Flood disasters monitoring require near real-time acquisition of water coverage information. As two major water extraction methods, thresholding has low accuracy in small water body areas, while machine learning approaches do not meet the demand for rapid monitoring. In this study, three global water distribution datasets, JRC-GSW, GSWD and cDSWE, were reorganized to obtain a combined multi-source water distribution dataset (CMWDD). The machine learning methods were then used to assist in determining the optimized thresholds. Results showed improved accuracy (OA = 98%, Kappa = 0.83). It solves the problem of low accuracy of the threshold method in the area of small and medium-sized water bodies, and provides flood inundation extent data with high temporal resolution.

Radiometric normalization is a critical step in producing consistent vegetation index data. However, to date, there have been few studies in the field of high-resolution vegetation index mapping. In this letter, a set of reference-based radiometric normalization methods, combining absolute and relative approaches, are evaluated for high-resolution vegetation index mapping. A framework incorporating a region-to-region [moment matching (MM)] method is introduced and compared to a traditional pixel-to-pixel method (M-estimation). Three modeling strategies (global, classification, and blocked) are evaluated for the two methods. Specifically, we describe the six sets of normalization experiments conducted with high-resolution images captured by China’s Gaofen-2 (GF-2) satellite and compare the results qualitatively and quantitatively. In addition, the sensitivity of the three parameters (purity threshold, number of categories, and block size) is analyzed. The results show that the MM method outperforms the M-estimation method visually. However, opposite performances are found in the quantitative evaluation. In summary, the blocked modeling strategy is recommended if the data volume is small. The findings of this letter will have important implications for the production of GF-2 vegetation index products at a large scale.

High-resolution seamless remote sensing (HRSRS) images are essential foundational data for natural resource monitoring and land-use assessment, etc. However, high spatial resolution (HR) Earth observation from satellites typically has a long revisit period, and optical images can be widely obscured by clouds. Furthermore, the geometry and radiation issues make the generation of HRSRS images a challenging task. In this letter, to systematically address these issues, we propose a robust and efficient framework designed to generate HRSRS images for regional-scale areas, integrating various mature image processing technologies and jointing super-resolution (SR) reconstruction and thick cloud removal for the first time. In particular, the frame-work can realize adaptive reconstruction of lost spatial information on demand based on the satellite observation coverage and thick cloud coverage information. The effectiveness and reliability of the framework was demonstrated by its successful application in generating a 1-m resolution quarterly seamless image of the city of Wuhan, Hubei province, China, using 34 images acquired by the Chinese Gaofen (GF)-1/2/6/7 satellites. The experimental results show that both the intermediate results and final results achieve a satisfactory visual and quantitative effect. For example, SR reconstruction improves the spatial resolution of low-resolution images from 2-m to 1-m, thus increasing the spatial frequency and entropy by about 0.62 and 0.12, respectively.

Current assessments of urban forest carbon storage were missing or largely underestimating their values due to limited spatial resolution. In this study, combining field plot measurements and satellite imagery, a wall-to-wall forest biomass map were generated at a very high spatial resolution (5 m) over urban areas in Wuhan City, China. Specifically, a series of characteristic metrics were extracted from Jilin-1 satellite images, including multispectral reflectances, vegetation indices, and texture features. The estimations of forest aboveground biomass from three machine learning models were evaluated at sampled field plot level. Results demonstrated that the random forest model achieved the highest accuracy using the leave-one-out cross-validation method, with a test set RMSE of 31.84 Mg/ha. However, discrepancies were observed in low biomass areas due to variations in vegetation species, leading to overestimation of lower values.

Current assessments of urban forest carbon storage were missing or largely underestimating their values due to limited spatial resolution. In this study, combining field plot measurements and satellite imagery, a wall-to-wall forest biomass map were generated at a very high spatial resolution (5 m) over urban areas in Wuhan City, China. Specifically, a series of characteristic metrics were extracted from Jilin-1 satellite images, including multispectral reflectances, vegetation indices, and texture features. The estimations of forest aboveground biomass from three machine learning models were evaluated at sampled field plot level. Results demonstrated that the random forest model achieved the highest accuracy using the leave-one-out cross-validation method, with a test set RMSE of 31.84 Mg/ha. However, discrepancies were observed in low biomass areas due to variations in vegetation species, leading to overestimation of lower values.

Solar-induced chlorophyll fluorescence (SIF) is an intuitive and accurate way to measure vegetation photosynthesis. Orbiting Carbon Observatory-2 (OCO-2)-retrieved SIF has shown great potential in estimating terrestrial gross primary production (GPP), but the discontinuous spatial coverage limits its application. Although some researchers have reconstructed OCO-2 SIF data, few have considered the uneven spatial and temporal distribution of the swath-distributed data, which can induce large uncertainties. In this article, we propose a spatiotemporal constrained light gradient boosting machine model (ST-LGBM) to reconstruct a contiguous OCO-2 SIF product (eight days, 0.05°), considering the data distribution characteristics. Two spatial and temporal constraining factors are introduced to utilize the relationships between the swath-distributed OCO-2 samples, combining the geographical regularity and vegetation phenological characteristics. The results indicate that the ST-LGBM method can improve the reconstruction accuracy in the missing data areas ($R^{2}= 0.79$), with an increment of 0.05 in$R^{2}$. The declined accuracy of the traditional light gradient boosting machine (LightGBM) method in the missing data areas is well alleviated in our results. The real-data comparison with TROPOspheric Monitoring Instrument (TROPOMI) SIF observations also shows that the results of the ST-LGBM method can achieve a much better consistency, in both spatial distribution and temporal variation. The sensitivity analysis also shows that the ST-LGBM can support stable results when using various input combinations or different machine learning models. This approach represents an innovative way to reconstruct a more accurate globally continuous OCO-2 SIF product and also provides references to reconstruct other data with a similar distribution.

Accurate mapping forest height at fine spatial resolution is essential for evaluating terrestrial ecosystem service. Yet, current assessments of forest height rely primarily on statistical or coarse scale model estimates, thus lack of spatial details for decision making at local scales. Recent advances in remote sensing technology provide great opportunities to fill this gap. Satellite data from radar and multispectral instruments are promising in providing spatial continuous observations. Here, we present a work that combined field measurements and satellite imagery to generate a wall-to-wall forest height map at a 30-m spatial resolution. Field plot data collected from October 2017 to April 2018 were used for model calibration and validation. A series of characteristic metrics were tested, including Landsat-8 multispectral reflectance and vegetation indices, Sentinel-1 C-band, and PALSAR-2 L-band SAR backscattering coefficients and difference index, and SRTM topographic variables. Our results indicate that a few variables from SRTM, Landsat, and Sentinel-1 show stronger relationships with forest height. We evaluated three types of models, including multiple linear regression (MLR), support vector regression (SVR), random forest (RF). Results show that RF model perform best (R2=0.44, RSME=4.7 m) compared with the other two methods (MLR, R2=0.31, RSME=5.0 m; SVR, R2=0.37, RSME=4.5 m).

Quickly extracting the spatial extent of flooding is necessary for disaster analysis and rescue planning. A large number of researches utilized optical remote sensing data to extract surface water extent. However, current data products derived from optical sensors are difficult to meet the need of rapid flood monitoring due to cloud cover. Radar remote sensing supports `all-weather' and `day-and-night' water information extraction. Here, we proposed an automatic thresholding approach to extract flood coverage using Sentinel-1 synthetic aperture radar (SAR) and prior classes information. Sixteen years of Landsat remote sensing image data were analyzed by Google earth engine (GEE) and prior classes of water and non-water were generated from composited dynamic water extent (cDSWE). We combined the distribution probability data of the prior classes of water with the extracting result to calculate the area of inundation. Two sites, Shouguang in Shandong province and Ji'an in Jiangxi province, were selected as representative areas for reservoir and floodplain. The results show that the overall classification accuracy for water is above 90%. Commission errors of water bodies ranged from 0% to 11.32%, and the omission errors ranged from 6.0% to 10.2%. Validations indicate the algorithm can effectively extract the spatial extent of reservoir or floodplain.

Two automated approaches, including Bayesian probability thresholding and regression tree based methods were utilized to detect the surface water extent with training dataset from prior class probabilities of water and non-water from two datasets. First, prior water and non-water masks were classified using SRTM Water Body Dataset (SWBD) and long-term summarized Dynamic Surface Water Extent (DSWE) class probabilities. Then, fully automatic algorithms were developed to derive water probability and classify surface water extent using Sentinel-1 data. Results over three representative study regions, including the Delmarva Peninsula, Florida Everglades and Prairie Pothole regions, indicate that the automated algorithm is efficient in monitoring open water inundation extent, and detection of partial water extent is possible using Sentienl-1 SAR data.

Time series of remote sensing data offers the opportunity to predict changes in vegetation extent and to estimate forest parameter change such as biomass. However, as sensors and technology advance, it is important to ensure that estimates obtained from different time periods or using different, but related, instruments are consistent in order to have confidence in detected change. This study compares estimates of biomass from small-footprint discrete-return LiDAR data and medium-footprint full-waveform LiDAR for Howland Experimental Forest, Maine, USA. Data were collected from both sensors during Summer 2009. Similar results were found using the same height metric with R2= 0.67, SE = 58.5 Mg ha-1and R2= 0.52, SE = 58.1 Mg ha-1, respectively. The predicted model of the relationship between LiDAR metrics and biomass was applied to data captured in 2003. Identified areas of change corresponded well with a map of forest management operations of varying intensities. Where sensitivity to change allows, vegetation age estimated using time series of Landsat observations, combined with biomass estimates, allows growth curves to be produced to monitor the effect of pests or disease, recovery rates following disturbance, or carbon sequestration.

This letter concentrates on scattering mechanism classification of polarimetric synthetic aperture radar (PolSAR) images. Scattering mechanism classes are defined as the combinations of dominant and secondary scattering mechanisms. With three metrics extracted from the observed coherency matrix, an unsupervised classifier is proposed to classify PolSAR pixels into eight combinations of surface scattering, double-bounce scattering, and volume scattering. When applying the proposed method to simulated data, the Kappa coefficient is 0.891. It effectively classifies the dominant mechanism, and the Kappa coefficient is 0.127 higher than that of the H/α method. Experiment using uninhabited aerial vehicle SAR data shows that the proposed method is able to identify secondary mechanism in forests and urban areas. This method is not only a good classifier free of specific polarimetric decomposition but also can serve as a preclassification step of sophisticated classification scheme.

The estimation of forest biomass from synthetic aperture radar (SAR) data is limited by the lack of forest structure information. Interferometric synthetic aperture radar (InSAR) provides a means for the extraction of forest structure. The crucial issue in InSAR application is to parameterize forest structure and to link the parameter with InSAR observations. Model-based analysis enables exploring the theoretical linkages between InSAR observations and forest structure free from temporal decorrelation effects. In this paper, a semicoherent model (SCSR) was first developed and verified. A series of simulations at L-band was then made for both homogeneous and heterogeneous forests generated from a forest growth model. The forest structure was parameterized by four height indices. Aside from the height of scattering phase center (HSPC), the depth of scattering phase center (DSPC) was also proposed to characterize the scattering phase center of InSAR. The results showed that the behavior of homogeneous forest on InSAR data was quite different from that of heterogeneous forest. Special care was needed when the retrieval algorithms of forest biomass developed on a homogeneous forest were applied to a heterogeneous forest. Crown size-weighted height (CWH) and Lorey's height were correlated with the HSPC at all polarizations and with the DSPC at copolarization in both cases of homogeneous and heterogeneous forests. These findings indicated that CWH could be an alternative biomass indicator of the Lorey's height for biomass estimation, which can be derived from the combination of InSAR data and the elevation of the forest canopy top from lidar or high-resolution stereo images.

Accurate estimates of aboveground biomass (AGB) from forest after disturbance could reduce the uncertainties in carbon budget of terrestrial ecosystem and provide critical information to related carbon policy. Yet the loss of carbon from forest disturbance and the gain from post-disturbance recovery have not been well assessed. In this study, sensitivity analysis was conducted to investigate: (1) influence of factors other than the change of AGB (i.e. distortion caused by incident angle, soil moisture) on SAR backscatter; (2) feasibility of cross-image calibration between multi-temporal and multi-sensor SAR data; and (3) possibility of applying normalized backscatter to detect the post-disturbance AGB recovery. A semi-automatic empirical model was proposed to reduce the incident angle effect. Then, a cross-image normalization procedure was performed in order to remove the radiometric distortions among multi-source SAR data. The results indicate that effect of incident angle and soil moisture on SAR backscatter could be reduced by the proposed procedure, and a detection of biomass changes is possible using multi-temporal and multi-sensor SAR data.

Biomass at local to regional scales is important for carbon cycle study and monitoring the ecosystem responses to natural and human activities. This paper directly quantify biomass and its changes at 1-ha (100 m) spatial resolution from LiDAR footprint-level waveform data. A large-footprint full-waveform LiDAR (LVIS) data were acquired in Penobscot County, Maine State (USA) in August, 2003 and 2009. Field data were collected during the October 2003, and August of 2009 to 2011. The developed models using field measurements at waveform footprints were applied to all LVIS waveforms within the study site. Plots at 0.25-ha, 0.5-ha and 1-ha were used to validate the biomass averaged from footprints measured in these plots. The effect of forest disturbances on LiDAR biomass prediction models was investigated in the study. The results show that: 1) the prediction accuracy of models at footprint-level was acceptable at various plot-levels; 2) the footprint-level models could be applied for forest biomass with consideration of forest disturbance; 3) the 1-ha (100 m) was a proper scale for mapping of forest biomass and its change detection.

Forest canopy height is an important indicator of standing biomass for management purposes as well as for the assessment of carbon storage. Theoretically, photogrammetry is one of remote sensing technologies which can be used to extract forest canopy height information. The Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) carried by the Advanced Land-Observing Satellite was designed to generate worldwide topographic data with its high-resolution and stereoscopic observation. A semi-automatic method for the extraction of digital surface model using PRISM data with ENVI is introduced. Tie points scattered evenly over the common area are manually selected on the stereo-image pair. Their ground coordinates are calculated using the geocoding parameters delivered along with the nadir image. Their elevations are extracted from SRTM by their ground coordinates. Ground control points (GCP) files needed in the stereo-processing of PRISM data can be composed by the image coordinates, ground coordinates and elevations. With the prepared tie points and GCP files, PRISM data can be stereo-processed automatically by ENVI. The results showed that the elevation from PRISM DSM is highly correlated with that from GLAS data and forest canopy height information is explicitly exhibited on it.

Forest plays a vital role in carbon, energy and water cycling of Earth System. Estimation of forest biomass has become essential to ecosystem studies. In this paper, three approaches, linear regression, Maximum Entropy and Look-up table based on model, were adopted. It was found that linear regression or Maximum Entropy met the saturation when biomass reached 170 Mg/ha, by use of either HV polarization or all polarizations. Look-up Table based on backscatter model could predict biomass well with the highest R2=0.74, and over 250Mg/ha without saturation problem.

Former NASA's Deformation, Ecosystem Structure, and Dynamics of Ice (DESDynl) satellite mission was to provide regional or global biomass carbon stock at regional, national, and global scales. Carbon Monitoring System UAVSAR and LVIS data acquired in August 2009 were used in this paper. In this paper we used Cloude's target decomposition theorem to decompose traditional polarimetric SAR data, and then to estimate biomass. By certification of LVIS derived biomass, the results show that only decomposed scattering elements are not enough for biomass retrieval, but they could improve the accuracy of biomass retrieval. The standard error for only pol method is 66.3Mg/ha, but for both polametric and decomposed data is only 64. lMg/ha.

Parameters of vegetation spatial structure have important effect on the carbon cycle and biodiversity of the ecosystems. How to estimate above-ground biomass is still a problem need to be worked out. In this paper we tried to use UAVSAR datasets to discuss the relation between backscattering coefficient and local incidence angle in different forest types. By the relation, a method based on scattering mechanism for correcting radiometric distortion caused by large range of incidence angle is developed. Biomass retrieval is based on incidence angle correction. The result shows good correlation between biomass and backscattering coefficient in 1 ha scale.