface visible and near-infrared reflectance instead of MODIS spectral albedo (Cheng and Liang, 2013).
References：
[1] Cheng, J., & Liang, S. (2014). Estimating the broadband longwave emissivity of global bare soil from the MODIS shortwave albedo product. Journal of Geophysical Research: Atmospheres, 119, 614-634. [Download]
[2] Cheng, J., Liang, S.L., Nie, A.X., & Liu, Q. (2018). Is There a Physical Linkage Between Surface Emissive and Reflective Variables Over Non-Vegetated Surfaces? Journal of the Indian Society of Remote Sensing, 46, 591-596. [Download]
[3] Cheng, J., & Liang, S. (2013). Estimating global land surface broadband thermal-infrared emissivity from the Advanced Very High Resolution Radiometer optical data. International Journal of Digital Earth, DOI: 10.1080/17538947.17532013.17783129. [Download]
[4] Cheng, J., Liang, S., Verhoef, W., Shi, L., & Liu, Q. (2016). Estimating the Hemispherical Broadband Longwave Emissivity of Global Vegetated Surfaces Using a Radiative Transfer Model. IEEE Transactions on Geoscience and Remote Sensing, 54, 905-917. [Download]
[5] Cheng, J., Liang, S., Weng, F., Wang, J., & Li, X. (2010). Comparison of Radiative Transfer Models for Simulating Snow Surface Thermal Infrared Emissivity. IEEE Journal in Special Topics in Applied Earth Observations and Remote Sensing, 3, 323-336. [Download]
PAR Products Based on MODIS Data
The PAR component of GLASS products is derived using a combination of data from multiple polar-orbiting and geostationary satellites. This approach leverages a refined look-up table (LUT) method that considers atmospheric water vapor, surface bi-directional reflectance distribution functions (BRDF), and surface elevation. This method facilitates the estimation of PAR at a high spatial resolution of 5 km and a temporal resolution of 3 hours. The model is validated with ground-based measurements, ensuring its reliability for applications in ecological and climate modeling. The validation shows high accuracy, making it a valuable tool for understanding and analyzing vegetation dynamics and photosynthetic productivity globally (Zhang et al., 2014).
[1] Zhang, X., Liang, S., Zhou, G., Wu, H., & Zhao, X. (2014). Generating Global Land Surface Satellite incident shortwave radiation and photosynthetically active radiation products from multiple satellite data. Remote Sensing of Environment, 152, 318-332. [Download]
FAPAR Products Based on MODIS Data
The GLASS fraction of absorbed photosynthetically active radiation (FAPAR) is defined as the black-sky FAPAR around 10:30 local solar time, which is an approximation of the daily average FAPAR. GLASS FAPAR is estimated from MODIS surface reflectance data and GLASS LAI data based on a Bi-LSTM model, which involves high-quality training samples and combines the strengths of the existing FAPAR products, as well as the temporal and spectral information from the MODIS surface reflectance data and other information (Ma et al., 2022). The 2022-2023 FAPAR product will be provided upon request.
FAPAR Products Based on AVHRR Data
The GLASS FAPAR product is derived from the GLASS LAI product, which is generated using a combination of MODIS and AVHRR surface reflectance data. The FAPAR algorithm employed in the GLASS product ensures physical consistency between the LAI and FAPAR retrievals by utilizing an approximate FAPAR calculation scheme that primarily considers the transmittance of PAR down to the soil, as expressed by the equation FAPAR = 1 - τPAR, where τPAR represents the transmittance of PAR. This method accounts for both direct and diffuse PAR and incorporates parameters such as the clumping index and solar zenith angle to estimate canopy transmittance. The GLASS FAPAR product, thus, provides a spatially complete and temporally continuous record of FAPAR from 2000 onwards, offering a more accurate and high-quality alternative for various applications compared to existing FAPAR products (Xiao et al., 2015).
References：
[1] Ma, H., Liang, S., Xiong, C., Wang, Q., Jia, A., & Li, B. (2022). Global land surface 250 m 8 d fraction of absorbed photosynthetically active radiation (FAPAR) product from 2000 to 2021. Earth System Science Data, 14(12), 5333-5347. [Download]
[2] Xiao, Z., Liang, S., Sun, R., Wang, J., & Jiang, B. (2015). Estimating the fraction of absorbed photosynthetically active radiation from the MODIS data based GLASS leaf area index product. Remote Sensing of Environment, 171, 105-117. [Download]
The GLASS DSR products are primarily produced using a hybrid algorithm based on MODIS data (Zhang et al., 2019) and a lookup table algorithm based on AVHRR data (Zhang et al., 2014).
DSR Products Based on MODIS Data
The hybrid algorithm based on MODIS data consists of two main parts: first, the surface shortwave net radiation is estimated using a direct estimation algorithm (Wang et al., 2015a) with Terra and Aqua satellite data. The main idea is to calculate TOA reflectance through radiative transfer models under different observation geometries and establish a linear regression relationship between TOA reflectance and surface shortwave net radiation under various atmospheric conditions. The coefficients obtained from the linear regression are then used to estimate the surface shortwave net radiation. Second, the obtained shortwave net radiation data are projected to a 5 km spatial resolution, and combined with GLASS broadband albedo data to produce spatially continuous daily mean surface shortwave radiation (Zhang et al., 2019).
DSR Products Based on AVHRR Data
The process of the lookup table algorithm based on AVHRR data is as follows: Step 1 involves directly using land surface reflectance products or other methods to obtain land surface reflectance data. Using the obtained reflectance, the TOA radiance values are calculated for all atmospheric conditions, ranging from the clearest to the cloudiest skies. Step 2 uses the lookup table algorithm to establish the relationship between the atmospheric condition index and TOA radiance. The atmospheric condition index is then determined by combining the obtained TOA radiance values. Step 3 uses the lookup table algorithm to establish the relationship between surface radiation flux and the atmospheric condition index. The DSR is further calculated based on the obtained atmospheric condition index (Zhang et al., 2014). A machine learning algorithm has been developed to estimate DSR from AVHRR data (Yang et al. 2018).
References：
[1] Zhang, X., Wang, D., Liu, Q., Yao, Y., Jia, K., He, T., Jiang, B., Wei, Y., Ma, H., & Zhao, X. (2019). An operational approach for generating the global land surface downward shortwave radiation product from MODIS data. IEEE Transactions on Geoscience and Remote Sensing, 57(7), 4636-4650. [Download]
[2] Zhang, X., Liang, S., Zhou, G., Wu, H., & Zhao, X. (2014). Generating Global Land Surface Satellite incident shortwave radiation and photosynthetically active radiation products from multiple satellite data. Remote Sensing of Environment, 152, 318-332. [Download]
[3] Wang, D., Liang, S., He, T., & Shi, Q. (2015). Estimation of daily surface shortwave net radiation from the combined MODIS data. IEEE Transactions on Geoscience and Remote Sensing, 53(10), 5519-5529. [Download]
[4] Yang, L., Zhang, X., Liang, S., Yao, Y., Jia, K., & Jia, A. (2018). Estimating Surface Downward Shortwave Radiation over China Based on the Gradient Boosting Decision Tree Method. Remote Sensing, 10, 185. [Download]
The GLASS Surface Long-Wave Radiation (LWR) products include surface UPward Long_Wave Radiation (LWUP), surface DowNward Long-Wave Radiation (LWDN), and Long-Wave Net Radiation (LWNT). These are further divided into longwave radiation products based on MODIS data and daily mean longwave radiation products based on AVHRR data.
LWR Products Based on MODIS Data
Under the general framework of the hybrid method, we developed linear and dynamic learning neural network (DLNN) models for estimating the global 1-km instantaneous clear-sky long-wave upwelling radiation (LWUP) from the top-of-atmosphere radiance of MODIS TIR channels 29, 31, and 32(Cheng and Liang 2016). At the same time, we developed an efficient hybrid method for estimating 1 km instantaneous clear-sky LWDN from MODIS TIR observations and the MODIS near-infrared column water vapor (CWV) data product(Cheng et al. 2017). The LWDN was formulated as a nonlinear function of LWUP estimated from the MODIS TOA radiance of channels 29, 31, and 32, as well as CWV and the MODIS TOA radiance of channel 29. Because the LWDN is overestimated over high-elevation area with extremely high CWV, we developed a power function relating LWDN to CWV as a complementary method. Regarding the longwave radiation at cloudy-sky, we estimated the LWDN using the single layer cloud model from MODIS cloud product; the cloud-sky LWUP was calculated from LST in MOD06/MYD06 and GLASS BBE product. Before 2000, we adopted the parameterization schemes to calculate clear-sky LWDN from reanalysis data (Guo et al. 2019), and using GLASS LST and BBE product to calculate clear sky LWUP. The bias of net longwave radiation (LWNT), derived from the difference between LWUP and LWDN, is 0.70 W/m², with a root mean square error of 26.7 W/m², which is significantly more accurate than other similar products (Zeng et al., 2020).
LWR Products Based on AVHRR Data
For all-sky longwave radiation, a densely connected convolutional neural network model was developed (Xu et al., 2022a) to produce the GLASS AVHRR spatiotemporally continuous daily mean longwave radiation dataset(Xu et al., 2022b). Specifically, using longwave radiation data from CERES, ERA5, and GLASS MODIS, a deep neural network model driven by longwave radiation samples fused based on the Random Forest (RF) algorithm was constructed. The model's primary inputs include AVHRR top-of-atmosphere shortwave reflectance (Zhan and Liang, 2022), thermal infrared brightness temperature observations, solar-observation geometry information, and near-surface meteorological data from ERA5 (temperature, water vapor, relative humidity). Shortwave observations are particularly helpful for estimating all-sky longwave radiation under high atmospheric water vapor and liquid water cloud content. Transfer learning was effectively applied to address the saturation effect of longwave radiation under high atmospheric water vapor conditions. Independent validation results indicate that the neural network model achieves high accuracy. Compared to CERES, ERA5, and GLASS MODIS longwave radiation products, the GLASS AVHRR longwave radiation data exhibit higher accuracy under the same ground station measurement validation conditions. The biases and root mean square errors (RMSE) for LWDN are -2.65 W/m² and 19.08 W/m², respectively; for LWUP, the biases and RMSE are -3.70 W/m² and 15.80 W/m², respectively; and for LWNT, the biases and RMSE are 0.49 W/m² and 16.29 W/m², respectively. In comparison, the RMSE ranges for the other three datasets are 20.95–27.82 W/m² (LWDN), 17.46–17.93 W/m² (LWUP), and 18.32–25.97 W/m² (LWNT).
References：
[1] Cheng, J., & Liang, S. (2016). Global Estimates for High-Spatial-Resolution Clear-Sky Land Surface Upwelling Longwave Radiation From MODIS Data. IEEE Transactions on Geoscience and Remote Sensing, 54, 4115-4129. [Download]
[2] Cheng, J., Liang, S., Wang, W., & Guo, Y. (2017). An efficient hybrid method for estimating clear-sky surface downward longwave radiation from MODIS data. Journal of Geophysical Research: Atmospheres, 122, 2616-2630. [Download]
[3] Zeng, Q., Cheng, J., & Dong, L. (2020). Assessment of the long-term high-spatial-resolution Global LAnd Surface Satellite (GLASS) surface longwave radiation product using ground measurements. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13, 2032-2055. [Download]
[4] Guo, Y., Cheng, J., & Liang, S. (2019). Comprehensive assessment of parameterization methods for estimating clear-sky surface downward longwave radiation. Theoretical and Applied Climatology, 1-14. [Download]
[5] Xu, J., Liang, S., & Jiang, B. (2022a). A global long-term (1981–2019) daily land surface radiation budget product from AVHRR satellite data using a residual convolutional neural network. Earth System Science Data, 14(5), 2315-2341. [Download]
[6] Xu, J., Liang, S., Ma, H., & He, T. (2022b). Generating 5 km resolution 1981–2018 daily global land surface longwave radiation products from AVHRR shortwave and longwave observations using densely connected convolutional neural networks. Remote Sensing of Environment, 280, 113223. [Download]
[7] Zhan, C., & Liang, S. (2022). Improved estimation of the global top-of-atmosphere albedo from AVHRR data. Remote Sensing of Environment, 269, 112836. [Download]
NR Products Based on MODIS Data
The GLASS NR products based on MODIS data employ deep learning methods to directly estimate net radiation from MODIS TOA band data (Chen et al., 2020). The GLASS NR product utilizes a hybrid machine learning approach, integrating a genetic algorithm with an artificial neural network (GA-ANN), to estimate daily average all-wave net radiation (NR) from MODIS data at high latitudes. This method addresses the limitations of traditional algorithms, such as greater uncertainty in surface and atmospheric satellite products at high latitudes and the lack of real-time cloud base height and temperature parameters. By employing the GA-ANN, the model optimizes weight and bias parameters, enhancing the accuracy of NR estimation without the need for auxiliary data.
NR Products Based on AVHRR Data
In contrast, the GLASS NR products based on AVHRR data use corrected ERA5 data to fill gaps, resulting in global, spatiotemporally continuous GLASS AVHRR daily net radiation products due to the limitations of available data sources. Recently, a deep learning method has also been developed to directly produce daily net radiation products from AVHRR TOA observation data (Xu et al., 2022).
The GLASS Net Radiation product is based on a relationship between all-wave net radiation and incident shortwave radiation in conjunction with other information (Jiang et al. 2018). The earlier version of the algorithm focused on day-time net radiation using a linear relationship (Jiang et al. 2015). After comprehensive evaluation of different machine learning techniques for the nonlinear relationship (Jiang et al. 2014), the multivariate adaptive regression splines (MARS) model was selected as the GLASS net radiation product algorithm.
References：
[1] Chen, J., He, T., Jiang, B., & Liang, S. (2020). Estimation of all-sky all-wave daily net radiation at high latitudes from MODIS data. Remote Sensing of Environment, 245, 111842. [Download]
[2] Xu, J., Liang, S., & Jiang, B. (2022). A global long-term (1981–2019) daily land surface radiation budget product from AVHRR satellite data using a residual convolutional neural network. Earth System Science Data, 14(5), 2315-2341. [Download]
[3] Jiang, B., Liang, S., Jia, A., Xu, J., Zhang, X., Xiao, Z., Zhao, X., Jia, K., & Yao, Y. (2018). Validation of the Surface Daytime Net Radiation Product From Version 4.0 GLASS Product Suite. Ieee Geoscience and Remote Sensing Letters, 1-5. [Download]
[4] Jiang, B., Zhang, Y., Liang, S., Wohlfahrt, G., Arain, A., Cescatti, A., Georgiadis, T., Jia, K., Kiely, G., Lund, M., Montagnani, L., Magliulo, V., Ortiz, P.S., Oechel, W., Vaccari, F.P., Yao, Y., & Zhang, X. (2015). Empirical estimation of daytime net radiation from shortwave radiation and ancillary information. Agricultural and Forest Meteorology, 211–212, 23-36. [Download]
[5] Jiang, B., Zhang, Y., Liang, S., Zhang, X., & Xiao, Z. (2014). Surface Daytime Net Radiation Estimation Using Artificial Neural Networks. Remote Sensing, 6, 11031-11050. [Download]
LST Products Based on MODIS Data
The GLASS all-weather LST products are derived from MODIS observation data, model data (GLDAS surface temperature, ERA5-land surface temperature), other GLASS satellite products (such as downward longwave radiation, DSR, albedo, LAI, angular information including observation zenith angle, solar zenith angle, and relative azimuth angle), and ground station observations. Machine learning algorithms were utilized to conduct experiments across the continental United States and globally, achieving spatiotemporally continuous, all-weather instantaneous and daily mean surface temperature estimates (Li et al., 2021a). Validation results using independent datasets indicate that the RMSE of the instantaneous model is approximately 2.7 K in both the continental US and global experiments, while the RMSE of the global daily mean surface temperature model is around 2.1 K. Compared to the MODIS official LST product, the GLASS all-weather instantaneous LST exhibits similar spatial distribution patterns and higher validation accuracy, filling in missing pixels and correcting LST anomalies caused by cloud misclassification, with particularly notable improvements in high-latitude regions. When comparing MODIS instantaneous LST images to the corresponding GLASS LST images over four days in 2010 (representing spring, summer, fall, and winter), it is evident that MODIS LST has substantial data gaps, while GLASS LST is spatially continuous. The GLASS daily mean LST product, an all-weather daily mean LST satellite product based on MODIS data, has a unified time reference and will play a crucial role in research areas such as climate change, agricultural monitoring, and drought monitoring.
LST Products Based on AVHRR Data
The GLASS product suite currently includes two sets of instantaneous Land Surface Temperature (LST) products based on AVHRR data. The first set uses a multi-algorithm ensemble approach to address the low accuracy of individual retrieval algorithms under large viewing angles and high water vapor content. This ensemble model integrates nine common split-window algorithms to construct the LST multi-algorithm ensemble retrieval model (Zhou et al., 2019; Ma et al., 2020). This LST product has been validated against ground station measurements. Preliminary validation results indicate that at the Baseline Surface Radiation Network (BSRN) Barrow station, the LST product has an RMSE of 2.89 K (Zhou et al., 2019). For six SURFRAD stations during 1995-2000, the product shows an average bias of 0.21 K and an average standard deviation of 2.48 K, with ranges of -1.59 to 2.71 K and 2.26 to 2.76 K, respectively (Ma et al., 2020). The second set of AVHRR LST products is based on an improved generalized split-window algorithm (Liu et al., 2019). This algorithm enhances the original split-window algorithm by adding quadratic terms of the brightness temperature differences between the two thermal infrared channels, thereby improving retrieval accuracy under high water vapor content conditions. Additionally, in the production process, to address the scarcity of ancillary data before 2000, the algorithm employs a vegetation index threshold method and a covariance/variance ratio method to estimate emissivity and atmospheric water vapor content, respectively, tailored to AVHRR data characteristics. Initial surface temperature is used instead of near-surface air temperature to determine coefficient groups, enabling the production of a global LST product entirely based on AVHRR data. Validation using direct temperature-based methods at six SURFRAD stations for NOAA-14 (1995-2000) indicates an RMSE of 2.2-4.1 K and an MBE of -0.4-2.0 K (Liu et al., 2019).
References：
[1] Li, B., Liang, S., Liu, X., Ma, H., Chen, Y., Liang, T., & He, T. (2021). Estimation of all-sky 1 km land surface temperature over the conterminous United States. Remote Sensing of Environment, 266, 112707. [Download]
[2] Zhou, J., Liang, S., Cheng, J., Wang, Y., & Ma, J. (2019). The GLASS Land Surface Temperature Product. IEEE Journal in Special Topics in Applied Earth Observations and Remote Sensing, 12, 10.1109/JSTARS.2018.2870130. [Download]
[3] Ma, J., Zhou, J., Göttsche, F. M., Liang, S., Wang, S., & Li, M. (2020). A global long-term (1981–2000) land surface temperature product for NOAA AVHRR. Earth System Science Data, 12(4), 3247-3268. [Download]
[4] Liu, X., Tang, B. H., Yan, G., Li, Z. L., & Liang, S. (2019). Retrieval of global orbit drift corrected land surface temperature from long-term AVHRR data. Remote Sensing, 11(23), 2843. [Download]
All-Weather/Gap-Free Land Surface Temperature Products Combining VIIRS and MODIS Data
All-Weather/Gap-Free Land Surface Temperature retrieval method is a novel method for estimating land surface temperature (LST) under cloudy conditions using data from the Visible Infrared Imaging Radiometer Suite (VIIRS) and the Moderate Resolution Imaging Spectroradiometer (MODIS). This method is based on the surface energy balance (SEB) principle and involves two major steps:
The method was applied to VIIRS and MODIS data and validated using ground measurements from various sites, showing promising results in terms of accuracy and the ability to capture realistic variability without abrupt discontinuities. It performed well across different sensors, seasons, and land cover types, and demonstrated an improvement over previous methods for cloudy-sky LST estimation. The overall conclusion is that this approach could be applied to similar satellite sensors for global real-time production of LST data, which is crucial for applications in climate monitoring, hydrological modeling, and environmental studies (Jia et al., 2022; Jia et al., 2023).
References：
[1] Jia, A., Liang, S., & Wang, D. (2022). Generating a 2-km, all-sky, hourly land surface temperature product from Advanced Baseline Imager data. Remote Sensing of Environment, 278, 113105. https://doi.org/10.1016/j.rse.2022.113105. [Download]
[2] Jia, A., Ma, H., Liang, S., & Wang, D. (2021). Cloudy-sky land surface temperature from VIIRS and MODIS satellite data using a surface energy balance-based method. Remote Sensing of Environment, 263, 112566. [Download]
[3] Jia, A., Liang, S., Wang, D., Ma, L., Wang, Z., & Xu, S. (2023). Global hourly, 5 km, all-sky land surface temperature data from 2011 to 2021 based on integrating geostationary and polar-orbiting satellite data. Earth System Science Data, 15(2), 869-895. [Download]
Land Surface Air Temperature Products Based on MODIS Data
The Land Surface Air Temperature retrieval algorithm for MODIS data introduced creates a high-resolution, daily mean Land surface air temperature product for mainland China covering 2003 to 2019. Utilizing MODIS thermal imagery and GLDAS data, this method accurately estimates temperatures under various sky conditions using a random forest approach (Chen et al., 2021). The algorithm incorporates data preprocessing for consistency and employs gap-filling techniques for cloud-covered periods. Validated against measurements from 2384 stations, the models demonstrate high accuracy with an R^2 consistently above 0.98, proving more reliable than existing reanalysis products.
References：
[1] Chen, Y., Liang, S., Ma, H., Li, B., He, T., & Wang, Q. (2021). An all-sky 1 km daily surface air temperature product over mainland China for 2003–2019 from MODIS and ancillary data. Earth System Science Data Discussions, 2021, 1-35. [Download]
FVC Products Based on MODIS Data
The GLASS MODIS FVC product algorithm is based on the GRNN model (Jia et al., 2015). However, the computational efficiency was unsatisfactory in the global FVC product production process. After comparing various machine learning methods, the MARS method, which offers higher computational efficiency and reasonable accuracy, was ultimately chosen (Yang et al., 2016). Extensive validation experiments have been conducted using the estimates from high-resolution satellite data and ground measurements (Jia et al. 2016; Jia et al. 2018). The details of the algorithms and the validation results are recently summarized (Jia et al. 2019).
FVC Products Based on AVHRR Data
The GLASS AVHRR FVC product is based on AVHRR data (Jia et al., 2019). The GLASS FVC algorithm for AVHRR data was also developed to be in concert with the GLASS MODIS FVC product. It was based on the GLASS MODIS FVC product to achieve continuity of FVC estimates from both AVHRR and MODIS data. First, a global training sample set of AVHRR reflectance and GLASS MODIS FVC is constructed, supported by global sampling points over an entire year. Then, the MARS model is trained using this sample set to develop an algorithm for estimating vegetation cover based on AVHRR reflectance data. Finally, the FVC estimated from AVHRR data is linearly corrected using GLASS MODIS FVC, resulting in an AVHRR FVC product consistent with GLASS MODIS FVC.
References：
[1] Jia, K., Liang, S., Liu, S.H., Li, Y.W., Xiao, Z.Q., Yao, Y.J., Jiang, B., Zhao, X., Wang, X.X., Xu, S., & Cui, J. (2015). Global Land Surface Fractional Vegetation Cover Estimation Using General Regression Neural Networks From MODIS Surface Reflectance. IEEE Transactions on Geoscience and Remote Sensing, 53, 4787-4796. [Download]
[2] Jia, K., Liang, S., Gu, X., Baret, F., Wei, X., Wang, X., Yao, Y., Yang, L., & Li, Y. (2016). Fractional vegetation cover estimation algorithm for Chinese GF-1 wide field view data. Remote Sensing of Environment, 177, 184-191. [Download]
[3] Jia, K., Liang, S.L., Wei, X.Q., Yao, Y.J., Yang, L.Q., Zhang, X.T., & Liu, D.Y. (2018). Validation of Global LAnd Surface Satellite (GLASS) fractional vegetation cover product from MODIS data in an agricultural region. Remote Sensing Letters, 9, 847-856. [Download]
[4] Jia, K., Yang, L., Liang, S., Xiao, Z., Zhao, X., Yao, Y., Zhang, X., Jiang, B., & Liu, D. (2019). Long-Term Global Land Surface Satellite (GLASS) Fractional Vegetation Cover Product Derived From MODIS and AVHRR Data. Ieee Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 12, doi:10.1109/jstars.2018.2854293. [Download]
[5] Yang, L., Jia, K., Liang, S., Liu, J., & Wang, X. (2016). Comparison of Four Machine Learning Methods for Generating the GLASS Fractional Vegetation Cover Product from MODIS Data. Remote Sensing, 8, 682. [Download]
The GLASS ET product algorithm employs a multi-algorithm ensemble approach using the Bayesian Model Averaging (BMA) method and eddy covariance data from 240 global FLUXNET sites. It integrates five traditional evapotranspiration algorithms (the MOD16 ET algorithm, an improved Penman-Monteith algorithm, NASA Jet Propulsion Laboratory's Priestley-Taylor algorithm, an enhanced Priestley-Taylor algorithm, and the University of Maryland's semi-empirical Penman ET algorithm) (Yao et al., 2014). This approach reduces the uncertainty of individual algorithms and enhances the accuracy of global land surface evapotranspiration estimates. The GLASS ET product primarily includes two sets of products: MODIS ET and AVHRR ET. The main features of the product are the use of an algorithm ensemble approach, long temporal coverage, high accuracy, and good temporal and spatial consistency.
ET Products Based on MODIS Data
The GLASS product employs a modified Priestley-Taylor (PT) algorithm to estimate terrestrial Evapotranspiration (ET) using MODIS data. This approach incorporates the Normalized Difference Vegetation Index (NDVI) and Apparent Thermal Inertia (ATI) derived from temperature changes over time to enhance the accuracy of ET estimation. The algorithm operates with net radiation, air temperature, diurnal temperature range, and NDVI as primary inputs, bypassing the complexities of aerodynamic resistance parameters. It is validated using ground-based flux tower data and demonstrates improved correlation with ground-measured ET compared to the original PT-JPL model (Yao et al. 2013).
ET Products Based on AVHRR Data
AVHRR data is utilized within the GLASS product to estimate Evapotranspiration (ET) through a process-based algorithm. GLASS Evapotranspiration product algorithm is based on the multi-model ensemble method, that is the Bayesian model averaging (BMA) method which merges five process-based ET algorithms to improve ET estimate (Yao et al. 2014).
References：
[1] Yao, Y.J., Liang, S., Cheng, J., Liu, S.M., Fisher, J.B., Zhang, X.D., Jia, K., Zhao, X., Qing, Q.M., Zhao, B., Han, S.J., Zhou, G.S., Zhou, G.Y., Li, Y.L., & Zhao, S.H. (2013). MODIS-driven estimation of terrestrial latent heat flux in China based on a modified Priestley-Taylor algorithm. Agricultural and Forest Meteorology, 171, 187-202. [Download]
[2] Yao, Y., Liang, S., Li, X., Hong, Y., Fisher, J.B., Zhang, N., Chen, J., Cheng, J., Zhao, S., Zhang, X., Jiang, B., Sun, L., Jia, K., Wang, K., Chen, Y., Mu, Q., & Feng, F. (2014). Bayesian multimodel estimation of global terrestrial latent heat flux from eddy covariance, meteorological, and satellite observations. Journal of Geophysical Research: Atmospheres, 119, 2013JD020864. [Download]
GPP Products Based on MODIS Data
GLASS GPP algorithm from MODIS data originates from EC-LUE model (Eddy Covariance – Light Use Efficiency) (Yuan et al. 2007). On the basis of the theory of light use efficiency, the EC-LUE model relies on two assumptions: first, that the fraction of absorbed PAR (fPAR) is a linear function of NDVI; second, that the realized light use efficiency, calculated from a biome-independent invariant potential LUE, is controlled by air temperature or soil moisture, whichever is most limiting. The original EC-LUE is driven by only four variables: normalized difference vegetation index (NDVI), photosynthetically active radiation (PAR), air temperature, and the Bowen ratio of sensible to latent heat flux (used to calculate moisture stress) (Yuan et al., 2007). The later version of EC-LUE substituted the Bowen ratio with the ratio of evapotranspiration (ET) to net radiation, and revised the RS-PM (Remote Sensing-Penman Monteith) model for quantifying ET (Yuan et al. 2010). To accurately indicate the long-term changes of GPP, the GLASS-GPP product used the latest version of EC-LUE, which integrates the impacts of several environmental variables: atmospheric CO2 concentration, radiation components and atmospheric water vapor pressure (VPD) (Yuan et al., in preparation). The EC-LUE model has been validated widely throughout North America, Europe and East Asia by using the measurements of eddy covariance towers (Yuan et al., 2007; 2010)(Li et al. 2013; Yuan et al. 2014). These validations showed that the EC-LUE model can successfully reproduce the spatial and temporal variabilities of GPP over the various ecosystem types. Several model comparisons also indicate the better performance of EC-LUE than other LUE models. Previous study compared EC-LUE model and MODIS-GPP products based on the measurements of eddy covariance towers at southeastern China, and found the EC-LUE model performed better than the MODIS algorithms (Xu et al., 2013). A recent study compared eight satellite-based GPP models over various major grassland ecosystem types and found the EC-LUE model performed best (Jia et al., 2018).
GPP Products Based on AVHRR Data
The GPP algorithm from AVHRR data utilizes revised Eddy Covariance-Light Use Efficiency (EC-LUE) and MODIS models to assess the effects of increased Vapor Pressure Deficit (VPD) on vegetation productivity. These satellite-based models indicate that rising VPD, influenced by higher air temperatures, reduces plant photosynthetic efficiency by increasing stomatal closure, thereby reducing CO2 intake and water retention. The revised EC-LUE model integrates the impacts of atmospheric CO2 concentration and modifies the light use efficiency based on VPD and other environmental factors to estimate GPP accurately across different terrestrial ecosystems. This methodology has been vital in demonstrating how recent increases in VPD have offset the benefits of CO2 fertilization, leading to a decline in global vegetation growth (Yuan et al., 2019).
References：
[1] Yuan, W.P., Liu, S., Zhou, G.S., Zhou, G.Y., Tieszen, L.L., Baldocchi, D., Bernhofer, C., Gholz, H., Goldstein, A.H., Goulden, M.L., Hollinger, D.Y., Hu, Y., Law, B.E., Stoy, P.C., Vesala, T., Wofsy, S.C., & AmeriFlux, C. (2007). Deriving a light use efficiency model from eddy covariance flux data for predicting dailygross primary production across biomes. Agricultural and Forest Meteorology, 143, 189-207. [Download]
[2] Yuan, W.P., Liu, S.G., Yu, G.R., Bonnefond, J.M., Chen, J.Q., Davis, K., Desai, A.R., Goldstein, A.H., Gianelle, D., Rossi, F., Suyker, A.E., & Verma, S.B. (2010). Global estimates of evapotranspiration and gross primary production based on MODIS and global meteorology data. Remote Sensing of Environment, 114, 1416-1431. [Download]
[3] Li, X., Liang, S., Yu, G., Yuan, W., Cheng, X., Xia, J., Zhao, T., Feng, J., Ma, Z., Ma, M., Liu, S., Chen, J., Shao, C., Li, S., Zhang, X., Zhang, Z., Chen, S., Ohta, T., Varlagin, A., Miyata, A., Takagi, K., Saiqusa, N., & Kato, T. (2013). Estimation of gross primary production over the terrestrial ecosystems in China. Ecological Modelling, 261–262, 80-92. [Download]
[4] Yuan, W., Cai, W., Xia, J., Chen, J., Liu, S., Dong, W., Merbold, L., Law, B., Arain, A., & Beringer, J. (2014). Global comparison of light use efficiency models for simulating terrestrial vegetation gross primary production based on the LaThuile database. Agricultural and Forest Meteorology, 192, 108-120. [Download]
[5] Yuan W P, Zheng Y, Piao S, Ciais P, Lombardozzi D, Wang Y P, Ryu Y, Chen G X, Dong W J, Hu Z M, Jain A K, Jiang C Y, Kato E, Li S H, Lienert S, Liu S G, Nabel J E M S, Qin Z C, Quine T, Sitch S, Smith W K, Wang F, Wu C Y, Xiao Z Q and Yang S. 2019. Increased atmospheric vapor pressure deficit reduces global vegetation growth. Science Advances, 5(8): eaax1396 [DOI: 10.1126/ sciadv.aax139]. [Download]
NPP Products Based on MODIS and AVHRR Data
The GLASS product utilizes MODIS and AVHRR data to estimate NPP through an advanced remote sensing-based approach. This method incorporates various environmental parameters and leverages the light use efficiency (LUE) model, which is calibrated and validated using eddy covariance flux tower measurements. The LUE model integrates factors such as photosynthetically active radiation, air temperature, and vegetation indices to simulate daily NPP. The spatial resolution of the GLASS NPP product is 0.05° latitude by 0.05° longitude, providing a comprehensive dataset for studying terrestrial ecosystem processes and their long-term variations (Zheng et al, 2020).
Based on GLASS GPP, the GLASS NPP product is generated by simulating the ratio of autotrophic respiration to GPP using 10 dynamic vegetation models from the TRENDY model comparison project. Inversion algorithms based on physical processes. GLASS Initial Grade productivity/net primary productivity (GPP/NPP) product utilization course The EC-LUE model independently developed by the question group, the development of multi generation models, and modeling The shape is gradually improving. At present, the model takes into account the photosynthetic efficiency of vegetation shade and shade leaves Differences in rates, effects of atmospheric drought, and effects of carbon dioxide fertilization Important influencing factors on vegetation photosynthesis (Zheng et al, 2020)
References：
[1] Zheng Y, Shen RQ, Wang YW, Li XQ, Liu SG, Chen JM, Ju WM, Zhang L, Yuan WP. 2020. Improved estimate of global gross primary production for reproducing its long-term variation, 1982–2017. Earth System Science Data, 12, 2725-2746. [Download]
AGB Products Based on MODIS Data
The GLASS AGB product from MODIS data is estimated and generated based on advanced land surface remote sensing products. The training data are derived from a compiled forest AGB benchmark dataset, which includes ground measurements, LiDAR, and high-accuracy biomass data. Predictor variables include GLASS Gross Primary Productivity (GPP), Albedo, Fractional Vegetation Cover (FVC), Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), Vegetation Optical Depth (VOD), and slope data. These inputs encompass data obtained from optical, microwave, and LiDAR sensors. The forest AGB prediction algorithm is the CatBoost ensemble algorithm. In previous studies, we found that CatBoost outperformed other algorithms such as Random Forest (RF), Gradient Boosted Regression Trees (GBRT), Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), and Multivariate Adaptive Regression Splines (MARS) in biomass prediction (Zhang et al., 2020). Using time-series remote sensing products and the CatBoost algorithm, the spatial distribution of global forest AGB was produced for every five years from 1985 to 2015, and the uncertainty of AGB estimation due to biomass benchmark data sampling was quantified. The Global Forest Aboveground Biomass (AGB) algorithm utilizes multiple high-resolution satellite products along with auxiliary data to generate comprehensive global AGB maps. This method employs machine learning techniques, specifically the Gradient Boosting Regression Tree (GBRT), to process and analyze data from various sources including LiDAR-derived biomass, satellite imagery, and forest inventory data. The algorithm focuses on different forest types, adapting the model to handle the unique characteristics of each. By integrating various datasets, such as Leaf Area Index (LAI), Gross Primary Productivity (GPP), canopy height, and climate variables, the algorithm is able to improve the accuracy and resolution of forest AGB estimation. The output is a high-resolution, time-series AGB map that is beneficial for assessing global carbon stocks and understanding environmental changes (Yang et al., 2020).
References：
[1] Zhang, Y., & Liang, S. (2020). Fusion of Multiple Gridded Biomass Datasets for Generating a Global Forest Aboveground Biomass Map. Remote Sensing, 12(16), 2559. [Download]
[2] Yang, L., Liang, S., & Zhang, Y. (2020). A new method for generating a global forest aboveground biomass map from multiple high-level satellite products and ancillary information. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13, 2587-2597. [Download]
SCE Products Based on MODIS Data
The GLASS SCE products from MODIS data are derived from visible and near-infrared imagery, employing an algorithm that processes the Normalized Difference Snow Index (NDSI) to distinguish snow from other surface features. Additionally, it employs a machine learning approach that incorporates environmental variables like temperature and elevation to refine snow cover classifications, particularly useful in the complex terrains of the Tibetan Plateau (Chen et al., 2015; Chen et al., 2016). These methodologies are complemented by significant references to observed changes in snow cover phenology and its impacts on radiative cooling, highlighting the algorithm's relevance in climatic and hydrological studies (Chen et al., 2017; Chen et al., 2015).
TPSCE Products Based on AVHRR Data
The Snow Cover Extent over the Tibetan Plateau (TPSCE) algorithm effectively integrates various satellite data sources to generate a comprehensive, daily snow cover extent record. This robust algorithm combines data from MODIS and AVHRR sensors, utilizing advanced cloud removal and interpolation techniques to address gaps in data, primarily caused by cloud cover (Chen et al., 2018). Moreover, the algorithm's capability to provide temporally consistent and gap-free snow cover datasets is crucial for long-term environmental monitoring across the northern hemisphere, as demonstrated in the comprehensive dataset spanning from 1981 to 2019 (Chen et al., 2021). The combined use of multi-source data and innovative processing techniques ensures that the TPSCE algorithm is not only accurate but also robust in capturing the dynamic changes in snow cover over the Tibetan Plateau, thus providing valuable insights for ecological, hydrological, and climatic research (Chen et al., 2016; Chen et al., 2018; Chen et al., 2021).
References：
[1] Chen, X., Liang, S., Cao, Y., He, T., & Wang, D. (2015). Observed contrast changes in snow cover phenology in northern middle and high latitudes from 2001–2014. Scientific Reports, 5, 16820. [Download]
[2] Chen, X., Liang, S., Cao, Y., & He, T. (2016). Distribution, attribution, and radiative forcing of snow cover changes over China from 1982 to 2013. Climatic Change, 137, 363-377. [Download]
[3] Chen, X., Long, D., Hong, Y., Liang, S., & Hou, A. (2017). Observed radiative cooling over the Tibetan Plateau for the past three decades driven by snow cover induced surface albedo anomaly. Journal of Geophysical Research: Atmospheres, 122, 6170-6185. [Download]
[4] Chen, X., Long, D., Liang, S., He, L., Zeng, C., Hao, X., & Hong, Y. (2018). Developing a composite daily snow cover extent record over the Tibetan Plateau from 1981 to 2016 using multisource data. Remote Sensing of Environment, 215, 284-299. [Download]
[5] Chen X N, Liang S L, He L, Yang Y P and Yin C. 2021. A temporally consistent 8-Day 0, 05° gap-free snow cover extent dataset over the northern hemisphere for the period 1981-2019. Earth System Science Data Discussions, 2021: 1-30 [DOI: 10.5194/essd-2021-279]. [Download]
Snow-Free Blue-Sky Land Surface Albedo Climatology Products Based on MODIS Data
Albedo is a crucial measure that affects the Earth's energy balance, influencing climate and ecological systems. Snow-Free Blue-Sky Albedo Climatology products method produces a new global 500m daily blue-sky land surface albedo climatology dataset using two decades of Moderate Resolution Imaging Spectroradiometer (MODIS) products accessed through Google Earth Engine. The method used in-situ measurements from 38 sites for validation and compared their dataset with existing albedo climatologies, including those from satellite products, reanalysis, and state-of-the-art climate models from the Coupled Model Intercomparison Projects Phase 6 (CMIP6), and found that new climatology provided more accurate and detailed albedo values, particularly for snow-covered surfaces, which are important for climate-related studies and model evaluations. The method analyzed albedo variations among different plant functional types (PFTs), and explored the impact of temporal aggregation on albedo variability, emphasizing the significance of daily-scale data for capturing albedo changes, especially in regions with distinct snow seasons(Jia, Wang, et al., 2022).
References：
[1] Jia, A., Wang, D., Liang, S., Peng, J., & Yu, Y. (2022). Global daily actual and snow‐free blue‐sky land surface albedo climatology from 20‐year MODIS products. Journal of Geophysical Research: Atmospheres, 127(8), e2021JD035987. [Download]
NDVI Products Based on MODIS Data
The GLASS NDVI product algorithm introduces an innovative approach for the reconstruction of the Normalized Difference Vegetation Index (NDVI) using a Long Short-Term Memory (LSTM) neural network. NDVI is a critical indicator used to analyze vegetation health and dynamics from satellite imagery. Traditional methods for NDVI estimation have been hampered by cloud coverage and atmospheric conditions, leading to inaccurate representations of surface vegetation. The LSTM model, trained with a combination of the Savitzky-Golay filter, GLASS leaf area index fitting, and upper envelope methods, has demonstrated substantial improvement in producing high-quality, globally representative NDVI samples. The trained LSTM effectively processes long-short temporal data to reconstruct continuous time series NDVI, showing great potential for applications in environmental monitoring and land cover change detection.
References：
[1] Xiong C, Ma H, Liang S, et al. Improved global 250 m 8-day NDVI and EVI products from 2000–2021 using the LSTM model[J]. Scientific Data, 2023, 10(1): 800. [Download]
EVI Products Based on MODIS Data
The GLASS EVI product algorithm extends the application of LSTM models to the Enhanced Vegetation Index (EVI) retrieval, which has been widely used alongside NDVI to capture the density and health of vegetation. EVI is less affected by atmospheric conditions and the soil background, and is especially sensitive in areas with dense vegetation cover. By leveraging high-quality, representative time-series samples, the LSTM model effectively predicts EVI values, even when the input data from the Moderate Resolution Imaging Spectroradiometer (MODIS) is incomplete or obscured by clouds. The study showcases that the LSTM network, with its adept handling of temporal sequences, can reconstruct spatially continuous, temporally consistent EVI datasets with reduced contamination from cloud cover and other atmospheric disturbances.
References：
[1] Xiong C, Ma H, Liang S, et al. Improved global 250 m 8-day NDVI and EVI products from 2000–2021 using the LSTM model[J]. Scientific Data, 2023, 10(1): 800. [Download]
We are here to distribute our datasets
© GLASS@HKUgeography. All Rights Reserved.