GM2.7

Remotely piloted airborne system (RPAS) based structure-from-motion (SfM) photogrammetry is a recognised tool in geomorphological applications. However, time constraints, methodological requirements and ignorance can easily compromise photogrammetric rigour in geomorphological fieldwork. Light RPAS mounted sensors often provide inherent low geometric stability and are thus typically calibrated on-the-job in a self-calibrating bundle adjustment. Solving interior (lens geometry) and exterior (position and orientation) camera parameters requires variation of sensor-object distance, view angles and surface geometry.

Deficient camera calibration can cause systematic errors resulting in final digital elevation model (DEM) deformation. The application of multi-sensor systems, common in geomorphological research, poses additional challenges. For example, the low contrast in thermal imagery of vegetated surfaces constrains image matching algorithms.

We present a pre-calibration workflow to separate sensor calibration and data acquisition that is optimized for geomorphological field studies. The approach is time-efficient (rapid simultaneous image acquisition), repeatable (permanent object), at survey scale to maintain focal distance, and on-site to avoid shocks during transport.

The presented workflow uses a stone building as a suitable 3D calibration structure (alternatively boulder or bridge) providing structural detail in visible (DJI Phantom 4 Pro) and thermal imagery (Workswell WIRIS Pro). The dataset consists of feature coordinates extracted from terrestrial laser scanner (TLS) scans (3D reference data) and imagery (2D calibration data). We process the data in the specialized software, vision measurement system (VMS) as benchmark and the widely applied commercial SfM photogrammetric software, Agisoft MetaShape (AM) as convenient alternative. Subsequently, we transfer the camera parameters to the application in an SfM photogrammetric dataset of a river environment to assess the performance of self- and pre-calibration using different image network configurations. The resulting DEMs are validated against GNSS reference points and by DEMs of difference. 

We achieved calibration accuracies below one-third (optical) and one-quarter (thermal) of a pixel. In line with the literature, our results show that self-calibration yields the smallest errors and DEM deformations using multi-scale and oblique datasets. Pre-calibration in contrast, yielded the lowest overall errors and performed best in the single-scale nadir scenario. VMS consistently performed better than AM, possibly because AM's software “black-box” is less customisable and does not allow purely marker-based calibration. Furthermore, we present findings regarding sensor stability based on a repeat survey.

We find that pre-calibration can improve photogrammetric accuracies in surveys restricted to unfavourable designs e.g. nadir-only (water refraction, sensor mount). It can facilitate the application of thermal sensors on surfaces less suited to self-calibration. Most importantly, multi-scale survey designs could potentially become redundant, thus shortening flight time or increasing possible areal coverage.

How to cite: Senn, J. A., Mills, J., Walsh, C. L., Addy, S., and Peppa, M.-V.: Assessment of sensor pre-calibration to mitigate systematic errors in SfM photogrammetric surveys, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-279, https://doi.org/10.5194/egusphere-egu22-279, 2022.

Near-continuous time series of 3D point clouds capture local landscape dynamics at a large range of spatial and temporal scales. These data can be acquired by permanent terrestrial laser scanning (TLS) or time lapse photogrammetry, and are being used to monitor surface changes in a variety of natural scenes, including snow cover dynamics, rockfalls, soil erosion, or sand transport on beaches.

Automatic methods are required to analyze such data with thousands of point cloud epochs (acquired, e.g., hourly over several months), each representing the scene with several million 3D points. Usually, no a-priori knowledge about the timing, duration, magnitude, and spatial extent of all spatially and temporally variable change occurrences is available. Further, changes are difficult to delineate individually if they occur with spatial overlap, as for example coinciding accumulation processes. To enable fully automatic extraction of individual surface changes, we have developed the concept of 4D objects-by-change (4D-OBCs). 4D-OBCs are defined by similar change histories within the area and timespan of single surface changes. This concept makes use of the full temporal information contained in 3D time series to automatically detect the timing and duration of changes. Via spatiotemporal segmentation, individual objects are spatially delineated by considering the entire timespan of a detected change regarding a metric of time series similarity (cf. Anders et al. 2021 [1]), instead of detecting changes between pairs of epochs as with established methods.

For hourly TLS point clouds, the extraction of 4D-OBCs improved the fully automatic detection and spatial delineation of accumulation and erosion forms in beach monitoring. For a use case of snow cover monitoring, our method allowed quantifying individual change volumes more accurately by considering the timespan of changes, which occur with variable durations in the hourly 3D time series, rather than only instantaneously from one epoch to the next. The result of our time series-based method is information-rich compared to results of bitemporal change analysis, as each 4D-OBC contains the full 4D (3D + time) data of the original 3D time series with determined spatial and temporal extent.

The objective of this contribution is to present how interpretable information can be derived from resulting 4D-OBCs. This will provide new layers that are supporting subsequent geoscientific analysis of observed surface dynamics. We apply Kalman filtering (following Winiwarter et al. 2021 [2]) to model the temporal evolution of individually extracted 4D-OBCs. This allows us to extract change rates and accelerations for each point in time, and to subsequently derive further features describing the temporal properties of individual changes. We present first results of this methodological combination and newly obtained information layers which can reveal spatial and temporal patterns of change activity. For example, deriving the timing of highest change rates may be used to examine links to external environmental drivers of observed processes. Our research therefore contributes to extending the information that can be extracted about surface dynamics in natural scenes from near-continuous time series of 3D point clouds.

References:

[1] https://doi.org/10.1016/j.isprsjprs.2021.01.015

[2] https://doi.org/10.5194/esurf-2021-103

How to cite: Anders, K., Winiwarter, L., and Höfle, B.: Automatic Extraction and Characterization of Natural Surface Changes from Near-Continuous 3D Time Series using 4D Objects-By-Change and Kalman Filtering, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4225, https://doi.org/10.5194/egusphere-egu22-4225, 2022.

The use of 3D point clouds has become ubiquitous in studying geomorphology. The richness of the acquired data, together with the high availability of 3D sensing technologies, enables a fast and detailed characterisation of the terrain and the entities therein. However, the key for a comprehensive study of landforms relies on detecting geomorphological features in the data. These entities are of complex forms that do not conform to closed parametric shapes. Furthermore, they appear in varying dimensions and orientations, and they are often seamlessly embedded within the topography. The large volume of the data, uneven point distribution and occluded regions present even a greater challenge for autonomous extraction. Therefore, common approaches are still rooted in utilising standard GIS tools on rasterised scans, which are sensitive to noise and interpolation methods. Schemes that investigate morphological phenomena directly from the point cloud use heuristic and localised methods that target specific landforms and cannot be generalised. Lately, machine-learning-based approaches have been introduced for the task. However, these require large training datasets, which are often unavailable in natural environments.

This work introduces a new methodology to extract 3D geomorphological entities from unstructured point clouds. Based on the level-set model, our approach does not require training datasets or labelling, requires little prior information about existing objects, and wants minor adjustments between different types of scenes. By developing the level-set function within the point cloud realm, it requires no triangulated mesh or rasterisation. As a driving force, we utilise visual saliency to focus on pertinent regions. As the estimation is performed pointwise, the proposed model is completely point-based, driven by the geometric characteristics of the surface. The result is three-dimensional entities extracted by their original points, as they were scanned in the field. We demonstrate the flexibility of the proposed model on two fundamentally different datasets. In the first scene, we extract gullies and sinkholes in an alluvial fan and are scanned by an airborne laser scanner. The second features pockets, niches and rocks in a terrestrially scanned cave. We show that the proposed method enables the simultaneous detection of various geomorphological entities, regardless of the acquisition technique. This is facilitated without prior knowledge of the scene and with no specific landform in mind. The proposed study promotes flexibility of form and provides new ways to quantitatively describe the morphological phenomena and characterise their shape, opening new avenues for further investigation.

How to cite: Arav, R., Poeppl, F., and Pfeifer, N.: Extraction of geomorphological entities from unstructured point clouds – a three-dimensional level-set-based approach, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11081, https://doi.org/10.5194/egusphere-egu22-11081, 2022.

Solifluction is the slow downslope movement of soil mass due to freeze-thaw processes. It is widespread on hillslopes in Polar and Alpine regions and contributes substantially to sediment transport. As solifluction lobe movement is in the order of millimeters to centimeters per year, it is tricky to measure with a high spatial and temporal resolution and accuracy. We developed a semi-automated approach to monitor movement of three solifluction lobes with different degrees of vegetation cover along an elevational gradient between 2,170 and 2,567 m in Turtmann Valley, Swiss Alps. Subsequently, we compared movement rates and patterns with environmental factors.

• For solifluction movement monitoring, we applied a combination of the Phantom 4 Pro Plus and Phantom 4 RTK (Real Time Kinematic) drones, image co-alignment and COSI-CORR (Co-registration of Optically Sensed Images and Correlation) to track movement on orthophotos between 2017 and 2021. This drone data acquisition and co-alignment procedure enable a simple, time-saving field setup without Ground Control Points (GCPs).
• Our high co-registration accuracy enabled us to detect solifluction movement if it exceeds 5 mm with sparse vegetation cover. Dense vegetation cover limited feature tracking but detected movement rates and patterns still matched previous measurements using classical total station measurements at the lowest, mostly vegetated lobe.
• In contrast to traditional solifluction monitoring approaches using point measurements, our monitoring approach provides spatially continuous movement estimates across the complete extend of the lobe. Lobe movement rates were highest at the highest elevations between 2,560 and 2,567 m (up to 14.0 cm/yr for single years) and lowest at intermediate elevations between 2,417 and 2,427 m (up to 2.9 cm/yr for single years). We found intermediate movement rates at lowest elevations between 2,170 and 2,185 m (up to 4.9 cm/yr for single years). In general, movement had the highest rates at the solifluction lobes center and the lowest rates at the front of solifluction lobes.
• We linked observed movement patters to environmental factors possibly controlling solifluction movement, such as geomorphic properties, vegetation species and coverage, soil properties determined from electrical resistivity tomography (ERT), and soil temperature data. The least movement at the lobe front is characterized by coarse material and plant species stabilizing the risers or plant species growing here due to the stable risers. Most movement at the lobe center is characterized by fine material and no vegetation or plant species promoting movement. The soil temperature data further suggests that snow cover reduced freezing rates at solifluction lobes and potentially decreased solifluction movement at the lobe between 2,417 and 2,427 m.

This study is the first to demonstrate the use of drone-based images and a semi-automated method to reach high spatiotemporal resolutions to detect subtle movements of solifluction lobes at timescales of years at sub-centimeter resolution. This provides new insights into solifluction movement and into drivers of and factors controlling solifluction movement and lobe development. Therefore, our semi-automated approach may have a great potential to uncover the fundamental processes to understand solifluction movement.

How to cite: Harkema, M., Eichel, J., Nijland, W., de Jong, S., Draebing, D., and Kattenborn, T.: Using high-resolution topography to solve “periglacial puzzles”: A semi-automated approach to monitor solifluction movement, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4763, https://doi.org/10.5194/egusphere-egu22-4763, 2022.

Data on grain sizes of pebbles in gravel-bed rivers are a well-known proxy for sedimentation and transport conditions, and thus a key quantity for the understanding of a river system. Therefore, methods have been developed to quantify the size of gravels in rivers already decades ago. These methods involve time-intensive fieldwork and bear the risk of introducing sampling biases. More recently, low-cost UAV (unmanned aerial vehicle) platforms have been employed for the collection of referenced images along rivers with the aim to determine the size of grains. To this end, several methods to extract pebble size data from such UAV imagery have been proposed. Yet, despite the availability of information on the precision and accuracy of UAV surveys, a systematic analysis of the uncertainty that is introduced into the resulting grain size distribution is still missing.

Here we present the results of three close-range UAV surveys conducted along Swiss gravel-bed rivers with a consumer-grade UAV. We use these surveys to assess the dependency of grain size measurements and associated uncertainties from photogrammetric models, in turn generated from segmented UAV imagery. In particular, we assess the effect of (i) different image acquisition formats, (ii) specific survey designs, and (iii) the orthoimage format used for grain size estimates. To do so, we use uncertainty quantities from the photogrammetric model and the statistical uncertainty of the collected grain size data, calculated through a combined bootstrapping and Monte Carlo (MC) modelling approach.

First, our preliminary results suggest some influence of the image acquisition format on the photogrammetric model quality. However, different choices for UAV surveys, e.g., the inclusion of oblique camera angles, referencing strategy and survey geometry, and environmental factors, e.g., light conditions or the occurrence of vegetation and water, exert a much larger control on the model quality. Second, MC modelling of full grain size distributions with propagated UAV uncertainties shows that measured size uncertainty is at the first order controlled by counting statistics, the selected orthoimage format, and limitations of the grain size determination itself, i.e., the segmentation in images. Therefore, our results highlight that grain size data are consistent and mostly insensitive to photogrammetric model quality when the data is extracted from single, undistorted orthoimages. This is not the case for grain size data, which are extracted from orthophoto mosaics. Third, upon looking at the results in detail, they reveal that environmental factors and specific survey strategies, which contribute to the decrease of the photogrammetric model quality, also decrease the detection of grains during image segmentation. Thereby, survey conditions that result in a lower quality of the photogrammetric model also lead to a higher uncertainty in grain size data.

Generally, these results indicate that even relative imprecise and not accurate UAV imagery can yield acceptable grain size data for some applications, under the conditions of correct photogrammetric alignment and a suitable image format. Furthermore, the use of a MC modelling strategy can be employed to estimate the grain size uncertainty for any image-based method in which individual grains are measured.

How to cite: Mair, D., Do Prado, A. H., Garefalakis, P., Lechmann, A., and Schlunegger, F.: Uncertainty of grain sizes from close-range UAV imagery in gravel bars, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3516, https://doi.org/10.5194/egusphere-egu22-3516, 2022.

The main type of research material is multi-season aerial photography of the oil mining karst river basin was carried out by unmanned aerial vehicle.

Visual photo delineation revealed the consequences of mechanical transformations, some hydrocarbon inputs (bitumization) and salts (technogenic salinization) were also identified. The last processes were verified using materials from direct geochemical surveys (chemical analyses of soils, surface waters and sets of ordinary photo of sample plots).

It has been established that mechanical transformations, as a rule, is detected by the color and shape of objects. Less often, it is necessary to additionally analyze indirect photo delineation signs: shape of the shadow, configuration of the borders, traces of heavy vehicle tracks. Photo delineation signs of technogenic salinization are turbidity of water and the acquisition of a bluish-whitish color; the change of the color of the water body to green-yellow; white ground salt spots. The bituminization process is sufficiently reliably identified only in the presence of open oil spills on the surface of soil or water. Despite the difficulty of photo delineation, the use of orthophotos allows to identify 13 new sites (26 in total in the studied area) of the processes of bitumization and technogenic salinization, which had not been noted during previous large-scale field survey.

The use of orthophotos to detect the processes of bitumization and technogenic salinization is effective, especially in combination with direct field studies. Conditions for using aerial photography to identify the consequences of oil mining technogenesis: pixel resolution should be equals or more precise than 20 cm / pixel (more desirable – equals or more precise than 10 cm / pixel), snowless shooting season, lack or low level of cloud cover, relatively low forest cover percent. The spatial distribution of the identified areas of all types of technogenesis indicates a close relationship with the location of oil mining facilities.

A promising direction for the development of the research is associated with the use of multispectral imaging, the improvement of attend field surveys, as well as the expansion of the experience of aerial photography of oil fields located in other natural conditions.

The reported study was funded by Russian Foundation for Basic Research (RFBR) and Perm Territory, project number 20-45-596018.

How to cite: Sannikov, P., Khotyanovskaya, Y., and Buzmakov, S.: Applicability of aerial photography for identifying of oil mining technogenesis: mechanical transformations, bitumization, technogenic salinization, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2643, https://doi.org/10.5194/egusphere-egu22-2643, 2022.

Photogrammetry and Structure-from-Motion have become widely assessed tools for geomorphological 3D reconstruction, and especially for monitoring remote and hardly accessible alpine environments. UAV-based photogrammetry enables large mountain areas to be modelled with high accuracy and limited costs. However, they still require a human intervention on-site. The use of fixed time-lapse cameras for retrieving qualitative and quantitative information on glacier flows have recently increased, as they can provide images with high temporal frequency (e.g., daily) for long-time spans, and they require minimum maintenance. However, in many cases, only one camera is employed, preventing the use of photogrammetry to compute georeferenced 3D models. This work presents a low-cost stereoscopic system composed of two time-lapse cameras for continuously and quantitatively monitoring the north-west tongue of the Belvedere Glacier (Italian Alps), by using a photogrammetric approach. Each monitoring station includes a DSLR camera, an Arduino microcontroller for camera triggering, and a Raspberry Pi Zero with a SIM card to send images to a remote server through GSM network. The instrumentation is enclosed in waterproof cases and mounted on tripods, anchored on big and stable rocks along the glacier moraines. The acquisition of a defined number of images and the timing can be arbitrary scheduled, e.g., 2 images per day acquired by each camera, around noon. A set of ground control points is materialized on stable rocks along the moraines and measured with topographic-grade GNSS receivers at the first epoch to orient stereo-pairs of images. From daily stereo-pairs, 3D models are computed with the commercial Structure from Motion software package Agisoft Metashape, and they can be used to detect morphological changes in the glacier tongue, as well as to compute daily glacier velocities. The work is currently focused on improving the orientation of stereo-pairs: the use of computer vision algorithms is under study to automatize the process and increase the robustness of consecutive orientation of stereo-images, e.g., by including images coming from different epochs in the same bundle block adjustment and dividing them afterwards for dense 3D reconstruction. Change detection can be then computed from 3D point clouds by using M3C2 algorithms. Although the stereoscopic system is already installed on the Belvedere Glacier and it is properly taking daily images of the glacier tongue, the processing workflow of stereo-pairs needs to be tuned and automatized to enable high-accurate continuous 3D photogrammetric monitoring of an alpine glacier, computing short-term and infra-seasonal ice volume variations and velocities, as well as detecting icefalls.

How to cite: Ioli, F., Bianchi, A., Cina, A., De Michele, C., and Pinto, L.: Time-lapse stereo-cameras and photogrammetry for continuous 3D monitoring of an alpine glacier, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7967, https://doi.org/10.5194/egusphere-egu22-7967, 2022.

Before modern remote sensing techniques, quantifying rock wall retreat due to rockfall events in the high alpine environment was limited to low-frequency post-event measurements for high-magnitude events. LiDAR and SFM now provide precise and accurate 3D models for computing 3D volume changes over time. Otherwise, mid- and low-sized events can remain unobserved due to the remoteness of the rockwalls and the lack of remnant evidence due to the rapid sequestration of ice in surrounding valley and cirque glaciers. To extend rockfall event measurement an initial measurement (t0) is necessary. The Mont-Blanc Massif (MBM, European Alps) High Resolution Topography Project is currently completing high-precision 3D models in the MBM using ground-based and aerial LiDAR, and drone-based structure-from-motion (SFM). In 2021, we began acquisition with initial measurements of 11 major sectors of the massif, representing about 80 km² of rock and ice slopes, between 1700m - 4810m in elevation. By choosing a study area with robust existent photographic and film archives, such as the MBM, it is possible to extend 3D models back in time for comparison with current datasets. Despite existent high-quality image archives, SFM processing is more challenging and error-prone than from contemporary images due to a lack of metadata, such as camera and lens type, precise dates of images, and the general degradation of the original material.  Despite these limitations, the use of historical-image-based SFM in combination with modern LiDAR data can allow the reconstruction of significant slopes of the MBM over several decades in order to i) obtain estimates of erosion rates, ii) to document rockfall events, and iii) to quantify the extent change and volume loss of hanging glaciers and ice aprons. We thus explore geomorphic processes in the high mountain environment in context of warming climate, as well as the limits of input data (image sets) in terms of practical output resolution.

How to cite: Uhlmann, D., Jaboyedoff, M., Derron, M.-H., Ravanel, L., Vicari, J., Wolff, C., Fei, L., Choanji, T., and Gutierrez, C.: High-resolution topography project on the rock walls of the Mont-Blanc massif to reconstruct volume change, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10943, https://doi.org/10.5194/egusphere-egu22-10943, 2022.

Pyrenean glaciers have shown a marked area and thickness decrease in the last century, especially in the last decades, and currently are highly threatened by climate change. Out of the 39 glaciers existing in the Pyrenees in 1984, 23 very small glaciers remain in this mountain range, from which only four have more than 10 ha. Probably, the most emblematic glacier of these four is Aneto glacier as it is located in the North-East face of the highest summit in the Pyrenees, the Aneto peak (3404 m a.s.l.). This work presents the Aneto glacier surface reconstruction from aerial images obtained in 1981, and its comparison with the glacier surface obtained in 2021 with Unmanned Aerial Vehicles (UAV) images.

The 1981 and 2021 images have been processed with Structure from Motion (SfM) algorithms to reconstruct the Digital Surface Model (DSM) of the glacier and nearby terrain. Taking advantage of the accurate geolocation of the UAV images in 2021 (GPS with RTK/PPK surveying), the DSM obtained has a precise representation of the glacier surface. Oppositely the aerial images of 1981 lack precise geolocation and thus require a post-processing analysis. The aerial images of the '80s have been firstly geolocated with Ground Control Points (GCPs) of known coordinates within the study area (summits, crests, and rock blocks with unaltered position). After this initial geolocation, the DSM of 1981 was generated with SfM algorithms. Nevertheless, this DSM still lacks a geolocation accuracy. To allow a comparison between the 1981 and the 2021 DSMs, the glacier surface in 1981 was registered to the 2021 surface with an Iterative Close Point (ICP) routine in the surrounding area of the glacier. The technique described in this work may be applicable to other historical aerial images, which may allow studying glacier evolutions all over the world for dates without field observations.

The surface comparison generated with images that have a temporal difference of 40 years has shown the dramatic area and thickness loss of this glacier, with areas decreasing more than 68 m, and an average thickness reduction of 31.5 m. In this period, the glacier has reduced its extent by about a 60%. There is a recent acceleration in the rate of shrinkage if we compare these data with the obtained for the period 2011-2021, in which area loss reaches 15% and thickness reduction almost reaches 10 m. During the 1981-2021 period the shrinkage rate is 0.78 m thickness/year and 1.5% area/year, meanwhile, during the 2011-2021 period the shrinkage rate is 0.99 m thickness/year and 2.7% area/year.

How to cite: Vidaller, I., Revuelto, J., Izagirre, E., García, J., Rojas-Heredia, F., and López-Moreno, J. I.: Comparison of 3D surfaces from historical aerial images and UAV acquisitions to understand glacier dynamics: The Aneto glacier changes in 40 years, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3163, https://doi.org/10.5194/egusphere-egu22-3163, 2022.

Mountain glaciers are responding in concert to a warming global climate over the past century. However, on interannual to decadal time scales, glaciers show temporally non-linear dynamics and spatially heterogeneous response, as a function of regional climate forcing and local geometry. Deriving long-term geodetic glacier change measurements from historical aerial photography can inform efforts to understand and project future response. 

We present interannual to decadal glacier and geomorphic change measurements at multiple sites across Western North America from the 1950s until present. Glacierized study sites differ in terms of glacial geometry and climatology, from continental mountains (e.g., Glacier National Park) to maritime stratovolcanoes (e.g., Mt. Rainier). Quantitative measurements of glacier and land surface change are obtained from Digital Elevation Models (DEMs) generated using the Historical Structure from Motion (HSfM) package. We use scanned historical images from the USGS North American Glacier Aerial Photography (NAGAP) archive and other aerial photography campaigns from the USGS EROS Aerial Photo Single Frames archive. 

The automated HSfM processing pipeline can derive high-resolution (0.5-2.0 m) DEMs and orthomosaics from scanned historical aerial photographs, without manual ground control point selection. We apply a multi-temporal bundle adjustment process using all images for a given site to refine both extrinsic and intrinsic camera model parameters, prior to generating DEMs for each acquisition date. All historical DEMs are co-registered to modern reference DEMs from airborne lidar, commercial satellite stereo or global elevation basemaps. The co-registration routine uses a multi-stage Iterative Closest Point (ICP) approach to achieve high relative alignment accuracy amongst the historical DEMs, regardless of reference DEM source. 

We examine the impact of regional climate forcing on glacier elevation change and dynamics using downscaled climate reanalysis products. By augmenting the record of quantitative glacier elevation change measurements and examining the relationship between climate forcing and heterogeneous glacier response patterns, we aim to improve our understanding of regional glacier mass change across multiple temporal scales, as well as inform management decisions impacting downstream water resources, ecosystem preservation, and geohazard risks.

How to cite: Knuth, F., Shean, D., McNeil, C., Schwat, E., and Bhushan, S.: Historical Structure From Motion (HSfM): An automated historical aerial photography processing pipeline revealing non-linear and heterogeneous glacier change across Western North America, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10190, https://doi.org/10.5194/egusphere-egu22-10190, 2022.

Long-term records of glaciers are more than ever crucial to understand their response to climate change. High-quality photogrammetric products, Digital Elevation Models (DEMs) and orthophotographs from early satellites are essential, as they offer a unique high-resolution view on the historical glacial dynamics. However, obtaining and producing high-resolution datasets from historical imagery can be a challenge.

In our study, we are extending available satellite images time series using images from Soviet Era KFA-1000 satellite cameras. Each KFA-1000 has a 1000 mm objective, holding 1800 frames in its magazine. Each frame is typically 18x18 cm or 30 × 30 cm, with an 80 km swath width, providing panchromatic images. They supplement the very sparse data period between aerial images and high-resolution modern satellites, giving us high-resolution insight of Antarctica and Greenland dating from 1974 to 1994. Since these images have been largely underused, they have the potential to improve our knowledge of glaciers and open new scientific perspectives. They could help us improve models in studies regarding, for instance the frontal position, the flow-velocity (by doing feature tracking), the surface elevation or the grounding line of the glaciers, etc. With a spatial resolution up to 2 m and images recorded in stereo geometry, they offer a valuable complement to other historical satellite archives such as the declassified American KH imagery. Here, we use structure-from-motion (SfM) to reconstruct former glacier surfaces and flow of main outlet glaciers in both Antarctica and Greenland. We compare and assess the quality of the results by comparing the produced DEMs with recent high-resolution imagery from Worldview’s ArcticDEM. We combine the historical DEMs with recent satellite imagery of the ice elevation and reconstruct the comprehensive history of volume change over southeast and northeast Greenland glaciers since the 90s. Mostly lost from sight for 50 years, we are now resurrecting these highly valuable records and will make them freely available to science and the public.

How to cite: Huiban, F., Dømgaard, M., Girod, L., Millan, R., Dehecq, A., Mouginot, J., Schomacker, A., Rignot, E., and Bjørk, A.: Expanding glacier time series of Antarctica and Greenland using Soviet Era KFA-1000 satellite images, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7374, https://doi.org/10.5194/egusphere-egu22-7374, 2022.

The USGS digitized many historical photos of Antarctica which could provide useful insights into this region from before the satellite era. However, these images are merely scanned and do not contain semantic information, which makes it difficult to use or search this archive (for example to filter for cloudless images). Even though there are countless semantic segmentation methods, they are not working properly with these images. The images are only grayscale, have often a poor image quality (low contrast or newton’s rings) and do not have very distinct classes, for example snow/clouds (both white pixels) or rocks/water (both black pixels). Furthermore, especially for this archive, these images are not only top-down but can also be oblique.

We are training a machine-learning based network to apply semantic segmentation on these images even under these challenging conditions. The pixels of each image will be labelled into one of the six different classes: ice, snow, water, rocks, sky and clouds. No training data was available for these images, so that we needed to create it ourselves. The amount of training data is therefore limited due to the extensive amount of time required for labelling. With this training data, a U-Net was trained, which is a fully convolutional network that can work especially with fewer training images and still give precise results.

In its current state, this model is trained with 67 images, split in 80% training and 20% validation images. After around 6000 epochs (approx. 30h of training) the model converges and training is stopped. The model is evaluated on 8 randomly selected images that were not used during training or validation. These images contain all different classes and are challenging to segment due to quality flaws and similar looking classes. The model is able to segment the images with an accuracy of around 75%. Whereas some classes, like snow, sky, rocks and water can be recognized consistently, the classes ice and clouds are often confused with snow. However, the general semantic structure of the images can be recognized.

In order to improve the semantic segmentation, more training imagery is required to increase the variability of each class and prepare the model for more challenging scenes. This new training data will include both labelled images from the TMA archive and from other historical archives in order to increase the variability of classes even more. It should be checked if the quality of the model can be further improved by including metadata of the images as additional data sources.

How to cite: Dahle, F., Lindenbergh, R., Tanke, J., and Wouters, B.: Semantic segmentation of historical images in Antarctica with neural networks, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10597, https://doi.org/10.5194/egusphere-egu22-10597, 2022.

Historical aerial imagery dating back to the mid-twentieth century offers high potential to distinguish anthropogenic impacts from natural causes of environmental change and reanalyze the long-term surface evolution from local to regional scales. However, the older portion of the imagery is often acquired in panchromatic grayscale thus making image classification a very challenging task.  This research aims to compare deep learning image colorisation methods, namely, , the Neural Style Transfer (NST) and the Cycle Generative Adversarial Network (CycleGAN), for colorizing archival images of Japanese river basins for land cover analysis. Historical monochrome images were examined with `4096 x 4096` pixels of three river basins, i.e., the Kurobe, Tenryu, and Chikugo Rivers. In the NST method, we used the transfer learning model with optimal hyperparameters that had already been fine-tuned for the river basin colorization of the archival river images (Ishii et al., 2021). As for the CycleGAN method, we trained the CycleGAN with 8000 image tiles of `256 x256` pixels to obtain the optimal hyperparameters for the river basin colorization. The image tiles used in training consisted of 10 land-use types, including paddy fields, agricultural lands, forests, wastelands, cities and villages, transportation land, rivers, lakes, coastal areas, and so forth. The training result of the CycleGAN reached an optimal model in which the root mean square error (RMSE) of colorization was 18.3 in 8-bit RGB color resolution with optimal hyperparameters of the dropout ratio (0.4), cycle consistency loss (10), and identity mapping loss (0.5). Colorization comparison of the two-deep learning methods gave us the following three findings. (i) CycleGAN requires much less training effort than the NST because the CycleGAN used an unsupervised learning algorithm. CycleGAN used 8000 images without labelling for training while the NST used 60k with labelling in transfer learning. (ii) The colorization quality of the two methods was basically the same in the evaluation stage; RMSEs in CycleGAN were 15.4 for Kurobe, 13.7 for Tenryu and 18.7 for Chikugo, while RMSE in NST were 9.9 for Kurobe, 15.8 for Tenryu, and 14.2 for Chikugo, respectively. (iii) The CycleGAN indicated much higher performance on the colorization of dull surfaces without any textual features, such as the river course in Tenryu River, than the NST. In future research work, colorized imagery by both the NST and CycleGAN will be further used for land cover classification with AI technology to investigate its role in image recognition. [Reference]: Ishii, R. et al.(2021) Colorization of archival aerial imagery using deep learning, EGU General Assembly 2021, EGU21-11925, https://doi.org/10.5194/egusphere-egu21-11925.

How to cite: Tanaka, S., Miyamoto, H., Ishii, R., and Carbonneau, P.: Comparison of deep learning methods for colorizing historical aerial imagery, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7686, https://doi.org/10.5194/egusphere-egu22-7686, 2022.