Satellite Datasets Research Articles

BDCN_UNet: Advanced shoreline extraction techniques integrating deep learning

The delineation of shorelines, marking the dynamic boundary between land and sea, is crucial for coastal management and ecological conservation. This research investigates various methods for automatic shoreline extraction from satellite images, utilizing cost-effective and regularly available remote sensing data. A deep learning framework is used for the comparison that combines semantic segmentation (UNEt) and edge detection (Bi-Directional Cascade Network, or BDCN) to make the results more accurate. The study focuses on the northern coast of Egypt, a region with diverse landscapes and significant human interventions. We’ve made progress in our methods by using advanced image processing, the CoastSat Toolkit, the BDCN_UNet framework, time series analysis, and deep learning networks to look at and keep an eye on Egypt’s North Coast’s future plans in more detail. This study captures the intricate interplay between natural and human factors in coastal environments, offering valuable insights for sustainable coastal development. The performance analysis of four models (UNet, Pyramid Scene Parsing Network, DeepLabV3, and BDCN_UNet) applied to three satellite datasets (Sentinel-2, PlanetScope, and Pleiades) showed that PlanetScope and Pleiades consistently yield higher accuracy (0.90–0.99), while Sentinel-2 proved more challenging, particularly for Pyramid Scene Parsing Network and DeepLabV3. UNet and BDCN_UNet exhibited robust performance across all datasets, especially with Pleiades delivering the best results overall.

Clouds reduce downwelling longwave radiation over land in a warming climate.

Clouds greatly influence the Earth's energy balance1,2. Observationally constraining cloud radiative feedback, a notably uncertain climate feedback mechanism3-5, is crucial for improving predictions of climate change5-7 but, so far, remains an elusive objective, and the feedback may be different over the ocean versus over land8-10. Here we show a local negative surface longwave cloud feedback over land at the southern Great Plains site, constrained by direct long-term observation of spectrally resolved downwelling longwave radiance11. This negative cloud feedback at thesouthern Great Plains site causes a -1.77 ± 1.15 W m-2 per decade change in downwelling longwave radiation and suggests that cloud changes may partially modulate the warming effect of increased greenhouse gas concentrations and atmospheric temperatures over land. Specifically, our results are derived from an optimal spectral fingerprinting method12-15 designed to separate surface longwave cloud feedback from other surface forcings and feedbacks, by making use of their unique spectral signatures13-18 in the long-term record of spectrally resolved radiances. Furthermore, we show that the results are not sitespecific: negative surface longwave cloud feedbacks, primarily induced by decreasing low cloud cover in warming climates, are commonly observed over land in reanalysis and satellite datasets. Our findings establish a pivotal observational benchmark of radiative forcing and feedback needed for validating climate model performance over land.

Application of Google Earth Engine to Monitor Greenhouse Gases: A Review

Google Earth Engine (GEE) is a cloud-based platform revolutionizing geospatial analysis by providing access to vast satellite datasets and computational capabilities for monitoring environmental and societal issues. It incorporates machine learning (ML) techniques and algorithms as part of its tools for analyzing and processing large geospatial data. This review explores the diverse applications of GEE in monitoring and mitigating greenhouse gas emissions and uptakes. GEE is a cloud-based platform built on Google’s infrastructure for analyzing and visualizing large-scale geospatial datasets. It offers large datasets for monitoring greenhouse gas (GHG) emissions and understanding their environmental impact. By leveraging GEE’s capabilities, researchers have developed tools and algorithms to analyze remotely sensed data and accurately quantify GHG emissions and uptakes. This review examines progress and trends in GEE applications, focusing on monitoring carbon dioxide (CO2), methane (CH4), and nitrous oxide/nitrogen dioxide (N2O/NO2) emissions. It discusses the integration of GEE with different machine learning methods and the challenges and opportunities in optimizing algorithms and ensuring data interoperability. Furthermore, it highlights GEE’s role in pinpointing emission hotspots, as demonstrated in studies monitoring uptakes. By providing insights into GEE’s capabilities for precise monitoring and mapping of GHGs, this review aims to advance environmental research and decision-making processes in mitigating climate change.

Attributing rainfall and drought variability across climate vulnerable area of Pakistan: Perspective from different satellite and ground-based datasets

Attributing rainfall and drought variability across climate vulnerable area of Pakistan: Perspective from different satellite and ground-based datasets

Urban tree species classification using multisource satellite remote sensing data and street view imagery

ABSTRACT Urban tree species distribution plays an important role in supporting both the urban forest carbon stocks estimation and sustainable urban ecological planning. Practical constraints, such as high labor costs and limited data coverage, yet present a major challenge in rapid census and classification of urban tree species at city scale. This paper aims to address this requirement by developing a large-scale urban tree species classification model through the combination of Global Ecosystem Dynamics Investigation (GEDI) data, GF-5 images, GF-6 images and Street View Imagery (SVI). The model employs Random Forest (RF) algorithm and incorporates reference data from both conventional forest inventory, comprising 866 samples and a large collection of virtual tree inventory plot automatically generated from SVI, totaling 8681 samples. From the results, the model achieves an overall accuracy of 77.89% and Kappa coefficient of 0.75. The model shows better recognition for tree species with unique crown morphology and clustered distributions. The finding indicates that satellite datasets can provide statistically significant contributions to urban tree species classification. Visible light spectrum, leaf-area and tree height metrics, as well as climate and geographic location metrics, exhibit notable value in the classification. Overall, this study can provide a viable research pathway for extending urban tree species classification to a larger scale.

GAN-Assisted Road Segmentation from Satellite Imagery

Geo-information extraction from satellite imagery has become crucial to carry out large-scale ground surveys in a short amount of time. With the increasing number of commercial satellites launched into orbit in recent years, high-resolution RGB color remote sensing imagery has attracted a lot of attention. However, because of the high cost of image acquisition and even more complicated annotation procedures, there are limited high-resolution satellite datasets available. Compared to close-range imagery datasets, existing satellite datasets have a much lower number of images and cover only a few scenarios (cities, background environments, etc.). They may not be sufficient for training robust learning models that fit all environmental conditions or be representative enough for training regional models that optimize for local scenarios. Instead of collecting and annotating more data, using synthetic images could be another solution to boost the performance of a model. This article proposes a GAN-assisted training scheme for road segmentation from high-resolution RGB color satellite images, which includes three critical components: (a) synthetic training sample generation, (b) synthetic training sample selection, and (c) assisted training strategy. Apart from the GeoPalette and cSinGAN image generators introduced in our prior work, this article explains in detail how to generate new training pairs using OpenStreetMap (OSM) and introduces a new set of evaluation metrics for selecting synthetic training pairs from a pool of generated samples. We conduct extensive quantitative and qualitative experiments to compare different image generators and training strategies. Our experiments on the downstream road segmentation task show that (1) our proposed metrics are more aligned with the trained model performance compared to commonly used GAN evaluation metrics such as the Fréchet inception distance (FID); and (2) by using synthetic data with the best training strategy, the model performance, mean Intersection over Union (mean IoU), is improved from 60.92% to 64.44%, when 1,000 real training pairs are available for learning, which reaches a similar level of performance as a model that is standard-trained with 4,000 real images (64.59%), i.e., enabling a 4-fold reduction in real dataset size.

Two-Level Supervised Network for Small Ship Target Detection in Shallow Thin Cloud-Covered Optical Satellite Images

Ship detection under cloudy and foggy conditions is a significant challenge in remote sensing satellite applications, as cloud cover often reduces contrast between targets and backgrounds. Additionally, ships are small and affected by noise, making them difficult to detect. This paper proposes a Cloud Removal and Target Detection (CRTD) network to detect small ships in images with thin cloud cover. The process begins with a Thin Cloud Removal (TCR) module for image preprocessing. The preprocessed data are then fed into a Small Target Detection (STD) module. To improve target–background contrast, we introduce a Target Enhancement module. The TCR and STD modules are integrated through a dual-stage supervision network, which hierarchically processes the detection task to enhance data quality, minimizing the impact of thin clouds. Experiments on the GaoFen-4 satellite dataset show that the proposed method outperforms existing detectors, achieving an average precision (AP) of 88.9%.

Comparative Trend Analysis of Precipitation Indices in Several Towns of the Sirba River Catchment (Burkina Faso) from CHIRPS and TAMSAT Rainfall Estimates

The increasingly frequent pluvial flood of West African urban settlements indicates the need to investigate the drivers of local rainfall changes. However, meteorological stations are few, unevenly distributed, and work irregularly. Daily satellite rainfall datasets can be used. Nevertheless, these products often need to be more accurate due to sensor errors and limitations in retrieval algorithms. The problem is, therefore, how to characterize rainfall where there is a need for ground-based rainfall records or incomplete series. This study aims to characterize urban rainfall using two satellite datasets. The analysis was carried out in the Sirba river catchment, Burkina Faso, using the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) and the Tropical Applications of Meteorology using SATellite and ground-based data (TAMSAT) datasets. Ten indices from the Expert Team on Climate Change Detection and Indices (ETCCDI) of precipitation were calculated, and their statistical trends were evaluated from 1983 to 2023. The study introduces two key innovations: a comparative analysis of precipitation trends using two satellite datasets and applying this analysis to towns within a previously understudied 39,138 km2 catchment area that is frequently flooded. Both datasets agree on the increase of (i) annual cumulative rainfall over all towns, (ii) five-day maximum rainfall over the town of Manni, (iii) rainfall due to very wet days in Gayéri, (iv) days of heavy rainfall in Bogandé, Manni and Yalgho, and (v) days of very heavy rainfall in Yalgho. These findings suggest the need for targeted pluvial flood prevention measures in towns with increasing trends in heavy rainfall.

Particulate matter estimation using satellite datasets: a machine learning approach.

In the present work, it is the first time an interpretable machine learning model has been developed for the estimation of Particulate Matter 10µm (PM10) concentrations over India using Aerosol Optical Depth (AOD) from two different satellites, i.e. INSAT-3D and Moderate Resolution Imaging Spectroradiometer (MODIS) for the period of 7years (2014 to 2020). Ground datasets of AOD are taken from the Aerosol Robotic Network (AERONET) for the validation of satellite-retrieved AOD. The observation of particulate matter (PM) data is acquired from the Central Pollution Control Board (CPCB) station across India. Analysis has been performed on a monthly basis for the given time period. The result shows that AOD products of MODIS exhibit good correlation with AERONET AOD whereas INSAT-3D AOD is not well correlated with AERONET AOD. However, after applying an error envelope and threshold-based filtering technique, we have found that INSAT-3D shows significant correlation with ground-level AOD with approximate correlation of 0.66 for Jaipur and 0.57 for Kanpur exhibiting almost similar performance as MODIS-derived AOD. Satellite AOD data together with ground PM concentration data is used to train the machine learning model (random forest) for the estimation of the PM distribution across India for the year 2020. An encouraging correlation of R-squared (R2) value 0.78 has been observed between the estimated and observed PM10 concentrations. The model demonstrates effective training, mitigating huge overestimation and underestimation. However, despite closely tracking the trends of estimated PM10 with observed PM10, few instances of overestimation persist. This suggests the need for an expanded training dataset to further refine and enhance the model's accuracy. Finally, the machine learning model used for PM10 estimation is found to be optimal for a calibrated satellite AOD product.

A Multitask Network for Joint Multispectral Pansharpening on Diverse Satellite Data.

Despite the rapid advance in multispectral (MS) pansharpening, existing convolutional neural network (CNN)-based methods require training on separate CNNs for different satellite datasets. However, such a single-task learning (STL) paradigm often leads to overlooking any underlying correlations between datasets. Aiming at this challenging problem, a multitask network (MTNet) is presented to accomplish joint MS pansharpening in a unified framework for images acquired by different satellites. Particularly, the pansharpening process of each satellite is treated as a specific task, while MTNet simultaneously learns from all data obtained from these satellites following the multitask learning (MTL) paradigm. MTNet shares the generic knowledge between datasets via task-agnostic subnetwork (TASNet), utilizing task-specific subnetworks (TSSNets) to facilitate the adaptation of such knowledge to a certain satellite. To tackle the limitation of the local connectivity property of the CNN, TASNet incorporates Transformer modules to derive global information. In addition, band-aware dynamic convolutions (BDConvs) are proposed that can accommodate various ground scenes and bands by adjusting their respective receptive field (RF) size. Systematic experimental results over different datasets demonstrate that the proposed approach outperforms the existing state-of-the-art (SOTA) techniques.

Environmental Factors Influencing Anthrax Distribution in an East African Protected Area

ABSTRACTAnthrax outbreaks caused by the soil‐borne bacterium Bacillus anthracis have been known to occur often in Africa, impacting both wildlife and livestock and occasionally infecting humans. Modelling the current distribution and predicting suitable habitats for this bacteria species is therefore critical for supporting effective planning and control measures for anthrax outbreaks. Despite its impact, the knowledge on the ecology of this pathogen is still limited. This can be due to intricate interactions between anthropogenic and environmental factors and the variety of species it affects. Therefore, detailed, site‐specific analyses are essential for contributing to the understanding of the ecology of anthrax. This study used the maximum entropy modelling algorithm (MaxEnt) method to predict suitable habitat and environmental conditions that may support anthrax distribution and spore survival in Tarangire National Park, in Tanzania. Model inputs included 14 predictors from World Climatic, Landsat satellite, and World Soil Information datasets, as well as 636 presence‐only occurrence data from park records from 2013 to 2023. The predicted suitable area favouring anthrax spores encompassed most of the savannah and open grassland in the northern part of the park, as well as narrow patches in the park's central region. The mean test AUC score was 0.91, and the predicted probability was 0.98 for anthrax presence. On average, the predicted probability of suitable conditions substantially improved with increasing levels of extractable phosphorus, exchangeable potassium, mean annual temperature, and soil pH. Furthermore, the probability of suitable habitat was highest at lower levels of fire frequency, exchangeable sodium, and extractable aluminium. The extensive track record of anthrax outbreaks in the area may be explained by the presence of these favourable soil characteristics and climate, which point to the long‐term occurrence of anthrax spores in the park. However, the existence of suitable habitat in the northern part of the park presents opportunities for mobilising resources to mitigate the situation through activities such as targeted disease surveillance and the use of controlled fire.

Distinct Energy Budgets of Mars and Earth

AbstractThe radiant energy budget (REB) is a fundamental physical parameter for planetary bodies, though studies constraining the REB for bodies beyond Earth are relatively limited. We generate the first meridional profiles of Mars' REB at seasonal and annual timescales through measurements based on long term multi‐instrument observations from spacecraft orbiting Mars. Then, we compare our findings to Earth's REB using contemporary satellite data sets. Each planet exhibits remarkably distinct seasonal REB distributions due to differences in their orbital, atmospheric, and surface properties. Annually, Earth's REB exhibits a tropical energy surplus and a deficit at the poles. In contrast, Mars' annual REB displays an inverted meridional distribution with significant hemispheric asymmetry. Additionally, global dust storms significantly modify the Martian REB. Our observations are employable in future studies to improve models on Mars' general circulation, meteorology, and polar ice cap evolution.

Global Terrestrial Water–Energy Coupling Across Scales

ABSTRACTTerrestrial water–energy coupling (WEC), in the form of a non‐linear relationship between Soil Moisture (SM) and evaporative fraction (EF, ratio of actual and potential evapotranspiration), controls critical ecohydrological processes. We investigate and parameterize the evolution of global SM–EF coupling from the field to remote‐sensing (RS)‐pixel. The field‐scale EF and SM were obtained from eddy covariance (EC) and SM sensors at the global FLUXNET and Texas Water Observatory sites. RS‐pixel‐scale EF and SM estimates were obtained from Moderate‐resolution Imaging Spectroradiometer (MODIS) and Soil Moisture Active Passive (SMAP) sensors, respectively. We estimate the effective thresholds of the WEC regimes from both EC and satellite datasets to highlight the influence of sub‐pixel‐scale heterogeneity, and, scaling and observational constraints on the evolution of WEC regimes from the field to RS‐pixel scale. We argue that the changes in land surface conditions add a temporal variability in the critical thresholds of terrestrial WEC at RS pixel scale. We compare the critical WEC thresholds of the water‐ and energy‐limited regimes with a SM drydown‐based approach and highlight the similarities between both methods in partitioning dominant WEC regimes. EF and SM are strongly coupled in dryland arid and semi‐arid regions compared to humid climates. WEC regimes and thresholds have strong interseason variability due to dynamic interactions between soil, vegetation and atmosphere at the RS‐pixel scale. In contrast, field‐scale SM‐EF coupling is influenced predominantly by soil conditions and land‐use/management practices. Hence, future development of Earth‐system/Land‐surface models must account for the inter‐scale differences in the coupling between terrestrial water and energy fluxes representative of the ‘effective’ processes at large spatial scales.

Iterative Methods for Vecchia-Laplace Approximations for Latent Gaussian Process Models

Latent Gaussian process (GP) models are flexible probabilistic nonparametric function models. Vecchia approximations are accurate approximations for GPs to overcome computational bottlenecks for large data, and the Laplace approximation is a fast method with asymptotic convergence guarantees to approximate marginal likelihoods and posterior predictive distributions for non-Gaussian likelihoods. Unfortunately, the computational complexity of combined Vecchia-Laplace approximations grows faster than linearly in the sample size when used in combination with direct solver methods such as the Cholesky decomposition. Computations with Vecchia-Laplace approximations can thus become prohibitively slow precisely when the approximations are usually the most accurate, that is, on large datasets. In this article, we present iterative methods to overcome this drawback. Among other things, we introduce and analyze several preconditioners, derive new convergence results, and propose novel methods for accurately approximating predictive variances. We analyze our proposed methods theoretically and in experiments with simulated and real-world data. In particular, we obtain a speed-up of an order of magnitude compared to Cholesky-based calculations and a 3-fold increase in prediction accuracy in terms of the continuous ranked probability score compared to a state-of-the-art method on a large satellite dataset. All methods are implemented in a free C++ software library with high-level Python and R packages. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Holocene fringing reef along southern Andaman and Swaraj Dweep shoreline

The Andaman and Nicobar Islands are rimmed by discontinuous fringing reef that is in general wider on western margin vs the eastern margin. This study characterizes the facies updip from the modern fringing reefs to the present shoreline of south Andaman and Swaraj Dweep, and describes in detail the coral terraces/carpets within and above the inter-tidal zone representing the Holocene Fringing reef. Field studies, satellite, and drone datasets have been utilized to map different facies, that include: coralgal boundstone, biodetrital-grainstone, beachrock, and coralgal rudstone. Multiple exposed microatolls as well as coral terraces (coral carpets) of Acropora and Porites (dated 8.7-8.4 ka BP) have been identified within the intertidal zone (Radhanagar Beach, Swaraj Dweep) indicating that Holocene fringing reef have down-stepped offshore to the current location of modern fringing reefs owing to either tectonics or eustasy. The eustatic sea-level fluctuations are relatively well established for the Holocene and we compute the tectonic uplift rates utilizing the stream-power-incision and linear-inversion model. A tectonic uplift rate of ~ 0.05 mm/yr (for Swaraj Dweep) during the past 100 ka is estimated, while taking into account a wide range of erodibility indexes and response time intervals. It is identified that the computed uplift rate is an amalgamation of the coseismic deformation along with the interseismic and aseismic surface deformation. Thus, not all exposed coral terraces/microatolls are exposed due to coseismic deformation (for example uplift in parts of Andaman due to earthquake in 2004). The average long-term uplift rates are a magnitude lower than the eustatic sea-level fall rates during Holocene, thus, we suggest that most of the Holocene fringing reefs are exposed due to eustatic sea-level fall and down-stepped to the current location of the modern fringing reefs. This would entail that the eustatic sea-level change rates would play a significant role in determining future of the modern fringing reef (catch-up vs keep up vs give up), and the coastal morphology of south Andaman and Swaraj Dweep, with implications for coastal inundation and stability in the scenario of climate change.

Assessment of CCMP in Capturing High Winds with Respect to Individual Satellite Datasets

High-wind structures were identified in the Cross-Calibrated Multi-Platform (CCMP) ocean wind vector reanalysis for comparison with winds measured by satellite radiometers, scatterometers, and synthetic aperture radar (SAR) instruments from February to October 2023. The comparison aims to evaluate bias, uncertainty, and spatial correlations with the goal of enhancing the accuracy of ocean wind datasets during tropical cyclones (TCs). In 10° longitude × 10° latitude blocks, each containing a TC, Soil Moisture Active Passive (SMAP) and Advanced Microwave Scanning Radiometer 2 (AMSR2) winds are 6.5 and 4.8% higher than CCMP, while Advanced Scatterometer (ASCATB) is 0.8% lower. For extratropical cyclones, AMSR2 and SMAP also show stronger winds with a 5% difference, and ASCATB is about 0.3% weaker compared to CCMP. The comparison between SAR and CCMP for TC winds, sampled at the locations and time frames of SAR tiles, indicates that SAR winds around TCs are about 9% higher than CCMP winds. Using empirically defined TC structural indices, we find that the TCs observed by CCMP are shifted in locations and lack a compact core region. A Random Forest (RF) regressor was applied to TCs in CCMP with corresponding SAR observations, nearly correcting the full magnitude of low bias in CCMP statistically, with a 15 m/s correction in the core region. The hierarchy of importance among the predictors is as follows: CCMP wind speed (62%), distance of SAR pixels to the eye region (21%) and eye center (7%), and distance of CCMP pixels to the eye region (5%) and eye center (5%).

Assessment of global carbon dynamics due to mining-induced forest cover loss during 2000–2019 using satellite datasets

Mining activities significantly contribute to forest cover loss (FCL), subsequently altering global carbon dynamics and exacerbating climate change. The present study aims to estimate the contributions of mining-induced FCL to carbon sequestration loss (CSL) and carbon dioxide (CO₂) emissions from 2000 to 2019 using the proxy datasets. For FCL analysis, the global FCL data at 30 m spatial resolution, developed by Hansen et al. (2013), was employed in the Google Earth Engine (GEE) cloud platform. Furthermore, for CSL and CO₂ emissions assessment, Moderate Resolution Imaging Spectroradiometer (MODIS)-based Net Primary Productivity (NPP) data and Zhang and Liang (2020)-developed biomass datasets were used, respectively. The outcomes of the study exhibited approximately 16,785.90 km2 FCL globally due to mining activities, resulting in an estimated CSL of ∼36,363.17 Gg CO₂/year and CO₂ emissions of ∼490,525.30 Gg CO₂. Indonesia emerged as the largest contributor to mining-induced FCL, accounting for 3,622.78 km2 of deforestation, or 21.58% of the global total. Brazil and Canada followed, with significant deforestation and CO₂ emissions. The findings revealed that mining activities are a major driver of deforestation, particularly in resource-rich regions, leading to substantial environmental degradation. The relative FCL was notably high in smaller countries like Suriname and Guyana, where mining activities constituted a significant proportion of total deforestation. The present study underscores the urgent need for robust regulatory frameworks, sustainable land management practices, and coordinated international efforts to mitigate the adverse environmental impacts of mining. The findings of this study can inform policymakers and stakeholders, leading to more effective conservation strategies and benefiting society by promoting environmental sustainability and resilience against climate change.

Evaluation of Three High-Resolution Satellite and Meteorological Reanalysis Precipitation Datasets over the Yellow River Basin in China

Evaluation of Three High-Resolution Satellite and Meteorological Reanalysis Precipitation Datasets over the Yellow River Basin in China

Research on Gating Fusion Algorithm for Power Grid Survey Data Based on Enhanced Mamba Spatial Neighborhood Relationship.

In power grid surveying, it is often necessary to fuse panchromatic and multispectral imagery for the design of power lines. Despite the abundance of deep learning networks for fusing these images, the results often suffer from spectral information loss or structural blurring. This study introduces a fusion model specifically tailored for power grid surveying that significantly enhances the representation of spatial-spectral features in remote sensing images. The model comprises three main modules: a TransforRS-Mamba module that integrates the sequence processing capabilities of the Mamba model with the attention mechanism of the Transformer to effectively merge spatial and spectral features; an improved spatial proximity-aware attention mechanism (SPPAM) that utilizes a spatial constraint matrix to greatly enhance the recognition of complex object relationships; and an optimized spatial proximity-constrained gated fusion module (SPCGF) that integrates spatial proximity constraints with residual connections to boost the recognition accuracy of key object features. To validate the effectiveness of the proposed method, extensive comparative and ablation experiments were conducted on GF-2 satellite images and the QuickBird (QB) dataset. Both qualitative and quantitative analyses indicate that our method outperforms 11 existing methods in terms of fusion effectiveness, particularly in reducing spectral distortion and spatial detail loss. However, the model's generalization performance across different data sources and environmental conditions has yet to be evaluated. Future research will explore the integration of various satellite datasets and assess the model's performance in diverse environmental contexts.

Synthesizing Landsat images using time series model-fitting methods for China’s coastal areas against sparse and irregular observations

ABSTRACT Long historical records and free accessibility have made Landsat data valuable for time-series analysis. However, Landsat time-series analysis is restricted for coastal areas due to the lack of sufficient numbers of clear images. The generation of synthetic Landsat images using model-fitting methods is accepted as an effective means of overcoming this problem. But for coastal areas, available observations are typically sparse and irregular, leading to ill-fitted models that distort synthetic Landsat images. In this study, we propose a linear harmonic model with different orders and implement it on an annual basis to generate synthetic Landsat images based on the Google Earth Engine (GEE) platform. First, we incorporated the available observations from adjacent years and filtered them by season to address the problems of data sparsity and irregularity. Then, we combined the threshold segmentation and ridge regression approaches to address the fake cloud problem, which sometimes contaminates synthetic Landsat images in summer. All cloud-free real Landsat images from five typical sites along China’s coasts during 1986–2020 were used to evaluate the accuracy and robustness of the synthetic Landsat images generated using our proposed methods. The principal results are as follows: (1) the R2 value of the linear harmonic model averaged across different Landsat bands and land cover types was 0.640, and the model was especially successful in simulating the surface reflectance in near NIR bands in forest and grassland areas; (2) 81.1% of the synthetic Landsat images covered by fake clouds were effectively restored, and there was the particular need to remove fake clouds for the synthetic Landsat images at low latitudes; (3) The mean absolute error for our synthetic Landsat images was 0.015, with the rate of 0.124; this was achieved under the clear-sky probability of only 36% and the average number of annual observations below 9, indicating a good performance. Compared with the continuous change detection and classification (CCDC) and seasonal median composite (SMC) methods, our proposed method offers advantages in both the accuracy and integrity of synthetic Landsat images. Our proposed method for synthesizing images also has potential for application to global coastal areas and other satellite datasets.