Longwave infrared (LWIR) sensors are widely used for measuring land surface radiation in ground and unmanned aerial vehicle (UAV) remote sensing missions. Although the land surface temperature (LST) retrieval algorithms for thermal in-frared (TIR) satellite sensors with narrow spectral response ranges have achieved good results, they are generally unsuitable for LWIR sensors. At present, the LST retrieval algorithm for LWIR data needs further investigation. In this study, an im-proved radiative transfer (IRT) algorithm based on the segmen-tation of spectral response function (SRF) is proposed for retriev-ing LST from LWIR data. The IRT algorithm is applied to three types of commonly used LWIR sensors. The simulation results show that the root-mean-squared error (RMSE) is lower than 0.1 K when the segmentation width is 0.2 μm. The higher the height of the sensor, the more obvious the fluctuation of the accuracy increases with the segmentation width. Using thein-situdata of the Heihe River basin (HRB) for validation, RMSEs are between 1.1 and 1.8 K, depending on different land cover types. The IRT algorithm can retrieve the relatively high-accuracy LSTs from LWIR data observed by a variety of LWIR sensors, and promote the collaborative application of multi-sensor LSTs, which is of great significance in ecological environment research.

With the development of deep learning, significant progress has been made in change detection (CD) methods for remote sensing (RS) images. However, many convolutional neural network (CNN)-based methods are constrained in capturing long-range dependencies due to the limitations of the receptive field. Transformers rely on self-attention mechanisms to effectively achieve global information modeling and are widely used in CD tasks. Nevertheless, transformer-based CD methods still suffer from issues such as pseudochanges and incomplete edges due to the lack of position and temporal correlations in bitemporal RS images. To deal with this issue, we propose a position-temporal awareness transformer (PT-Former), which models position and temporal relations in bitemporal images. Specifically, a Siamese network attached to a position-aware embedding module (PEM) serves as a feature encoder to extract the features of changed areas. Then, a temporal difference perception module (TDPM) is designed to capture the cross-temporal shift and enhance the difference perception ability during cross-temporal interaction. Meanwhile, the contextual information of the ground object is aggregated by the fusion block, and the spatial relation is reconstructed under the guidance of bitemporal features. The experimental results validate the superiority of PT-Former on three benchmark datasets, including the season-varying CD (SVCD) dataset, the learning vision and RS laboratory building CD (LEVIR-CD) dataset, and the WHU-CD dataset confirming the potential of PT-Former for CD tasks in RS images. The code will be available athttps://github.com/liuyk29/PT-Former.

Pan-sharpening aims to increase the spatial resolution of the low-resolution multispectral (LrMS) image with the guidance of the corresponding panchromatic (PAN) image. Although deep learning (DL)-based pan-sharpening methods have achieved promising performance, most of them have a two-fold deficiency. For one thing, the universally adopted black box principle limits the model interpretability. For another thing, existing DL-based methods fail to efficiently capture local and global dependencies at the same time, inevitably limiting the overall performance. To address these mentioned issues, we first formulate the degradation process of the high-resolution multispectral (HrMS) image as a unified variational optimization problem, and alternately solve its data and prior subproblems by the designed iterative proximal gradient descent (PGD) algorithm. Moreover, we customize a Local-Global Transformer (LGT) to simultaneously model local and global dependencies, and further formulate an LGT-based prior module for image denoising. Besides the prior module, we also design a lightweight data module. Finally, by serially integrating the data and prior modules in each iterative stage, we unfold the iterative algorithm into a stage-wise unfolding network, Local-Global Transformer Enhanced Unfolding Network (LGTEUN), for the interpretable MS pan-sharpening. Comprehensive experimental results on three satellite data sets demonstrate the effectiveness and efficiency of LGTEUN compared with state-of-the-art (SOTA) methods. The source code is available at https://github.com/lms-07/LGTEUN.

Acquiring the satellite land surface temperature (LST) with high spatiotemporal resolutions is pressing in the land surface biophysical process. However, most current LST products hardly satisfy this requirement. LST Downscaling provides an effective way to solve this issue by introducing driving factors, but existing methods usually ignore the weights of descriptors. In this letter, based on the Geographically Weighted Regression (GWR) and Random Forest (RF), a new downscaling method (i.e., WGWR) considering the weights of LST descriptors is proposed. To examine the performance of WGWR, the 100-m Landsat-8 TIRS and Terra ASTER LSTs are aggregated to 1000 m as the simulated coarse LSTs, and then the coarse LSTs are downscaled to 100 m using WGWR, RF, and GWR. Meanwhile, the original 100-m LSTs are used as validation references. Results indicate that the proposed WGWR outperforms RF and GWR: for RF (GWR), the RMSEs can be reduced by 0.34 K (0.26 K) in Zhangye and 0.22 K (0.1 K) in Beijing. Compared to RF and GWR, WGWR also yields better image quality: the downscaled LST images have neither obvious smoothing effect nor boundary effect and maintain the details of the image at high spatial resolution. Validation based onin-situLST indicates that the downscaled LST based on WGWR has better agreement with thein-situLST, and the RMSE is reduced by 0.57 K. The proposed WGWR contributes to obtain high spatio-temporal resolution LSTs and promote hydrological, meteorological, and ecological studies.

Lunar exploration has put further demands on high spatial resolution of the lunar surface temperature (LuST) data. However, available LuST data with coarse resolution provided by current remote sensing platforms (e.g., LRO Diviner) are unable to meet the requirement. Therefore, it is necessary to develop new methods to obtain LuST data with high spatial resolution. Inspired by the study of land surface temperature (LST) downscaling, here, we perform the exploratory study of the LuST downscaling. The core of the exploration work consists of two parts: 1) the 16 pixels per degree (ppd) Diviner Gridded Data Records (GDRs) images are downscaled to 128 ppd using a downscaling model based on geographically weighted regression (GWR), in order to explore the feasibility of downscaling LuST; and 2) two sets of experiments with scales of 4 and 8 are conducted and the growth characteristics of the errors are observed to determine whether it is feasible to obtain higher-spatial-resolution data. The evaluation metrics for the experiments in two specific craters with scale of 8 are 4.03/5.42 K for mean absolute error, 5.89/8.05 K for root mean square error and 0.037/0.048 for normalized root mean square error, respectively. The LuST downscaling study is feasible, and it provides an effective way for obtaining high-spatial-resolution LuST data, which has positive implications for lunar exploration.

To mine the spectral-spatial information of target pixel in hyperspectral image classification (HSIC), convolutional neural network (CNN)-based models widely adopt patch-based input pattern, where a patch represents its central pixel and the neighbor pixels play auxiliary roles in the classification process. However, compared to the central pixel, its neighbor pixels often have different contributions for classification. Although many existing patch-based CNNs could adaptively emphasize the spatial neighbor information, most of them ignore the latent relationship between the center pixel and its neighbor pixels. Moreover, efficient spectral-spatial feature extraction has been a difficult yet vital topic for HSIC. To address the mentioned problems, a central vector oriented self-similarity network (CVSSN) is proposed for HSIC. Specifically, based on two similarity measures, we firstly design an adaptive weight addition based spectral vector self-similarity module (AWA-SVSS) in input space and a Euclidean distance based feature vector self-similarity module (ED-FVSS) in feature space to fully mine the central vector oriented spatial relationships. Besides, a spectral-spatial information fusion module (SSIF) is formulated as a new pattern to fuse the central 1D spectral vector and the corresponding 3D patch for efficient spectral-spatial feature learning of the subsequent modules. Moreover, we implement a channel spatial separation convolution module (CSS-Conv) and a scale information complementary convolution module (SIC-Conv) for efficient spectral-spatial feature learning. Extensive experimental results on four popular HSI data sets demonstrate the effectiveness and efficiency of the proposed method compared with other state-of-the-art methods. The source code is available athttps://github.com/lms-07/CVSSN

For the abundant spectral and spatial information recorded in hyperspectral images (HSIs), fully exploring spectral-spatial relationships has attracted widespread attention in hyperspectral image classification (HSIC) community. However, there are still some intractable obstructs. For one thing, in the patch based processing pattern, some spatial neighbor pixels are often inconsistent with the central pixel in land-cover class. For another thing, linear and nonlinear correlations between different spectral bands are vital yet tough for representing and excavating. To overcome these mentioned issues, an adaptive mask sampling and manifold to Euclidean subspace learning (AMS-M2ESL) framework is proposed for HSIC. Specifically, an adaptive mask based intra-patch sampling (AMIPS) module is firstly formulated for intra-patch sampling in an adaptive mask manner based on central spectral vector oriented spatial relationships. Then, based on distance covariance descriptor, a dual channel distance covariance representation (DC-DCR) module is proposed for modeling unified spectral-spatial feature representations and exploring spectral-spatial relationships, especially linear and nonlinear interdependence in spectral domain. Furthermore, considering that distance covariance matrix lies on the symmetric positive definite (SPD) manifold, we implement a manifold to Euclidean subspace learning (M2ESL) module respecting Riemannian geometry of SPD manifold for high-level spectral-spatial feature learning. Additionally, we introduce an approximate matrix square-root (ASQRT) layer for efficient Euclidean subspace projection. Extensive experimental results on three popular HSI data sets with limited training samples demonstrate the superior performance of the proposed method compared with other state-of-the-art methods. The source code is available at https://github.com/lms-07/AMS-M2ESL.

Semantic segmentation is an extremely challenging task in high-resolution remote sensing (HRRS) images as objects have complex spatial layouts and enormous variations in appearance. Convolutional neural networks (CNNs) have excellent ability to extract local features and have been widely applied as the feature extractor for various vision tasks. However, due to the inherent inductive bias of convolution operation, CNNs inevitably have limitations in modeling long-range dependencies. Transformer can capture global representations well, but unfortunately ignores the details of local features and has high computational and spatial complexity in processing high-resolution feature maps. In this paper, we propose a novel hybrid architecture for HRRS image segmentation, termed EMRT, to exploit the advantages of convolution operations and Transformer to enhance multi-scale representation learning. We incorporate the deformable self-attention mechanism in the Transformer to automatically adjust the receptive field, and design an encoder-decoder architecture accordingly to achieve efficient context modeling. Specifically, the CNN is constructed to extract feature representations. In the encoder, local features and global representations at different resolutions are extracted by the CNN and Transformer, respectively, and fused in an interactive manner. Moreover, a separate spatial branch is designed to extract multi-scale contextual information as queries, and global dependencies between features at different scales are efficiently established by the decoder. Extensive experiments on three public remote sensing datasets demonstrate the superiority of EMRT and indicate that the overall performance of our method outperforms state-of-the-art methods. Code is available at https://github.com/peach-xiao/EMRT.

Unmanned aerial vehicle (UAV) thermal infrared (TIR) remote sensing is an important way to obtain land surface temperature (LST) with high spatial and temporal resolutions. Due to wide spectral response function (SRF) ranges of UAV thermal imagers, currently available LST retrieval methods suitable for satellite sensors may induce significant uncertainty when applied to UAV sensors. Despite that some methods have been proposed to retrieve LST from UAV remote sensing, studies considering the adverse effect caused by the SRF ranges are still rare. Here, we present a so-called Temperature Retrieval for UAV Broadband thermal imager data (TRUB) method to retrieve LST from UAV broadband thermal imager data. TRUB’s core includes two parts: 1) a simple lookup table (LUT) algorithm for reducing the uncertainty induced by the wide SRF ranges; and 2) models suitable for UAV remote sensing for estimating the atmospheric parameters. Validation from the Heihe River Basin shows that the LST retrieved by TRUB, of which the root mean square error (RMSE) and mean bias error (MBE) is 1.71 and −0.02 K, respectively, is highly consistent with thein situLST. TRUB is helpful to reduce the uncertainty caused by the wide SRF ranges of UAV thermal imagers and quantify the influence of atmosphere, thus can obtain UAV remote-sensing LST with better accuracy in large-area operating missions.

Land surface temperature (LST) is an important parameter for series land surface processes, models and applications. The accuracy of LST directly influenced its application. Therefore, a reasonable validation method is meaningful to assess the accuracy of LST datasets. In this study, an in-situ observation representativeness assessment method was proposed. Based on this method, the JPSS VIIRS LST product was validated against in-situ LST at 7 ground sites over the Heihe River Basin during the HiWATER experiments period. Results show that about 70%, 28%, 43%, 68%, 42%, 35% and 25% of the FOV LST for ARS, DMS, DSL, EBO, HHL, JYL, and SDQ are able to well represent the corresponding LST of VIIRS pixels, respectively and determined the representativeness period of each site. The VIIRS LST has a high correlation with the in-situ LST with an accuracy of 2.30 K - 5.76 K at daytime and 1.26 K-2.68 K at nighttime, respectively.

A method of decreasing the uncertainty in validating moderate resolution remotely sensed Land surface temperature (LST) was established on a heterogeneity surface in Northwest China. Two Landsat-8 images and in situ measured data were used in the paper. The LST and multiband reflectance as well as the reflectance coefficient of variation have a significant regression relation. For two Landsat-8 images, the determination coefficients were both above 0.88, and root mean square errors (RMSEs) were 1.2 K and 1.5 K respectively. Meanwhile, the regression relation isn't sensitive to the change of spatial resolution. High resolution LST which was obtained by using the regression relation and in situ measured LST were used in validating the Moderate Resolution Imaging Spectroradiometer (MODIS) daily LST product MOD11A1. The final result shown that the method can effectively decrease the validation uncertainty.

Remotely sensed land surface temperature (LST) is influenced by the viewing angles and ground structure. By selecting a sparsely vegetated surface as the study area, the effects of structural parameters of land surface on remotely sensed LST are analyzed in this paper. Results demonstrate that both the density of canopy and size of canopy have significant influences on the remote sensing observations of LST. The difference between the LSTs obtained at nadir and off-nadir views also varies according to the acquisition season. This work is expected to be beneficial for quality control of remotely sensed LST.

With the rapid development of new satellite thermal sensors and applications of land surface temperature (LST), research on finding effective algorithms to retrieve accurate LST from satellite thermal infrared (TIR) data is becoming more and more important. In this study, multiple algorithms for retrieving LST from Landsat-8 Thermal Infrared Sensor (TIRS) data are validated and intercompared in an extremely arid region, Northwest China. According to the validation and intercomparison, we find that the radiative transfer equation (RTE) based method with TIRS band 1 (10.60-11.19 μm) has the highest accuracy, while the single-channel (SC) method using TIRS band 2 (11.50-12.51 μm) yielded the lowest accuracy. The accuracies of split-window (SW) algorithms are slightly lower than the RTE based method. However, the SW algorithms have better applicability than the RTE based method. The most suitable SW algorithm for Landsat-8 TIRS data in the study area is recommended. This study will be beneficial for developing the LST product from Landsat-8 data for the study area.

Although many challenges exist, validation of the satellite land surface temperature (LST) product over heterogeneous surface can provide new and in-depth understandings of the product. Lessons learned from the validation are important to improve the satellite LST product. In order to better understand the relationship between LSTs measured through different approaches and instruments and test the possibility to upscale the ground measured LST over heterogeneous surface, a MUlti-Scale Observation Experiment on land Surface temperature (MUSOES) was designed and conducted in an extremely arid region in Northwest China. The experiment was concentrated at two typical sites (i.e. HHL - sparsely forest, and SDQ - open shrubland). It began from July 2014 and have run normally for two years. First results of this experiment have been presented here. MUSOES provides a basis to examine the upscaling of the ground measured LST to the LST at the satellite pixel scale over the heterogeneous surface.

Airborne TIR remote sensing can obtain land surface temperature (LST) with high spatial resolution. However, the swath width of airborne stripes is usually limited. Therefore, it is necessary to generate the LSTs for a large area through temporal normalization of LSTs derived from different stripes. By selecting an agricultural oasis as the study area, this study compares the diurnal temperature cycle (DTC) model and polynomial regression (PR) technique in the temporal normalization of the LSTs derived from the Thermal Airborne Spectrographic Imager (TASI) data. The results show that the DTC model has better accuracy in normalizing the LSTs. However, the PR technique is simple and requires less ancillary data. The DTC method can normalize the LST to any specific time and generate temporally continuous LSTs, while the PR method can only do relative normalization. This study is helpful to reduce the temperature differences of different airborne stripes and obtain airborne LSTs with both high spatial and temporal resolutions.

Thermal cameras are helpful instruments for measuring surface temperatures in field experiments. However, previous studies haven't detailed the method of deriving component temperatures of vegetation and soil over heterogeneous surfaces. In addition, the sources contributing to uncertainties in the derived component temperatures require further investigation. We present a study wherein the component temperatures of a dynamically developing maize field were derived from thermal images. The sources influencing the derived component temperatures have been investigated and different parameterization schemes for atmospheric downwelling radiation have been compared. The results demonstrate that the thermal cameras provide a feasible method of deriving the component temperatures. If the thermal camera is mounted at approximately 30 m above the target and then the atmospheric upwelling radiation and transmittance is ignored, a 1.0-2.0 K error for the component temperatures may occur.

The two-source energy balance (TSEB) model was successfully applied to estimate evaporation (E), transpiration (T), and evapotranspiration (ET) for land covered with vegetation, which has significantly important applications for the terrestrial water cycle and water resource management. However, the current composite temperature separation approaches are limited in their effectiveness in arid regions. Moreover, E and T are difficult to measure on the ground. In this letter, the ground-measured soil and canopy component temperatures were used to estimate E, T, and ET, which were better validated with observed ratios of E (E/ET%) and T (T/ET%) using the stable oxygen and hydrogen isotopes, and the ET measurements using an eddy covariance (EC) system. Our results indicated that even under the strongly advective conditions, the TSEB model produced reliable estimates of the E/ET% and T/ET% ratios and of ET. The mean bias and root-mean-square error (RMSE) of E/ET% were 1% and 2%, respectively, and the mean bias and RMSE of T/ET% were -1% and 2%, respectively. In addition, the model exhibited relatively reliable estimates in the latent heat flux, with mean bias and RMSE values of 31 and 61 W · m-2, respectively, compared with the measurements from the EC system. These results demonstrated that a robust soil and vegetation component temperature calculation was crucial for estimating E, T, and ET. Moreover, the separate validation of E/ET% and T/ET% provides a good prospect for TSEB model improvements.