High spatial and temporal resolution Earth Observation (EO) data are essential for global environmental monitoring. A cost-effective and feasible approach to obtain such data involves integrating observations from multiple satellites equipped with complementary sensor attributes. This integration aims to generate a calibrated and consistent Analysis Ready Dataset (ARD) featuring optimized spatial and temporal resolutions. Our recent research produced a global daily 30-m Seamless Data Cube (SDC) of land surface reflectance by fusing Landsat, MODIS, and AVHRR from 1985 to 2022. Our experiment results demonstrated the superior capabilities of SDC for monitoring global land dynamics. In this paper, we will discuss our current progress, lessons learned, and future plans for the global SDC applications.

Change detection in remote sensing is crucial for land use/land cover (LULC) change awareness. However, large-scale change detection suffers from limited contextual understanding of spatial intricacies of large changes in existing CNN-based methods. Overlapped receptive fields lead to weight sharing across feature sliders, contributing to limited detection ability on large dense multi-class changes. To address this problem, this paper presents a receptive convolution operation for large-scale multi-class change detection from high-resolution remote sensing images. Different from existing CNN-based networks, our architecture involves receptive convolutions with a large kernel size to guarantee focus on different receptive field features. Experiments on SECOND datasets show that the proposed method achieves better performance than previous counterparts. Furthermore, a large-scale LULC change detection is conducted to demonstrate the ability in large-scale applications.

In high-resolution remote sensing images (RSIs), complex composite object detection (e.g., coal-fired power plant detection and harbor detection) is challenging due to multiple discrete parts with variable layouts leading to complex weak inter-relationship and blurred boundaries, instead of a clearly defined single object. To address this issue, this article proposes an end-to-end framework, i.e., relational part-aware network (REPAN), to explore the semantic correlation and extract discriminative features among multiple parts. Specifically, we first design a part region proposal network (P-RPN) to locate discriminative yet subtle regions. With butterfly units (BFUs) embedded, feature-scale confusion problems stemming from aliasing effects can be largely alleviated. Second, a feature relation Transformer (FRT) plumbs the depths of the spatial relationships by part-and-global joint learning, exploring correlations between various parts to enhance significant part representation. Finally, a contextual detector (CD) classifies and detects parts and the whole composite object through multirelation-aware features, where part information guides to locate the whole object. We collect three remote sensing object detection datasets with four categories to evaluate our method. Consistently surpassing the performance of state-of-the-art methods, the results of extensive experiments underscore the effectiveness and superiority of our proposed method.

As one of the most important application for remote sensing monitoring, land cover mapping has witnessed notable advancements in data acquisition, algorithmic diversity, and classification accuracy. Despite the instrumental role data-driven algorithms have played in the development of global land cover products, their inherent limitations as “black box” methods often fall short of meeting end-users’ specific requirements. In this study, built upon the foundation of the earlier land cover monitoring platform [FROM-GLC plus(FGP)], a data and knowledge dual-driven framework (FGP 2.0) was developed as a user-adaptive framework for intelligent remote sensing land cover mapping. By incorporating ontology-based semantic descriptions with advanced data-driven algorithms, FGP 2.0 provides the capacity for both traditional annual mapping and emerging dynamic mapping. Our results illustrate that FGP 2.0 significantly improves the overall accuracy of annual maps by ~5%, and dynamic maps by ~20% compared to FGP. Moreover, an operational dynamic mapping tool has been developed on the Google Earth engine (GEE), enabling the generation of near-real-time land cover maps for any given place. With an extensible and flexible mapping framework, FGP 2.0 demonstrates the potential of customized land cover monitoring results to suit different application scenarios. This innovative approach not only meets the current demand for reliable annual and dynamic land cover maps but also sets a new benchmark for the integration of geoscientific expertise with machine learning techniques in remote sensing monitoring.

High-resolution land cover mapping over large areas is a challenging task due to the lack of high-quality labels. A potential solution is to leverage the existing knowledge contained in the freely available lower-resolution land cover products. However, the relatively low resolution and low accuracy of the products lead to numerous inaccurate labels, which harms the performance of the neural network. This article addresses the challenge by jointly optimizing the network parameters and correcting the noisy labels with a novel online noise correction approach and a synergistic noise correction loss. By incorporating the information entropy as a measurement to determine the probable correct labels, the proposed noise correction approach learns to make effective correction of the noisy labels during training and eventually boosts the performance with a training set containing less noisy labels. Experimental results show that the proposed method can effectively correct the noisy labels and reduce their negative impact on network training. By employing the proposed method, we produce a refined high-resolution (3-m) land cover map from a lower-resolution (10-m) product in China and improve the accuracy from 74.96% (10-m) to 81.32% (3-m). Such an approach that can effectively learn from noisy data sets leads to many potential opportunities for using and magnifying existing knowledge and results.

Ocean surface current is a very important parameter of ocean dynamic environment. The observation and prediction of ocean surface current has attracted more and more concern. Doppler Scatterometer (DopScat) is a new type of radar for ocean surface wind and current field remote sensing. The ocean surface current inversion method of DopScat impacts the measurement accuracy directly. In this paper, we establishes the Maximum Likelihood Estimation(MLE) method to retrieve the ocean surface current and wind simultaneously. The retrieval accuracy for different position in cross-track, wind speed, and current speed are analyzed. The retrieval results show that the RMS of inversion current speed and direction can be smaller than 0.18m/s and 25°respectively, for medium wind speed condition. This is submitted for the special session of “New Developments of Chinese Oceanographic and Meteorological Satellites”.

Microwave remote sensing is one of the most useful methods for observing the ocean parameters. The Doppler frequency of the radar echoes can be used for ocean surface current speed retrieval. While the effect of the ocean currents and waves are interactional. In this paper the suitable ocean wave elevation spectrum and directional distribution function are selected by comparing the ocean Doppler spectrum in C band with the empirical geophysical model function (CDOP). The simulation results show that the ocean surface current speed error is sensitive to the wind speed and wind direction error. With VV polarization, the ocean surface current speed error is about 0.15 m/s when the wind speed error is 2 m/s, and the ocean surface current speed error is smaller than 0.3 m/s when the wind direction error is within 20° in the cross wind direction. This is submitted for the special session of “New Developments of Chinese Oceanographic and Meteorological Satellites”.

The ocean surface current is a very important parameter of ocean dynamic environment. It is connected to global climate change, marine environment forecasting, marine navigation, engineering security, and so on. The observation and prediction of ocean surface current have attracted more and more concern. Doppler Scatterometer (DopScat) is a new type of radar for ocean surface wind and current field remote sensing. The ocean surface current inversion method of DopScat impacts the measurement accuracy directly. In this paper, we establish the simulation model of a DopScat and provide the radial velocity error model. The numerical ocean surface Doppler spectrum model is also introduced and validated with the empirical geophysical model function in C-band (CDOP). The suitable ocean wave elevation spectrum and directional distribution function are selected. What is more, this paper establishes the maximum likelihood estimation (MLE) method to retrieve the ocean surface current and wind simultaneously. The retrieval accuracy for different positions in cross track, different wind speeds, and different current speeds are analyzed. At last, the global ocean current field is observed by DopScat and the ocean current is retrieved. In our simulation, the orbit parameters and observation geometry of DopScat are the same as that of HY-2A scatterometer. The retrieval results show that global current speed standard deviation can be smaller than 0.18 m/s for five days and$0.5 {^{\circ }} \times 0.5 {^{\circ }}$grid average.

As the launch of radars, such as the Precipitation Radar (PR) on Tropical Rainfall Measuring Mission (TRMM)[1] satellite and the Surface Wave Investigation and Monitoring (SWIM) on China France Oceanography SATellite (CFOSAT)[2], that operate in low incident angles, more and more NRCS data at low incident angle will be obtained. In order to retrieve the sea surface wind speed using low incident angle NRCS, the empirical GMF of NRCS with wind speed is established. The empirical nadir reflection coefficient |R(0)2| are calculated and the empirical relationship between mean square slop s(u) and wind speed is established. The mean square slop s(u) can be retrieved by fitting the NRCS at certain incident angles with the theoretical Gaussian GMF model. Then the wind speeds are calculated using the empirical corresponding relation between mean square slop and wind speed. The retrieved wind speeds are compared with Tao and NDBC buoy. The results show that the standard deviation (STD) and bias of retrieved wind speeds are smaller than 1.7m/s and 0.1m/s respectively.

Simplex volume is the most commonly used parameter for endmember extraction. However, when outliers exist in the image, the maximum-volume-criterion (MVC)-based methods tend to extract them as endmembers. Those outlier endmembers could be either physically meaningless or not representative enough for prevalent land covers. This is the biggest bottleneck preventing MVC-based methods from being extended from theoretical analysis to practical applications. This is mainly due to the limitation of the simplex volume formula itself, which is only determined by simplex vertices and completely ignoring the statistics of the data cloud. Usually, the simplex with vertices containing outliers has a larger volume than the one with vertices only containing true endmembers; thus, outliers are more favorably extracted as endmembers. Usually, the outliers are distributed in the direction of low information content. When extracted endmembers contain outliers, the overall information content (OIC) of the data cloud projected onto the endmember subspace will be definitely reduced. Motivated by this fact, we present the concept of statistical volume and develop a new endmember extraction method, which is named statistical volume analysis (SVA). The algorithm simultaneously utilizes the geometrical property of the simplex and the statistical characteristic of the projected data in the endmember subspace. Therefore, SVA not only can find a simplex with a large volume but also can get a large OIC of the projected data. Experiments with both simulated and real data show that SVA can compete with state-of-the-art methods in extracting endmembers of prevalent land covers. Moreover, it is capable of avoiding extracting outliers as endmembers.

Transition rules are the core of urban cellular automata (CA) models. Although the logistic cellular automata (Logistic-CA) is commonly used for rules extraction, it cannot always achieve satisfactory performance because of the spatial heterogeneity and the inherent complexity of urban expansion. This article presents an ensemble-urban cellular automata (Ensemble-CA) model to achieve better transition rules. First, an uncertainty map that assesses the performance of transition rules spatially was achieved. Then, two auxiliary models (i.e. classification and regression tree, CART; and artificial neural network, ANN), both of which have been stabilized with a Bagging algorithm, were prepared for integration using a proposed self-adaptive -nearest neighbors (-NN) combination algorithm. Thereafter, those unconfident sites were replaced with the ensemble output. This model was applied to Guangzhou, China, for an urban growth simulation from 2003 to 2008. Static validation confirmed that this ensemble framework (i.e. without substitution of uncertain sites) can achieve better performance (0.87) in terms of receiver operating characteristic (ROC) statistics (area under the curve, AUC), and outperformed the best single model (ANN, 0.82) and other common strategies (e.g. weighted average, 0.83). After the substitution of unconfident sites, the AUC of Logistic-CA was elevated from 0.78 to 0.81. Subsequently, two urban growth mechanisms (i.e. pixel- and patch-based) were implemented separately based on the integrated transition rules. Experimental results revealed that the accuracy obtained from simulation of the Ensemble-CA increased considerably. The obtained kappa outperformed the single model, with improvements of 1.74% and 2.76% for pixel- and patch-based approaches, respectively. Correspondingly, landscape similarity index (LSI) improvements of these two mechanisms were 4.24% and 1.82%.

Geoscience Laser Altimeter System data have been widely used in forest canopy height extraction. It is still challenging over rugged areas. In this paper, we propose a forest canopy height extraction method consisting of the Savitzky-Golay filter and fitting, Sigbeg determination based on the fitting results, and slope correction for rugged areas, particularly for slopes ranging from 5°to 15°. The method was applied to both the Xinlin Forest, China, and Santa Rosa National Park, Costa Rica. The performance of this method was validated by field measurement and Laser Vegetation Imaging Sensor data. The goodness of fit (R2) reached 0.73 and 0.78, respectively, and root-mean-squared errors (RMSEs) were 2.27 and 3.75 m over the two areas, respectively.

An ensemble method was used to combine three surface soil moisture products, retrieved from passive microwave remote sensing data, to reconstruct a monthly soil moisture data set for China between 2003 and 2010. Using the ensemble data set, the temporal and spatial variations of surface soil moisture were analyzed. The major findings were: 1) The ensemble data set was able to provide more realistic soil moisture information than individual remote sensing products; 2) The soil moisture variation trends derived from the three retrieval products and the ensemble data differ from each other but all data sets show the dominant drying trend for the summer, and that most of the drying regions were in major agricultural areas; 3) Combining soil moisture trends with land surface temperature trends derived from Moderate Resolution Imaging Spectroradiomete, the study domain was divided into four categories. Regions with drying and warming trends cover 33.2%, the regions with drying and cooling trends cover 27.4%, the regions with wetting and warming trends cover 21.1% and the regions with wetting and cooling trends cover 18.1%. The first two categories primarily cover the major grain producing areas, while the third category primarily covers non arable areas such as Northwest China and Tibet. This implies that the moisture and heat variation trends in China are unfavorable to sustainable development and ecology conservation.

To provide high-quality data for time-series change detection of lake water level, an automatic and robust algorithm for lake water footprint (LWF) identification is developed. Based on the Ice, Cloud, and Land Elevation Satellite GLA14 data file, six parameters were taken as features of an algorithm for LWF identification, and they are elevation difference between adjacent footprints, waveform width, number of peaks, reflectivity, kurtosis, and skewness of laser echoes. The sensitivity of each parameter was discussed, and elevation difference between adjacent footprints was proved to be most effective. The algorithm was described as a combination of these six parameters, and the thresholds of each parameter were set through statistics of LWF covering Peiku Co in Tibet, China, from 2003 to 2009. The performance of this classification algorithm was evaluated by the user's accuracy and producer's accuracy. Greater than 94% is achieved for all four tested lakes with 97% being the best result of producer's accuracy, and the user's accuracy ranges from 97.9% to 90% for these four lakes.

Although the Geoscience Laser Altimeter System (GLAS) onboard the NASA Ice, Cloud and Land Elevation Satellite was not designed for urban applications, its 3-D measurement capability over the globe makes it a nice feature for consideration in monitoring urban heights. However, this has not been previously done. In this paper, we report some preliminary assessment of the GLAS data for building height and density estimation in a suburb of Beijing, China. Building heights can be directly calculated from a GLAS data product (GLA14). Because GLA14 limits height levels to six in each ground footprint, we developed a new method to remove this restriction by processing the raw GLAS data. The maximum heights measured in the field at selected GLAS footprints were used to validate the GLAS measurement results. By assuming a constant incident energy and surface reflectance within a GLAS footprint, the building density can be estimated from GLA14 or from our newly processed GLAS data. The building density determined from high-resolution images in Google Earth was used to validate the GLAS estimation results. The results indicate that the newly developed method can produce more accurate building height estimation within each GLAS footprint ($R^{2} = 0.937$,$\hbox{rmse} = 6.4\ \hbox{m}$, and$n = 26$) than the GLA14 data product ($R^{2} = 0.808$,$\hbox{rmse} = 11.5\ \hbox{m}$, and$n = 26$). However, satisfactory estimation results on building density cannot be obtained from the GLAS data with the methods investigated in this paper. Forest cover could be a challenge to building height and density estimation from the GLAS data. It should be addressed in future research.

Introducing the sensor spectral response into the wavelet-based image fusion methods may produce images closer to which obtained by the ideal sensor. But the wavelet-based image fusion methods are complex because of the wavelet decomposition. This paper presents a simplified image fusion method using sensor spectral response which derives from a wavelet-based one, but need not to do the wavelet decomposition at the last calculation. The experimental results demonstrate that it provides good performance both in processing speed and image fusion quality.

Field spectrometer data and leaf area index (LAI) measurements were collected on the same day as the Earth Observing 1 satellite overpass for a study site in the Patagonia region of Argentina. We first simulated the total at-sensor radiances using MODTRAN 4 for atmospheric correction. Then ground spectroradiometric measurements were used to improve the retrieved reflectance for each pixel on the Hyperion image. Using the improved pixel-based surface reflectance spectra, 12 two-band "vegetation indices (VIs)" were constructed using all available 168 Hyperion bands. Finally, we evaluated the correlation of each possible vegetation index with LAI measurements to determine the most effective bands for forest LAI estimation. The experimental results indicate that most of the important hyperspectral bands with high R/sup 2/ are related to bands in the shortwave infrared (SWIR) region and some in the near-infrared (NIR) region. The bands are centered near 820, 1040, 1200, 1250, 1650, 2100, and 2260 nm with bandwidths ranging from 10-300 nm. It is notable that the originally defined VIs that use red and NIR bands did not produce higher correlation with LAI than VIs constructed with bands in SWIR and NIR regions.

A correlation analysis was conducted between forest leaf area index (LAI) and two red edge parameters: red edge position (REP) and red well position (RWP), extracted from reflectance image retrieved from Hyperion data. Field spectrometer data and LAI measurements were collected within two days after the Earth Observing One satellite passed over the study site in the Patagonia region of Argentina. The two red edge parameters were extracted with four approaches: four-point interpolation, polynomial fitting, Lagrangian technique, and inverted-Gaussian (IG) modeling. Experimental results indicate that the four-point approach is the most practical and suitable method for extracting the two red edge parameters from Hyperion data because only four bands and a simple interpolation computation are needed. The polynomial fitting approach is a direct method and has its practical value if hyperspectral data are available. However, it requires more computation time. The Lagrangian method is applicable only if the first derivative spectra are available; thus, it is not suitable to multispectral remote sensing. The IG approach needs further testing and refinement for Hyperion data.

Data acquired with Advanced Land Imager (ALI), Hyperspectral Imager (Hyperion) and Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), were used to estimate and map forest LAI. Analysis methods include 1) simulating the total at-sensor radiances using MODTRAN4, 2) modifying retrieved surface reflectance with ground spectroradiometric measurements, 3) constructing 6-term LAI prediction models to predict pixel-based LAI, and 4) mapping LAI. The experimental results indicate that the retrieval of surface reflectance is the most successful with AVIRIS, followed by Hyperion and ALI. AVIRIS data can produce more reasonable LAI map than the other two sensors. The results also indicate that Hyperion data have potentially extensive application values in bio-parameter extraction at varied scales.