Fig 4 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Validation image of an explosive volcanic eruption at Arenal volcano. A) Raw image. B) Manually labelled plume mask. C) Predicted plume mask (yellow = class 1, purple = class 0). D) Difference image -false positives are displayed in yellow, false negatives are displayed in purple, correct classifications are black. There are very few false negatives in this image. False positives are confined to around the edges of the plume, therefore are not too problematic and could also be a result of poor manual labelling as well as model prediction errors. This image is relatively representative of all validation images, errors are primarily confined to the edges of the plumes. Model accuracy here is 0.9970. Image source: https://commons.wikimedia.org/wiki/File: Arenal_strombolian_eruption_2008.JPG. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
Source publication
Tracking explosive volcanic phenomena can provide important information for hazard monitoring and volcano research. Perhaps the simplest forms of monitoring instruments are visible-wavelength cameras, which are routinely deployed on volcanoes around the globe. Here, we present the development of deep learning models, based on convolutional neural n...
Contexts in source publication
Context 1
... predicted mask of a representative validation image is displayed in Fig. 4; as with all subsequent discussion, this relates to the UNetDenseNet121 model. In this case the model replicates the manual labelling very well, with an accuracy of 99.7%. Labelling differences are solely confined to small areas around the edges of the plume -subjective boundaries which could also be mislabelled by manual labelling in ...
Context 2
... architecture, where memory limitations meant that a size of 224 × 224 was used, and PSPNet, where image dimensions must be divisible by 48, therefore 240 × 240 was used. Note that no step to preserve aspect ratio during resizing was taken; however, images with notably large ratios were first cropped to a ratio close to 1:1. We highlight that, as Fig. 4 shows, these slight distortions do not seem to significantly impact the performance of the model. All images were then normalised to a range between 0 and 1, as is common practice in deep learning tasks. Due to loading convention in the OpenCV library (Bradski, 2000), there is some confusion over images being passed to models in ...
Citations
... Our problem requires separating the largest turbulent motions arising from individual column pulses from the complex and moving background of the column exterior. The advent of advanced video segmentation (feature identification and classification) algorithms including Recurrent Convolutional Neural Networks (R-CNN's) and Long Short-Term Memory Networks (LSTM-CNN's) provides a promising way forward for rapid and automated quantitative analyses of video and thermal imagery (e.g., Wilkes et al., 2022;Witsil & Johnson, 2020). However, such supervised machine learning techniques require extensive training with well-curated data sets from field and laboratory studies or simulations spanning the full range of spatio-temporal dynamics involved in the evolutions shown in Figure 1 and that we characterize in detail below. ...
... Moreover, the same principles for capturing time-dependent eruption dynamics apply for other monitoring techniques for which relationships between measured source properties and column dynamic states can be established, such as Doppler radar (e.g., Bonadonna et al., 2011;Donnadieu, 2012;Freret-Lorgeril et al., 2020), video or ultraviolet imagery (e.g., Woitischek, Mingotti, et al., 2021), or acoustic monitoring (e.g., De Angelis et al., Watson et al., 2021). We underscore the conclusions of other recent studies and emphasize the value of multi-instrument, community data sets to create rapid-analysis AI tools for real time monitoring of volcanic columns (Cigna et al., 2020;Dye & Morra, 2020;Guerrero Tello et al., 2022;Korolev et al., 2021;Wilkes et al., 2022;Witsil & Johnson, 2020). ...
Volcanic eruption columns typically have unsteady source conditions, where mass and heat fluxes from the vent evolve or fluctuate on time scales from seconds to hours. However, integral plume models routinely assume source conditions that are statistically stationary, and the degree to which source unsteadiness influences the mechanics of column rise and air entrainment has not been established with quantitative predictions. We address this knowledge gap by examining eruptions with varying unsteady character at Sabancaya Volcano, Peru. Using a novel tracking algorithm based on spectral clustering, we track the spatiotemporal evolution of coherent turbulent structures in columns using ground‐based, thermal infrared imagery. For turbulent structures tracked in time and space, we calculate the power law decay exponent of excess temperature with height. In general, the starting pulses of transient events are characterized by power law exponents matching theoretical predictions for an instantaneous point release of buoyancy (i.e., a thermal), which evolve with sustained emissions to values consistent with steady plumes. Our results support previous findings from field evidence and laboratory experiments that entrainment and gravitational stability in unsteady volcanic columns are inadequately captured by time‐averaging or constant entrainment coefficients. We propose a quantitative definition for column source unsteadiness which captures the timing and magnitude of source fluctuations on time scales that influence entrainment mechanics, and which provisionally predicts our observed differences in power law behavior. We argue for systematic experimental and numerical studies of the relationship between source unsteadiness and entrainment to implement unsteady entrainment parameterizations for integral plume models.
... Other authors were also able to classify rock images in order to improve petrographic analysis time through ternary diagrams [22,23]. CNN has been used to classify explosive volcanic plumes [24], fossil identification [25], The petroleum density is vital to the oil and gas industry because it implies reservoir recovery's cost reduction together with refined products' quality, which can reduce production costs for companies [5,8]. The • API gravity decreases with the light compounds' loss as well as petroleum quality [3,6,9]. ...
... Other authors were also able to classify rock images in order to improve petrographic analysis time through ternary diagrams [22,23]. CNN has been used to classify explosive volcanic plumes [24], fossil identification [25], and unstructured geological text data clustering [26]. Koeshidayatullah et al. [27] used transfer learning [28] to classify 4000 carbonate petrographic images in six classes as well as nine object detection classes. ...
Chromatographic oil analysis is an important step for the identification of biodegraded petroleum via peak visualization and interpretation of phenomena that explain the oil geochemistry. However, analyses of chromatogram components by geochemists are comparative, visual, and consequently slow. This article aims to improve the chromatogram analysis process performed during geochemical interpretation by proposing the use of Convolutional Neural Networks (CNN), which are deep learning techniques widely used by big tech companies. Two hundred and twenty-one chromatographic oil images from different worldwide basins (Brazil, the USA, Portugal, Angola, and Venezuela) were used. The open-source software Orange Data Mining was used to process images by CNN. The CNN algorithm extracts, pixel by pixel, recurring features from the images through convolutional operations. Subsequently, the recurring features are grouped into common feature groups. The training result obtained an accuracy (CA) of 96.7% and an area under the ROC (Receiver Operating Characteristic) curve (AUC) of 99.7%. In turn, the test result obtained a 97.6% CA and a 99.7% AUC. This work suggests that the processing of petroleum chromatographic images through CNN can become a new tool for the study of petroleum geochemistry since the chromatograms can be loaded, read, grouped, and classified more efficiently and quickly than the evaluations applied in classical methods.
... The utility of camera-based systems in detecting ultraviolet (UV) radiation has been widely recognized and scrutinized within the research community, as summarized in Table 2 [2,20,22,30,[58][59][60][61][62][63][64]. This growing body of evidence attests to the capacity of image sensors to yield meaningful UV irradiance data, suggesting that prevalent devices such as smartphones and digital cameras could be deployed as pragmatic, accessible tools for UV radiation research. ...
... This growing body of evidence attests to the capacity of image sensors to yield meaningful UV irradiance data, suggesting that prevalent devices such as smartphones and digital cameras could be deployed as pragmatic, accessible tools for UV radiation research. Using low-cost UV cameras to measure how much sulphur dioxide comes out of volcanoes with UV light [64] 320 and 330 nm ...
... With the aid of deep learning algorithms, every pixel in an image can be categorized and labeled, facilitating the identification of pixel clusters belonging to different classes [117]. Semantic segmentation endeavors to attribute a class label to every pixel in an image [64]. The final category, instance segmentation, is a specialized form of image segmentation focusing on the detection of instances of objects and delineating their boundaries. ...
The growing demand for sustainable and energy-efficient buildings has highlighted the need for reliable and accurate methods to detect fenestration deterioration and assess UV radiation transmission. Traditional detection techniques, such as spectrophotometers and radiometers, discussed in Part I, are often expensive and invasive, necessitating more accessible and cost-effective solutions. This study, which is Part II, provides an in-depth exploration of the concepts and methodologies underlying UV bandpass-filtered imaging, advanced image processing techniques, and the mechanisms of pixel transformation equations. The aim is to lay the groundwork for a unified approach to detecting ultraviolet (UV) radiation transmission in fenestration glazing. By exploiting the capabilities of digital imaging devices, including widely accessible smartphones, and integrating them with robust segmentation techniques and mathematical transformations, this research paves the way for an innovative and potentially democratized approach to UV detection in fenestration glazing. However, further research is required to optimize and tailor the detection methods and approaches using digital imaging, UV photography, image processing, and computer vision for specific applications in the fenestration industry and detecting UV transmission. The complex interplay of various physical phenomena related to UV radiation, digital imaging, and the unique characteristics of fenestration glazing necessitates the development of a cohesive framework that synergizes these techniques while addressing these intricacies. While extensively reviewing existing techniques, this paper highlights these challenges and sets the direction for future research in the UV imaging domain.
... Nonetheless, the sampling rate and resolution of the captured images are of a different order of magnitude than the object of study of this work. Other authors use images captured by satellite to study tracking and segment deformable objects such as icebergs (YoungHyunKoo, 2021; Barbat et al., 2019), other natural phenomena such as volcanic eruptions to monitor ash plumes (Wilkes et al., 2022;Guerrero Tello et al., 2022) or cloud segmentation for cloudiness determination (Xie et al., 2020). ...
Wildfires have significant impacts on the environment, society, and economy. Consequently, understanding its dynamics is crucial to evaluate such effects. Nonetheless, monitoring and measuring the burned area by traditional, non-automatic methods remains time-consuming and challenging.
For several years, automatic semantic segmentation models have been used to describe natural phenomena, but deep learning models have recently achieved very competitive results. However, this new breed of models typically needs annotated datasets of significant dimensions. Nonetheless, datasets for real-time burnt area segmentation are often scarce.
In this article, we create tools to support the benchmarking for testing and validating burned area segmentation models in a wildfire context. As such, we propose a new manually annotated dataset for segmentation of forest fire burned area based on a video captured by a UAV to train and evaluate semantic segmentation models. We suggest specific temporal consistency metrics to validate burned area polygons generated by the models in successive frames of non-annotated data. We also explore deep learning-based techniques and establish baselines, including IoU values superior to 95% on the test set.
... This region-growing technique scans for contiguous pixels according to a criterion of similarity (e.g., pixels below or above a threshold as in Teodoro et al., 2008;Teodoro and Goncalves, 2011) departing from a location of plume origin (also referred to as seed). In addition to the approaches mentioned, a new generation of artificial intelligence (AI) algorithms, including machine learning and deep learning, has been applied for the detection of various types of plumes such as volcanic ash (Guerrero Tello et al., 2022;Wilkes et al., 2022), fire smoke (Khan et al., 2021), atmospheric gases (Finch et al., 2022) or dust (Berndt et al., 2021) plumes. However, to the best of our knowledge, no studies have yet addressed the challenge of identifying turbid coastal plumes applying AI approaches. ...
Turbid coastal plumes carry sediments, nutrients, and pollutants. Satellite remote sensing is an effective tool for studying water quality parameters in these turbid plumes while covering a wide range of hydrological and meteorological conditions. However, determining boundaries of turbid coastal plumes poses a challenge. Traditionally, thresholds are the approach of choice for plume detection as they are simple to implement and offer fast processing (especially important for large datasets). However, thresholds are site-specific and need to be re-adjusted for different datasets or when meteorological and hydrodynamical conditions differ. This study compares state-of-the-art threshold approaches with a novel algorithm (PLUMES) for detecting turbid coastal plumes from satellite remote sensing, tested for Patos Lagoon, Brazil. PLUMES is a semi-supervised, and spatially explicit algorithm, and does not assume a unique plume boundary. Results show that the thresholds and PLUMES approach each provide advantages and limitations. Compared with thresholds, the PLUMES algorithm can differentiate both low or high turbidity plumes from the ambient background waters and limits detection of coastal resuspension while automatically retrieving metrics of detected plumes (e.g., area, mean intensity, core location). The study highlights the potential of the PLUMES algorithm for detecting turbid coastal plumes from satellite remote sensing products, which can have significantly positive implications for coastal management. However, PLUMES, despite its demonstrated effectiveness in this study, has not yet been applied to other study sites.
Volcanic plumes typically have unsteady source conditions, where mass and heat fluxes from the vent evolve or fluctuate on time scales from seconds to hours. However, integral plume models routinely assume source conditions that are statistically stationary, and the degree to which source unsteadiness influences the mechanics of plume rise and air entrainment has not been established with quantitative predictions. We address this knowledge gap by examining eruptions with varying unsteady character at Sabancaya Volcano, Peru. We develop a novel tracking algorithm based on spectral clustering, which tracks the spatiotemporal evolution of coherent turbulent structures in plumes using ground-based, thermal infrared imagery. For turbulent structures tracked in time and space, we calculate the power law decay exponent of excess temperature with height. In general, more unsteady or transient events are characterized by power law exponents matching theoretical predictions for an instantaneous point release of buoyancy (i.e. a thermal), which evolve with sustained emissions to values consistent with steady plumes. Our results support previous findings from field evidence and laboratory experiments that entrainment and gravitational stability in unsteady volcanic plumes are inadequately captured by time-averaging or constant entrainment coefficients. We propose a quantitative definition for plume source unsteadiness which captures the timing and magnitude of source fluctuations on time scales that influence entrainment mechanics, and which provisionally predicts our observed differences in power law behavior. We argue for systematic experimental and numerical studies of the relationship between source unsteadiness and entrainment to develop unsteady entrainment parameterizations for integral plume models.
Volcanic eruptions are routinely simulated as sustained flows of gas and ash, similar to the mechanics of a jet engine.
However, most eruptions in nature are unsteady at the source vent, meaning the flow rate and heat content of erupted material varies substantially over time scales ranging from seconds to hours.
This variation impacts mixing of eruption plumes with the background atmosphere (a process called entrainment), ultimately affecting how high plumes rise and where they disperse hazardous ash.
To better understand how unsteady conditions influence eruption behavior and hazard, we analysed infrared camera imagery of eruption plumes at Sabancaya Volcano, Peru.
By developing a new algorithm which tracks individual turbulent eddies in the rising plume, we measure how the heat content in the plumes evolve with entrainment of atmosphere.
Our measurements show the plume mixing process evolving between theoretical predictions for sustained, jet-like flows and single, brief pulses, as a result of unsteady, evolving conditions at the plume source.
We use our measurements to propose a mathematical framework for quantifying unsteadiness in volcanic plumes, enabling future experiments and computer simulations that include unsteady effects.
Ultimately, this will lead to improved forecasts of ash dispersal and resulting hazards for unsteady eruptions.
