Remote sensing-based evapotranspiration (ET) has been widely used in the study of global climate change, water resources management and precision agriculture. However, due to the relative coarser spatial resolution of thermal infrared data obtained by remote sensing, the retrievals of fine resolution ET through different remote sensing-based models were full of challenge. In this paper, a general ET downscaling method based on the land surface temperature-vegetation index (Ts-VI) triangle was proposed. 990 m resolution ET datasets obtained by aggregating 90 m surface energy balance algorithm for land (SEBAL)-derived estimates from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data over a spatial dimension of 9.9 km by 9.9 km around the AmeriFlux US-Ne1 site were downscaled to 90 m by using this new proposed Ts-VI-based ET downscaling method. Compared with the original 90 m ASTER ET, the 90 m downscaled ET results had a mean absolute error (MAE) of 19.2~40.2 W/m2, a root mean square error (RMSE) of 28.8~52.0 W/m2and a bias of 1.7~2.4 W/m2.

Accurate quantification of ecosystem water use efficiency (eWUE), defined as the ratio of gross primary production (GPP) and evapotranspiration (ET), is vital to deepen our understanding of global water and carbon cycles. However, the influence of varying abiotic and biotic factors on GPP and ET is still not thoroughly understood, and thus accurate estimation of GPP and ET is still challenging, which may introduce uncertainties into eWUE. Here, we applied the random forest (RF) machine learning model to directly estimate the 8-day observed eWUE collected from the 197 globally distributed flux sites involved in the FLUXNET2015 dataset. Additionally, the RF model was also intercompared with the widely used Moderate Resolution Imaging Spectroradiometer (MODIS) and Penman-Monteith-Leuning version 2 (PMLv2) products. Our results show that the RF model could well reproduce the 8-day observed eWUE, as indicated by the root mean square error (RMSE) = 1.01 g C Kg-1H2O, the coefficient of determination (R2) = 0.66, and the mean prediction error (Bias) = 0.00 g C Kg-1H2O. More importantly, the RF model showed considerable improvements over the MODIS and PMLv2 products in simulating 8-day eWUE, with decreasing the RMSE by 1.03 g C Kg-1H2O and 0.86 g C Kg-1H2O, increasing the R2by 0.65 and 0.49, and reducing the Bias by 0.64 g C Kg-1H2O and 0.32 g C Kg-1H2O, respectively. This study indicates a promising avenue for using machine learning models to simulate eWUE directly.

Accurately quantifying ecosystem water use efficiency (WUE) is essential for advancing our understanding of carbon and water exchanges between the land surface and atmosphere. Routinely, WUE is estimated by first predicting gross primary production (GPP) and evapotranspiration (ET) and then calculating WUE as the ratio of GPP to ET. However, this approach can lead to amplified errors in WUE estimates due to uncertainties in GPP and ET predictions. Here, we proposed a novel random forest (RF)-based WUE estimation model, referred to as the DRF model, which directly predicts WUE as the targeted variable to improve WUE estimation. The DRF model was trained using a combination of remote sensing (RS), meteorological reanalysis, and digital elevation model (DEM) datasets, along with in situ WUE observations at 261 global flux tower sites from the FLUXNET2015 and AmeriFlux FLUXNET datasets. Moreover, the DRF model was intercompared with the routine WUE estimation method using the RF model (the IRF model) as well as the widely used Moderate-Resolution Imaging Spectroradiometer (MODIS) and Penman-Monteith–Leuning version 2 (PMLv2) products in WUE estimation. Our results demonstrated that the DRF model well-reproduced 8-day in situ WUE, with the root-mean-square error (RMSE) of 1.07 g C kg−1 H2O, the coefficient of determination ($R^{2}$) of 0.59, and the mean bias error (Bias) of 0.00 g C kg−1 H2O, and showed significant improvement over the IRF model with the RMSE of 1.20 g C kg−1 H2O,$R^{2}$of 0.50, and Bias of −0.09 g C kg−1 H2O. Moreover, the DRF model considerably outperformed the MODIS product (RMSE =1.93 g C kg−1 H2O,$R^{2} =0.01$, and Bias$= -0.49$g C kg−1 H2O) and the PMLv2 product (RMSE =1.70 g C kg−1 H2O,$R^{2} =0.22$, and Bias =0.25 g C kg−1 H2O). Finally, the DRF model better captured seasonal fluctuations of in situ WUE than the other three models/products. Our study indicates that the DRF model is a promising alternative to routine WUE estimation methods and has the potential to produce more accurate global WUE estimates in future studies.

Accurate quantification of terrestrial gross primary production (GPP) is essential for enhancing our in-depth understanding of the global carbon budget and climate change [1] [2] . The two-leaf light use efficiency (TL-LUE) model, considering more deeply the disparities of photosynthesis capacity between sunlit and shaded leaves, has been proven to be a more efficient and potent approach than the big-leaf light use efficiency (BL-LUE) model for global GPP simulations [3] . However, the TL-LUE model is theoretically applicable for sunny days and fails to reflect the real configuration of the canopy under overcast and cloudy days, since the direct radiation could be shaded by clouds and thus all the leaves within the canopy are in reality shaded leaves (i.e., no sunlit leaves should exist). This mismatch between the theory and reality could definitely introduce a certain degree of systematic errors into the GPP simulations by the TL-LUE model. Here, we proposed a revised two-leaf light use efficiency (RTL-LUE) model for improving GPP estimation through better quantifying the sunlit and shaded leaf area index (LAI) under different sky conditions.

Land surface temperature (LST) is an important physical parameter at the interface between the Earth’s surface and the atmosphere. Accurately quantifying LST uncertainty is essential for the generation of a long-term and consistent LST Climate Data Record (CDR) or Earth System Data Record (ESDR) from either multiple sensors or algorithms. In this study, a physical-based method was proposed to quantify the uncertainty of LST retrieval from satellite thermal infrared (TIR) data using the generalized split-window (GSW) algorithm. LST uncertainties were parameterized as a function of brightness temperature at the top of the atmosphere (TOA) and surface emissivity in two split-window channels, which are two key input parameters in the GSW algorithm, as well as their uncertainties. The performance of the parameterized uncertainty model was evaluated according to the simulation dataset at six prescribed viewing zenith angles (VZAs) of 0°, 33.56°, 44.42°, 51.32°, 56.25°, and 60°, with a root mean squared error (RMSE) of 0.001 K. The coefficients of the parameterized uncertainty model at arbitrary VZA within a sensor’s field of view (FOV) can be obtained by linear interpolation of the coefficients at the six prescribed VZAs. Once the coefficients of the parameterized uncertainty model for each pixel are available, total LST uncertainties can be quantified on a pixel-by-pixel basis. As an example, the parameterized uncertainty model was applied to actual MODIS data for displaying the spatial distribution of LST uncertainties. The results indicate that the parameterized uncertainty model can characterize the spatial variation in LST uncertainties well over various land cover types.

Land surface temperature (LST) reflects the cold and hot conditions of the land surface and is one of the most important geophysical parameters in the study and research of the land–atmosphere system. Passive microwave (PMW) is one of the primary techniques for obtaining spatially continuous LST at regional, continental, and global scales. However, there is an orbital gap in the LST retrieved from PMW (PMW LST) due to the scanning scheme of the PMW sensor, which limits the application of PMW LST, so it is necessary for the proposed some methods to fill the orbital gap of PMW LST. In this study, a new orbital gap-filling method based on a deep neural network (DNN) was developed to address the issue of PMW LST orbital gaps. This method first established the DNN model based on the nonlinear relationship between AMSR2 LST and 11 environmental variables and then used the DNN model to generate a new spatially continuous LST product, namely, DNN-LST, and, finally, used DNN-LST to fill the orbital gaps of AMSR2 LST to generate the daytime/nighttime spatially seamless gap-filled LST (GF-LST) product for China from 2012 to 2020. GF-LST can more correctly represent the spatiotemporal variation of surface temperature in China than AMSR2 LST because it has continuous spatial texture information and no obvious boundary reconstruction effect. After verifying the accuracy of GF-LST products through simulated gap region validation and in situ validation, it can be found that: 1) DNN-LST in simulated gap regions showed high accuracy during the daytime and nighttime on July 15, 2012–2020, and the mean values of bias and root mean square error (RMSE) compared with AMSR2 LST at day (night) were, respectively, −0.08 K (−0.22 K) and 1.89 K (2.23 K); 2) the accuracy of DNN-LST was the best in autumn (mean RMSE values of 1.43 K at day and 1.89 K at night) and the worst in winter (mean RMSE values of 2.35 K at day and 2.36 K at night), no matter during daytime or nighttime, in different seasons in 2015–2017; 3) the RMSE value of DNN-LST during nighttime was slightly higher than the RMSE value of DNN-LST during daytime; and 4) the accuracy of DNN-LST was equivalent to AMSR2 LST, that is, the unbiased RMSE (ubRMSE) of DNN-LST and AMSR2 LST was all about 4 K compared with in situ LST, but the ubRMSE of DNN-LST was slightly lower than AMSR2 LST. The above accuracy validation analysis shows that DNN-LST has good robustness and good spatial consistency with AMSR2 LST and can be well used to fill the orbital gap of AMSR2 LST to generate spatial seamless GF-LST product.

The monthly mean land surface temperature (MMLST) reflects more stable intra- and interannual temperature variations, and therefore, it has a wider range of applications than instantaneous land surface temperature (LST). This study aimed to generate a high-resolution global MMLST product by temporally upscaling the Moderate Resolution Imaging Spectroradiometer (MODIS) 1-km instantaneous LST. First, six current methods were comprehensively evaluated using cross-validation technology. These six methods are the cross combinations of two temporal aggregation schemes: the average by observations (ABO) and average by days (ABD), and three conversion models: the diurnal temperature cycle model (DTC), the simple average of two instantaneous LSTs (TSA), and a weighted average model for multiple instantaneous LSTs (MWA). The analysis with measurements from 235 flux stations worldwide revealed that the choice of conversion model considerably affected the overall retrieval accuracy, whereas the influence of the aggregation scheme was minor. From the conversion model standpoint, MWA performed best, followed by DTC, and finally TSA; this order remained the same even if DTC and TSA were improved with mean bias correction. Notably, the errors of ABDMWA decreased as the number of daily mean LST (NOD) increased, whereas the errors of ABOMWA were not related to NOD. Accordingly, we deduced that the optimal strategy for estimating MMLST is using ABOMWA when NOD is$\ge 20$. Subsequently, we adopted this combination method to process MODIS instantaneous LSTs and produced a global 1-km MMLST dataset for the years 2003–2020. The validation showed a satisfactory accuracy with a root mean square error (RMSE) of 1.6 K. The intercomparison with MMLSTs from geostationary (GEO) satellites (containing complete LST daily cycle) presented a good agreement (biases < 0.3 K and STDs < 2 K). Compared with atmospheric infrared sounder (AIRS) L3 monthly standard physical retrieval (AIRS3STM) product which had the same temporal span, the newly generated product exhibited a high consistency in reflecting temporal variations of global temperature. Most importantly, it had a prominently better ability to retrieve spatial details of temperature variations due to its higher resolution. Our new method and product show promising prospects for applications in global change studies, where accurate spatially resolved MMLST data are one of the fundamental geophysical variables required.

Accurate quantification of gross primary production (GPP) in agroecosystems not only improves our ability to understand global carbon budget but also ensures basic human survival supplements. Here, we improved the MODIS-GPP algorithm by two main perspectives: (1) taking the seasonal variations of maximum light use efficiency (LUE) into modeling consideration; (2) separately parameterizing maximum LUE with a recently proposed vegetation index (VI) NIRv during vegetative stage and senescence stage. Performances of the revised and traditional MODIS-GPP algorithms were tested at three FLUXNET crop sites planted with maize and soybean. The revised model was well validated, indicated by the root mean square error (RMSE), coefficient of determination$(\mathrm{R}^{2})$and Bias being 2.33$\text{gC m}^{-2}$day${}^{-1}, 0.91$and 0.48$\text{gC m}^{-2}$da y -l for maize, respectively, and being 1.51$\text{gC} \mathrm{m}^{-2}\text{day}^{-1},0.91$and 0.43$\text{gC m}^{-2}$day$-1$for soybean, respectively. Overall, compared to the traditional MODIS-GPP algorithm, the proposed algorithm reduced RMSE by 29.6% and 27.4%, increased$\mathrm{R}^{2}$by 10.9% and 10.9%, and reduced Bias by 41.5% and 36.8% for maize and soybean, respectively. This paper demonstrates that incorporating seasonal fluctuations of maximum LUE into MODIS-GPP algorithm and distinguishing the different photosynthesis rates among vegetative and senescence stages significantly benefit the retrieval accuracy of daily model-estimated GPP.

Gross Primary Production (GPP) and evapotranspiration (ET) play a critical role of the global carbon, water and energy cycle. Accurate quantification of the global GPP and ET could improve our ability to understand global climate change and energy budget. However, most of the GPP and ET remote sensing models fail to take the coupled relationship between vegetation transpiration (Et) and photosynthesis into consideration. More importantly, these models might ignore the difference of transpiration and photosynthesis rate in different groups of leaves (sunlit and shaded). Here, we coupled the estimates of daily GPP and ET at 84 global forest sites based on the Two-Leaf Light Use Efficiency model and the Penman-Monteith equation that were linked by the Ball-Berry conductance model. The developed model was well calibrated with the root mean square error (RMSE) and the coefficient of determination (R2) being 1.97 gC/m2 day and 0.77 for GPP respectively, and being 21.91 W/m2and 0.65 for ET, respectively. In the meantime, the validation results demonstrated the good performance of the coupled model, with the RMSE and R2 being 1.95 gC/m2 day and 0.77 for GPP, respectively, and being 21.34 W/m2and 0.66 for ET, respectively.

Evapotranspiration (ET) is an important variable in hydrological cycle and widely used in the study of water management and climate change. This paper developed a Random Forest (RF) model for global daily ET estimation over vegetated areas with 500 m spatial resolution using merely MODIS data. 255 in-situ sites from AmeriFlux network, FLUXNET Network and National Tibetan Plateau Data Center (TPDC) of China have been used to evaluate the RF model. Results indicated that the RF model-estimated global daily ET exhibited reasonable accuracy compared to the in-situ observations using MODIS datasets as inputs, with root mean square error (RMSE) between 0.52-0.97 mm/d over six different land-cover types representing forest, shrubland, cropland, savanna, grassland and wetland. The models generally achieved the best performance in shrubland, grassland and savanna, while provided the worst in wetland.

Fine spatial resolution land surface temperature (LST) data derived from a thermal infrared remote sensing image are essential to the study of land surface energy, water and carbon cycles. As an alternative and effective way to obtain fine spatial resolution LST, a large number of LST downscaling methods have been proposed in recent decades to enhance coarse resolution LST to fine resolution. However, the drawbacks of the random selection of scaling factors and the establishment of statistical regression relationships are obvious. In this context, a general and physical LST downscaling method based on surface energy balance (DTsEB) is proposed in this study. Moderate Resolution Imaging Spectroradiometer (MODIS) LST data at 990 m spatial resolution were downscaled to 90 m by using this new proposed SEB-based LST downscaling method in this study. Compared with the concurrent 90 m resolution Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) LST data, the downscaled results have a mean absolute error (MAE) of 1.37 K and a root mean square error (RMSE) of 1.84 K.

Evapotranspiration (ET), including soil evaporation and vegetation transpiration, is a vital component of the water cycle and energy exchange. The end-member-based soil and vegetation energy partitioning approach (ESVEP model) for estimating ET is the first model considering the differing responses of soil water content at the upper surface layer and at the deeper root zone layer. In this paper, we have improved the ESVEP model by 1) improving estimates of canopy resistance from three parts: stomatal conductance, cuticular conductance and leaf boundary-layer conductance; 2) adding the influence of atmosphere pressure and atmosphere temperature on resistance; 3) dividing aerodynamic resistance into convective resistance and radiative heat transfer resistance. Compared to the original ESVEP model, the improved algorithm is more applicable for different biome types and has a great potential in operational estimation of regional and global evapotranspiration. Due to the underestimation of soil temperature, the estimated ET is biased, which means that the biome properties and resistance are required to be further reparameterized in future study.

As the main process parameter of water and energy exchange, evapotranspiration (ET) is defined as the water being converted from liquid to gaseous and from land surface to atmosphere. Potential evapotranspiration (ETO) is defined as the evapotranspiration when water supply is sufficient of the land surface and reflect the ability of the surface to supply moisture. In this study, we explored the relationship between annual averaged ET (ET/ETO) and annual averaged 36.5 GHz emission, and provided a new train of thought of how to use passive microwave data to estimate annual averaged evapotranspiration. We found a non-linear relationship with a R2 of 0.52 between annual averaged 36.5 GHz emission and observed annual evapotranspiration at 28 flux tower sites of Asia and North America. We estimated ET and ETO of China and found a linear relationship with a R2 of 0.51 between the annual averaged (ET/ET0)1/2and the annual averaged 36.5 GHz emission at 9 flux tower sites of China.

Continuous monitoring of high spatial resolution evapotranspiration (ET) is critical for water resources management at both regional and local scales. This research employs a multi-sensor satellite data fusion approach (ESTARFM: Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model) combined with a Two-Source N95 model and a constant evaporative fraction method to compute daily ET at 30 m spatial resolution. Two schemes are followed: the first scheme is to apply ESTARFM on the LST data to estimate daily ET at 30 m spatial resolution. The second scheme is to apply ESTARFM on the ET derived from MODIS and Landsat 8 images. The results show that the ET fused by both schemes is in good agreement with the reference ET data from the Landsat 8, while the first scheme (applying the ESTARFM on LST) is observed with more variations.

Linking the terrestrial water cycles, carbon cycles and energy exchange, evapotranspiration (ET), which combines the surface evaporation and plant transpiration, is a key land surface parameter in water and heat balance of land, lake or river surface, and is central to earth system science. In this study, based on the MERRA reanalysis dataset and MODIS NDVI and LAI product, a support vector machine was used to estimate the land surface ET at sites and global scales. The results showed that, the support vector machine model probably could explain 60%–80% of the land surface ET change at 242 global FLUXnet sites when ten indicators while 56%–79% when five indicators were used to drive the model. For different vegetable cover sites, compared with EC observations, the results of evergreen broadleaf forest was worse than others.

Evapotranspiration (ET) is the combination process of the surface evaporation and plant transpiration which occur simultaneously, and it links the terrestrial water cycles, carbon cycles and energy exchange. In this study, based on the observations from 242 global FLUXnet sites, with daily average temperature, relative humidity, wind speed, incident solar radiation, NDVI and observed ET as input data, we used a support vector machine to estimate the land surface daily ET at nine different vegetation type sites. The results show that, for all vegetation type sites, when the predicted ET was validated with the eddy covariance measurements, the support vector machine algorithm underestimates the ET and probably could explain 71%-86% of the land surface ET change.