Hyperspectral Satellite Imagery Research Articles

A Small-Sample Classification Strategy for Extracting Fractional Cover of Native Grass Species and Noxious Weeds in the Alpine Grasslands.

The fractional cover of native grass species (NGS) and noxious weeds (NW) provides a more comprehensive understanding of grassland health in the alpine grasslands. However, coverage extraction of NGS and NW from satellite hyperspectral imagery can be challenging due to the small spectral and spatial feature difference, insufficient training samples, and the lack of effective fractional cover extraction methods. In this research, firstly, a feature optimization method is proposed to optimize the difference feature between NGS and NW. Secondly, a spectral-spatial constrained re-clustering training sample extension method (SSCTSE) is proposed to increase the number of training samples. Thirdly, a composite three-kernel SVM method (CTK-SVM) is developed to produce fractional cover maps of NGS and NW. The experimental results show that (1) the feature optimization method is effective in preserving the spectral and spatial difference features while eliminating invalid features; (2) the SSCTSE algorithm is capable of significantly increasing the number of training samples; (3) the fractional cover maps of NGS and NW are produced with the CTK-SVM method with overall accuracies of approximately 65%, and the RMSEs of NGS and NW are approximately 16% and 11%, respectively. The results provide a foundation for the fractional cover extraction of different grass species in alpine grasslands based on satellite hyperspectral imagery.

A multi-source approach to mapping habitat diversity: Comparison and combination of single-date hyperspectral and multi-date multispectral satellite imagery in a Mediterranean Natural Reserve

The increasing availability of spaceborne hyperspectral satellite imagery opens new opportunities for forest habitat mapping and monitoring, but the limitation of its generally low temporal resolution must be considered. In this study, we compare the ability of single-date PRISMA (PRecursore IperSpettrale della Missione Applicativa), the hyperspectral satellite from the Italian Space Agency, with that of both single-date and multi-date Sentinel-2 (S2) and PlanetScope (PS) to detect and correctly classify various EUNIS habitat types distributed over a relatively small spatial extent (6000 ha) in a natural reserve in Central Italy. The case study deals with multiple levels of spectral similarity, as the dominant canopy species of the target forest habitat classes belong to the same genus (Quercus spp., both deciduous and evergreen species) as well as of different taxa (Pinus and Fraxinus spp.). We performed a pixel-based classification with the Random Forest algorithm using a set of 28 spectral indices computed on PRISMA bands, 22 on S2, and 12 on PS. A Canopy Height Model (CHM) was also used as an input variable for the classification. Our results showed that PRISMA considerably outperforms the two multispectral satellites in single-date classifications, with an overall accuracy of 84 % compared to PlanetScope's 69 % and Sentinel-2's 72 %. Regarding the comparison between multi-date multispectral and single-date hyperspectral, 10-fold cross-validation results revealed that S2 achieves an out-of-bag error rate of approximately 16 %, while PRISMA achieves 17 % and PS 19 %. This demonstrates that a combination of spectral indices calculated during the growing season can capture phenological or physiological differences among the target species, which consequently results in a significant improvement in the classification accuracy of the multispectral sensors. Ultimately, classification results from all three sensors were combined to create probability maps for each forest class, identifying areas classified with a higher degree of certainty by each satellite tested and potentially contributing to forest management by defining areas with varying conservation levels.

Estimation of soil organic matter content by combining Zhuhai-1 hyperspectral and Sentinel-2A multispectral images

Hyperspectral satellite imagery has significant advantages in rapidly monitoring soil organic matter (SOM) content over a large area. However, limitations in the timely acquisition of site-specific data may affect its effectiveness due to weather influences and revisit cycles. This paper proposes a novel method for estimating SOM content that combines the high temporal resolution of the Zhuhai-1 hyperspectral satellite image and Sentinel-2A multispectral satellite image to broaden the spectral range of Zhuhai-1 images. Multisource features, including spectral bands, topographic features, textural features, and spectral indexes, are extracted from the combined image and digital elevation model. An improved genetic algorithm (IGA) is proposed to optimize feature selection and extreme gradient boosting (XGBoost) is then applied to estimate SOM content. The proposed method was validated using 197 topsoil samples and satellite images collected from a demonstration area in Lishu County, Jilin Province, China. The results indicate that the estimation accuracy using the combined image was greater than using one single image. Compared with only using the Sentinel-2A and Zhuhai-1 images, the coefficient of determination (R2) values improved from 0.61 and 0.73 to 0.82, the ratio of the prediction to the deviation (RPD) values improved from 1.62 and 1.93 to 2.35, and the ratio of performance to the interquartile distance (RPIQ) values improved from 2.16 and 2.58 to 3.15, respectively. Furthermore, the XGBoost algorithm outperformed the random forest algorithm in terms of model accuracy and mapping reliability. The use of multisource features improved the R2 value from 0.45 to 0.82 compared to only using 31 spectral bands of Zhuhai-1 and Sentinel-2A image, with contribution rates in descending order of spectral indexes (48.2%), topographic features (24.7%), spectral bands (19.9%), and textural features (7.2%). This paper thus presents a promising method for efficient periodic mapping of the SOM content by combining data from a hyperspectral satellite constellation and a multispectral satellite image.

A hybrid algorithm for estimating total nitrogen from a large eutrophic plateau lake using Orbita hyperspectral (OHS) satellite imagery

Total nitrogen concentration (CTN) enrichment is the primary cause of natural water eutrophication. Accurately estimating CTN and its spatiotemporal dynamics is crucial for formulating monitoring and control measures to alleviate lake eutrophication. A hybrid model was proposed for estimating CTN in optically complex inland waters by incorporating the relationship between CTN and water optical active components for Zhuhai-1 Orbita hyperspectral (OHS) imagery. Compared with other semi-analytical algorithms, the re-adjusted reference wavelength QAA716 shows the best performance in aph(λ) retrieval. The hybrid model for CTN estimation achieves an root mean square deviation (RMSD) of 0.20 mg/L, a mean absolute percentage deviation (MAPD) of 6.96 %, and a unbiased mean absolute percentage deviation (UMAPD) of 6.96 %, with aph(569) accuracy exerting the greatest influence. Ground-satellite synchronous validation demonstrates robust performance, with an RMSD of 0.28 mg/L, a MAPD of 14.49 %, and a UMAPD of 14.92 %. The hybrid model was applied to OHS observations of Lake Dianchi from April 2019 to September 2021. The analysis revealed a generally decreasing trend in CTN during this timeframe. The above results demonstrate that the robustness and applicability of the proposed CTN hybrid model for inland waters with complex optical properties. Furthermore, satellite-based data products provide valuable information for formulating lake management strategies.

A combined Remote Sensing and GIS-based method for Local Climate Zone mapping using PRISMA and Sentinel-2 imagery

In the last decade, several methods have been developed for Local Climate Zone (LCZ) mapping, encompassing Remote Sensing and Geographic Information Systems (GIS) −based procedures. Combined approaches have also been proposed to compensate for intrinsic limitations that characterized their separate application. Recent work has disclosed the potential of hyperspectral satellite imagery for improving LCZ identification. However, the use of hyperspectral data for LCZ mapping is yet to be fully unfolded. A combined Remote Sensing and GIS-based method for LCZ mapping is proposed to exploit the integration of hyperspectral PRISMA and multispectral Sentinel-2 images with ancillary urban canopy parameter layers. Random Forest algorithm is applied to the feature sets to obtain the LCZ classification. The method is tested on the Metropolitan City of Milan (Italy), for the period from February to August 2023. A spectral separability analysis is carried out to investigate the improvement in LCZ identification using PRISMA in comparison to Sentinel-2 data, as well as improvements in LCZ spectral separability on PRISMA pan-sharpened images. The resulting maps’ quality is evaluated by extracting accuracy metrics and performing inter-comparisons with maps computed from the LCZ Generator benchmark tool. Inter-comparisons yield promising results with a mean Overall Accuracy increase of 16% using PRISMA for each LCZ class. Furthermore, we find that PRISMA improves the detection of LCZs compared to Sentinel-2, with a mean Overall Accuracy increase of 5%, in line with the higher spectral separability of PRISMA spectral signatures computed on the training samples.

PRISMA Hyperspectral Satellite Imagery Application to Local Climate Zones Mapping

Abstract. The urban heat island effect exacerbates the vulnerability of cities to climate change, emphasizing the need for sustainable urban planning driven by data evidence. In the last decade, the Local Climate Zone (LCZ) model emerged as a key tool for categorizing urban landscapes, aiding in the development of urban temperature mitigation strategies. In this work, the contribution of hyperspectral satellite imagery to LCZ mapping, leveraging the Italian Space Agency (ASI)’s PRISMA satellite, is investigated. Mapping performances are compared with traditional multispectral-based LCZ mapping using Sentinel-2 satellite imagery. The Random Forest algorithm is utilized for LCZ classification, with evaluation conducted through spectral separability analysis and accuracy assessment between PRISMA and Sentinel-2 derived LCZ maps as well as with the benchmark LCZ Generator mapping tool. An initial experiment on the effect of PRISMA image pan-sharpening on LCZ spectral separability is also presented. Results obtained for Milan (Northern Italy) demonstrate the potential of hyperspectral imagery in enhancing LCZ identification compared to multispectral data, with promising improvements in LCZ maps overall accuracy. Finally, air temperature patterns within each LCZ class are explored, qualitatively confirming the influence of urban morphology on thermal comfort.

The application of PRISMA hyperspectral satellite imagery in the delineation of distinct hydrothermal alteration zones in the Chilean Andes: The Marimaca IOCG and the Río Blanco-Los Bronces Cu-Mo porphyry districts

This study provides an overview of the application of hyperspectral imagery acquired by the Italian satellite mission “PRecursore IperSpettrale della Missione Applicativa” (PRISMA), for mapping hydrothermal and supergene mineral alteration zones associated with different Cu deposits in the Chilean Andes. Study areas include the Marimaca Copper Project in the Naguayán district (Antofagasta Province), hosting Late Jurassic – Early Cretaceous iron oxide-copper–gold mineralizations and the Río Blanco-Los Bronces district (Santiago Region) known for its Late Miocene – Early Pliocene copper-molybdenum porphyry deposits. These mineral systems exhibit hydrothermal alteration haloes extending over several kilometers, with upward and outward mineral zonation.To characterize the surface-exposed alteration, the satellite imagery was processed using a multiple feature extraction workflow targeting the relative abundances and compositions of specific supergene and hydrothermal alteration minerals, including Fe-oxides and hydroxides (hematite-goethite), phyllosilicates (micas-kaolinite-chlorite), hydroxyl-bearing sulfates (alunite-natroalunite-jarosite) and epidote. The results reveal a zonation from proximal sulfates (natroalunite to alunite) and Al-rich white mica (ranging from 0.3 km-wide for the area identified to the east of the Marimaca deposit to 1 km-wide in correspondence of the Los Sulfatos deposit, respectively) to distal Al-poor white mica (0.2 to 0.5 km-wide, respectively) and an outer chlorite-epidote zone. In the Marimaca Project area, the structurally-controlled alteration evolves west to east from chlorite-rich to phengitic white mica (Al-poor). A NNE-SSW trending supergene leached cap covers a 6 km-long and up to 2 km-wide area from north to south. Our study documents how spaceborne hyperspectral imaging spectroscopy can support mineral exploration by enabling non-invasive reconnaissance mapping of the outcropping rocks, providing specific targeting information for follow-up field surveys and drilling campaigns.

Ontology-Based Data Observatory for Formal Knowledge Representation of UXO Using Advanced Semantic Web Technologies

As landmines and other unexploded ordnances (UXOs) present a great risk to civilians and infrastructure, humanitarian demining is an essential component of any post-conflict reconstruction. This paper introduces the Minefield Observatory, a novel web-based datastore service that semantically integrates diverse data in humanitarian demining to comprehensively and formally describe suspected minefields. Because of the high heterogeneity and isolation of the available minefield datasets, extracting relevant information to determine the optimal course of demining efforts is time-consuming, labor-intensive and requires highly specialized knowledge. Data consolidation and artificial intelligence techniques are used to convert unstructured data sources and store them in an ontology-based knowledge database that can be efficiently accessed through a Semantic Web application serving as the Minefield Observatory user interface. The MINEONT+ ontology was developed to integrate diverse mine scene information obtained through non-technical surveys and remote sensing, such as aerial and hyperspectral satellite imagery, indicators of mine presence and absence, contextual data, terrain analysis information, and battlefield reports. The Minefield Observatory uses the Microdata API to embed this dataset into dynamic HTML5 content, allowing seamless usage in a user-centric web tool. A use-case example was provided demonstrating the viability of the proposed approach.

Comparing Laboratory and Satellite Hyperspectral Predictions of Soil Organic Carbon in Farmland

Mapping soil organic carbon (SOC) accurately is essential for sustainable soil resource management. Hyperspectral data, a vital tool for SOC mapping, is obtained through both laboratory and satellite-based sources. While laboratory data is limited to sample point monitoring, satellite hyperspectral imagery covers entire regions, albeit susceptible to external environmental interference. This study, conducted in the Yuncheng Basin of the Yellow River Basin, compared the predictive accuracy of laboratory hyperspectral data (ASD FieldSpec4) and GF-5 satellite hyperspectral imagery for SOC mapping. Leveraging fractional order derivatives (FODs), various denoising methods, feature band selection, and the Random Forest model, the research revealed that laboratory hyperspectral data outperform satellite data in predicting SOC. FOD processing enhanced spectral information, and discrete wavelet transform (DWT) proved effective for GF-5 satellite imagery denoising. Stability competitive adaptive re-weighted sampling (sCARS) emerged as the optimal feature band selection algorithm. The 0.6FOD-sCARS RF model was identified as the optimal laboratory hyperspectral prediction model for SOC, while the 0.8FOD-DWT-sCARS RF model was deemed optimal for satellite hyperspectral prediction. This research, offering insights into farmland soil quality monitoring and strategies for sustainable soil use, holds significance for enhancing agricultural production efficiency.

Machine and Deep Learning Regression of Chlorophyll-a Concentrations in Lakes Using PRISMA Satellite Hyperspectral Imagery

The estimation of Chlorophyll-a concentration is crucial for monitoring freshwater ecosystem health, particularly in lakes, as it is closely linked to eutrophication processes. Satellite imagery enables synoptic and frequent evaluations of Chlorophyll-a in water bodies, providing essential insights into spatiotemporal eutrophication dynamics. Frontier applications in water remote sensing support the utilization of machine and deep learning models applied to hyperspectral satellite imagery. This paper presents a comparative analysis of conventional machine and deep learning models—namely, Random Forest Regressor, Support Vector Regressor, Long Short-Term Memory, and Gated Recurrent Unit networks—for estimating Chlorophyll-a concentrations. The analysis is based on data from the PRecursore IperSpettrale della Missione Applicativa (PRISMA) hyperspectral mission, complemented by low-resolution Chlorophyll-a concentration maps. The analysis focuses on three sub-alpine lakes, spanning Northern Italy and Switzerland as testing areas. Through a series of modelling experiments, best-performing model configurations are pinpointed for both Chlorophyll-a concentration estimations and the improvement of spatial resolution in predictions. Support Vector Regressor demonstrated a superior performance in Chlorophyll-a concentration estimations, while Random Forest Regressor emerged as the most effective solution for refining the spatial resolution of predictions.

Mapping Local Climate Zones in Lausanne (Switzerland) with Sentinel-2 and PRISMA imagery: comparison of classification performance using different band combinations and building height data

ABSTRACT The Local Climate Zone (LCZ) concept has emerged as a valuable classification system for climate-related studies. The World Urban Database and Access Portal Tools (WUDAPT) protocol provides a framework for generating a LCZ segmentation which relies on the supervised classification of multispectral imagery. However, since LCZ is based on the physical and thermal properties of the urban surfaces, more insightful information on the surface reflectivity characteristics – which is provided by hyperspectral sensors – may be beneficial for improving the LCZ classification. This assumption is investigated in this study by comparing the classification performance of a supervised algorithm applied to multispectral (Sentinel-2) and hyperspectral (PRISMA) satellite imagery. The study area is the city of Lausanne (Switzerland). Experiments are performed considering these sensors and different band combinations, including the building height layer as an additional band. Preliminary outcomes show that PRISMA imagery yields satisfying results in terms of classification accuracy while not outperforming Sentinel-2. An improvement is achieved by leveraging the first 10 PRISMA Principal Components which allows to retain the uncorrelated information out of the original bands. These first results will be validated in future investigations by improving image pre-processing and exploiting a larger number of seasonal PRISMA acquisitions.

Marine plastic detection using PRISMA hyperspectral satellite imagery in a controlled environment

ABSTRACT The amount of plastics on seawater is causing ecosystem damage to both land areas and water bodies. It is proven that once plastic particles reach the sea, they will be degraded. As their detection is not easy with in situ sampling, remote sensing techniques could help detect and evaluate their impact on ecosystems. To understand the main limitations of detecting floating plastic material through remote images, an a pilot experiment was carried out on an artificial floating plastic target deployed at the Aegean Sea (Greece) within the Plastic Litter Project 2021. Hyperspectral PRISMA, multispectral PlanetScope and visible Unmanned Aerial Systems (UAS) acquisitions were analysed. A nonlinear unmixing technique was applied to derive the spectral signature of the plastic target; finally, the linear unmixing approach allowed for determining the plastic percentage occupation at the pixel level. Different band combinations of the PRISMA data were selected to evaluate which provided the best result for the detection; one of these band combinations was retrieved via the principal component analysis. Only slight differences were achieved using the different PRISMA band-sets. As expected, since the artificial plastic target had a diameter comparable with the PRISMA spatial resolution (~30 m), its detection was a challenging task caused by the water influence in the pixel (mixed pixels). The detection was realized by benefitting from the high amount of available spectral bands, as confirmed by the comparative test with a PlanetScope image used at its original spatial resolution (3 m) and after degrading it at the PRISMA level. Results demonstrated that an optimal target detection was possible with few spectral bands, only taking advantage of the high spatial resolution (compared with the target dimension). Indeed, unreliable plastic fractions were derived at the PRISMA spatial resolution with a limited number of spectral bands.

A Review of Ocean Color Algorithms to Detect Trichodesmium Oceanic Blooms and Quantify Chlorophyll Concentration in Shallow Coral Lagoons of South Pacific Archipelagos

The oceanic waters of the Southwest Tropical Pacific occupy a vast region including multiple Pacific Island Countries. The state of these waters is determinant for fisheries and the blue economy. Ocean color remote sensing is the main tool to survey the variability and long-term evolution of these large areas that are important for economic development but are affected by climate change. Unlike vast oligotrophic gyres, tropical waters are characterized by numerous archipelagos and islands, with deep and shallow lagoons subjected to the large impacts of the land. Strikingly large dendritic phytoplankton (Trichodesmium) blooms with high levels of chlorophyll, developing within archipelagos, as well as coastal enrichments from various origins may be observed. Algorithms to detect the presence of Trichodesmium have been developed or adapted, as well as algorithms to estimate the chlorophyll concentration ([Chl-a)]. Adapting existing [Chl-a] algorithms does not always yield high, i.e., sufficient, accuracy. A review of published regional bio-optical algorithms developed taking into account the specific phytoplankton composition and minimizing the adverse impacts of particles and the seabed bottom on [Chl-a] determination is presented, as well the bio-optical database that allowed their development. The interest of such algorithms for a variety of applications and scientific accomplishments is highlighted, with a view to further addressing the main biology and biogeochemistry questions, e.g., to determine the true impact of diazotrophs and assess lagoon [Chl-a] variability with the highest confidence. This work anticipates the use of future coarse and high-spatial-resolution and multi- and hyper-spectral satellite imagery in the Pacific.

Towards practical semi-empirical models for the estimation of leaf and canopy water contents from hyperspectral reflectance

The knowledge of crop water status provides valuable information for agricultural water management. Most remote sensing methods for estimating water contents are based on statistical models and spectral features in the shortwave infrared region where water absorption dominates. However, the effects of leaf structure (Ns) and dry matter (expressed as leaf mass per area, LMA) on the spectral features in this region are not fully considered when estimating leaf water content (LWC) and canopy water content (CWC). This issue would lead to the data-specific empirical model, which lacks generality and is not transferable among different conditions. To fill this gap, this study for the first time evaluated the Ns and LMA effects on the spectral features, including the widely used normalized difference (ND)-type vegetation index (VI) and wavelet feature (WF) from wavelet transform. The leaf and canopy radiative transfer model simulations showed that the ND-type VI was sensitive to Ns at both leaf and canopy levels, whereas WF was sensitive to LMA at the canopy level. For example, the relationships between WF and CWC varied clearly with the level of LMA, especially for high values of CWC. Next, the optimal water index (OWI = ND1200,1500) and optimal WF (OWF = WF1600,8) were determined by minimizing their correlations with the Ns and LMA based on the leaf and canopy simulations. Unlike the existing spectral features applicable to only leaf or canopy level alone, OWI and OWF showed more consistent performance on estimating LWC and CWC. OWI- and OWF-based semi-empirical models were calibrated using model simulations and then evaluated using the independent measured datasets. The semi-empirical model based on OWI exhibited moderate estimation accuracy of LWC (R2 = 0.62, RMSE = 24.33 g/m2) and CWC (R2 = 0.69, RMSE = 0.15 kg/m2), outperforming previous ND-type VIs. In comparison, the best estimation accuracies for LWC (R2 = 0.80, RMSE = 17.47 g/m2) and CWC (R2 = 0.85, RMSE = 0.11 kg/m2) were obtained by the OWF-based semi-empirical model when the prior information of LMA was incorporated at the canopy level. These findings have great potential for mapping crop water status over large regions with universal semi-empirical models and hyperspectral satellite imagery.

A Novel Adaptively Optimized PCNN Model for Hyperspectral Image Sharpening

Hyperspectral satellite imagery has developed rapidly over the last decade because of its high spectral resolution and strong material recognition capability. Nonetheless, the spatial resolution of available hyperspectral imagery is inferior, severely affecting the accuracy of ground object identification. In the paper, we propose an adaptively optimized pulse-coupled neural network (PCNN) model to sharpen the spatial resolution of the hyperspectral imagery to the scale of the multispectral imagery. Firstly, a SAM-CC strategy is designed to assign hyperspectral bands to the multispectral bands. Subsequently, an improved PCNN (IPCNN) is proposed, which considers the differences of the neighboring neurons. Furthermore, the Chameleon Swarm Optimization (CSA) optimization is adopted to generate the optimum fusion parameters for IPCNN. Hence, the injected spatial details are acquired in the irregular regions generated by the IPCNN. Extensive experiments are carried out to validate the superiority of the proposed model, which confirms that our method can realize hyperspectral imagery with high spatial resolution, yielding the best spatial details and spectral information among the state-of-the-art approaches. Several ablation studies further corroborate the efficiency of our method.

A hyperspectral inversion framework for estimating absorbing inherent optical properties and biogeochemical parameters in inland and coastal waters

The simultaneous remote estimation of biogeochemical parameters (BPs) and inherent optical properties (IOPs) from hyperspectral satellite imagery of globally distributed optically distinct inland and coastal waters is a complex, unsolved, non-unique inverse problem. To tackle this problem, we leverage a machine-learning model termed Mixture Density Networks (MDNs). MDNs outperform operational algorithms by calculating the covariance between the simultaneously estimated products. We train the MDNs on a large (N = 8237) dataset of co-aligned, in situ measured, hyperspectral remote sensing reflectance (Rrs), BPs, and absorbing IOPs from globally representative optically distinct inland and coastal waters. The estimated IOPs include absorption due to phytoplankton (aph), chromophoric dissolved organic matter (acdom), and non-algal particles (anap). The estimated BPs include chlorophyll-a, total suspended solids, and phycocyanin (PC). MDNs dramatically reduce uncertainty in the retrievals, relative to operational algorithms, when using a 50/50 dataset split, where the MDNs are trained on a randomly selected half of the in situ dataset and validated on the other half. Our model is shown to have higher, or equivalent, generalization performance than the calculated operational algorithms available for all BPs and IOPs (except PC) via a leave-one-out cross-validation assessment. The MDNs are sensitive to uncertainties in the hyperspectral satellite Rrs, resulting from instrument noise and atmospheric correction; there is a difference of ∼37.4–62.8% (using median symmetric accuracy) between the MDNs' estimates derived from co-located satellite-derived Rrs and in situ Rrs. Of the IOPs, acdom and anap are less sensitive to uncertainties in hyperspectral satellite imagery relative to aph, with remote estimates of aph exhibiting incorrect spectral shape and magnitude relative to in situ measured IOPs. Despite the uncertainties in satellite derived Rrs, the spatial distributions of BPs and IOPs in MDN-derived product maps of Lake Erie and the Curonian Lagoon, based on imagery taken with the Hyperspectral Imager for the Coastal Ocean (HICO) and PRecursore IperSpettrale della Missione Applicativa (PRISMA), are confirmed via co-aligned in situ measurements and agree with the literature's understanding of these well-studied regions. The consistency and accuracy of the model on HICO and PRISMA imagery, despite radiometric uncertainties, demonstrate its applicability to future hyperspectral missions, such as the Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) mission, where the simultaneous estimation model will serve as a key part of phytoplankton community composition analysis.

Band-Optimized Bidirectional LSTM Deep Learning Model for Bathymetry Inversion

Shallow water bathymetry is of great significance in understanding, managing, and protecting coastal ecological environments. Many studies have shown that both empirical models and deep learning models can achieve promising results from satellite imagery bathymetry inversion. However, the spectral information available today in multispectral or/and hyperspectral satellite images has not been explored thoroughly in many models. The Band-optimized Bidirectional Long Short-Term Memory (BoBiLSTM) model proposed in this paper feeds only the optimized bands and band ratios to the deep learning model, and a series of experiments were conducted in the shallow waters of Molokai Island, Hawaii, using hyperspectral satellite imagery (PRISMA) and multispectral satellite imagery (Sentinel-2) with ICESat-2 data and multibeam scan data as training data, respectively. The experimental results of the BoBiLSTM model demonstrate its robustness over other compared models. For example, using PRISMA data as the source image, the BoBiLSTM model achieves RMSE values of 0.82 m (using ICESat-2 as the training data) and 1.43 m (using multibeam as the training data), respectively, and because of using the bidirectional strategy, the inverted bathymetry reaches as far as a depth of 25 m. More importantly, the BoBiLSTM model does not overfit the data in general, which is one of its advantages over many other deep learning models. Unlike other deep learning models, which require a large amount of training data and all available bands as the inputs, the BoBiLSTM model can perform very well using equivalently less training data and a handful of bands and band ratios. With ICESat-2 data becoming commonly available and covering many shallow water regions around the world, the proposed BoBiLSTM model holds potential for bathymetry inversion for any region around the world where satellite images and ICESat-2 data are available.

Spatial-Convolution Spectral-Transformer Interactive Network for Large-Scale Fast Refined Land Cover Classification and Mapping Based on ZY1-02D Satellite Hyperspectral Imagery

Satellite hyperspectral imagery is an important data source for large-scale refined land cover classification and mapping, but the high spatial heterogeneity and spectral variability at low spatial resolution and the high computation cost for massive data remain challenges in the research community. In recent years, convolutional neural network (CNN) models with the capability for feature extraction have been widely used in hyperspectral image classification. However, incomplete feature extraction, inappropriate feature fusion, and high time consumption are still the major problems for CNN applications in large-scale fine land cover mapping. In this study, a Spatial-Convolution Spectral-Transformer Interactive Network (SCSTIN) was proposed to integrate 2D-CNN and Transformer into a dual-branch network to enhance feature extraction capabilities by exploring spatial context information and spectral sequence signatures in a targeted manner. In addition, spatial-spectral interactive fusion (SSIF) units and category-adaptive weighting (CAW) as two feature fusion modules were also adopted between and after the two feature extraction branches to improve efficiency in feature fusion. The ZY1-02D hyperspectral imagery was collected to conduct the experiments in the study area of the eastern foothills of the Helan Mountains (EFHLM), covering an area of about 8800 km2, which is the largest hyperspectral dataset as far as we know. To explore the potential of the proposed network in terms of accuracy and efficiency, SCSTIN models with different depths (SCSTIN-4 and SCSTIN-2) were performed. The results suggest that compared with the previous eight advanced hyperspectral image classifiers, both SCSTIN models achieved satisfactory performance in accuracy and efficiency aspects with low complexity, where SCSTIN-4 achieved the highest accuracy and SCSTIN-2 obtained higher efficiency. Accordingly, the SCSTIN models are reliable for large-scale fast refined land cover classification and mapping. In addition, the spatial distribution pattern of diverse ground objects in EFHLM is also analyzed.

AI-Accelerated Decision Making in Forest Management

This paper is a continuation of the research on the application of artificial intelligence in counting trees with the use of methods for the automatic analysis of photogrammetric data in forests of the continental region. This paper is devoted to the AI application in accelerating decision making processes in forest management. It also discusses how RGB imagery from drones could replace aerial and satellite hyperspectral imagery and automatically detect unhealthy and dead trees. Experimental research was conducted to verify whether Faster R-CNN can automatically detect and classify snag and trees weakened by diseases on aerial RGB data, enabling a quick response to forest-threatening factors. The research is based on photogrammetric data taken in areas of forest districts subordinate to the Regional Directorate of State Forests in Zielona Góra. Non-metric imagery data was collected from drones and small airplanes with a photogrammetric container and postprocessed with respect to the photogrammetric constraints. The results show that in specific cases aerial and satellite hyperspectral imagery can be replaced by RGB orthomosaics in order to decrease the time needed for forestry treatments.

Neural Network Emulation of Synthetic Hyperspectral Sentinel-2-like Imagery with Uncertainty.

Hyperspectral satellite imagery provides highly-resolved spectral information for large areas and can provide vital information. However, only a few imaging spectrometer missions are currently in operation. Aiming to generate synthetic satellite-based hyperspectral imagery potentially covering any region, we explored the possibility of applying statistical learning, i.e. emulation. Based on the relationship of a Sentinel-2 (S2) scene and a hyperspectral HyPlant airborne image, this work demonstrates the possibility to emulate a hyperspectral S2-like image. We tested the role of different machine learning regression algorithms (MLRA) and varied the image-extracted training dataset size. We found superior performance of Neural Network (NN) as opposed to the other algorithms when trained with large datasets (up to 100'000 samples). The developed emulator was then applied to the L2A (bottom-of-atmosphere reflectance) S2 subset, and the obtained S2-like hyperspectral reflectance scene was evaluated. The validation of emulated against reference spectra demonstrated the potential of the technique. R 2 values between 0.75-0.9 and NRMSE between 2-5% across the full 402-2356 nm range were obtained. Moreover, epistemic uncertainty is obtained using the dropout technique, revealing spatial fidelity of the emulated scene. We obtained highest SD values of 0.05 (CV of 8%) in clouds and values below 0.01 (CV of 7%) in vegetation land covers. Finally, the emulator was applied to an entire S2 tile (5490x5490 pixels) to generate a hyperspectral reflectance datacube with the texture of S2 (60Gb, at a speed of 0.14sec/10000pixels). As the emulator can convert any S2 tile into a hyperspectral image, such scenes give perspectives how future satellite imaging spectroscopy will look like.