Accurate remote analyses of high-alpine landslides are a key requirement for future alpine safety. In critical stages of alpine landslide evolution, UAS (unmanned aerial system) data can be employed using image registration to derive ground motion with high temporal and spatial resolution. However, classical area-based algorithms suffer from dynamic surface alterations and their limited velocity range restricts detection, resulting in noise from decorrelation and hindering their application to fast landslides. Here, to reduce these limitations we apply for the first time the optical flow-time series to landslides for the analysis of one of the fastest and most critical debris flow source zones in Austria. The benchmark site Sattelkar (2130–2730 m asl), a steep, high-alpine cirque in Austria, is highly sensitive to rainfall and melt-water events, which led to a 70,000 m³ debris slide event after two days of heavy precipitation in summer 2014. We use a UAS data set of five acquisitions (2018–2020) over a temporal range of three years with 0.16 m spatial resolution. Our new methodology is to employ optical flow for landslide monitoring, which, along with phase correlation, is incorporated into the software IRIS. For performance testing, we compared the two algorithms by applying them to the UAS image stacks to calculate time-series displacement curves and ground motion maps. These maps allow the exact identification of compartments of the complex landslide body and reveal different displacement patterns, with displacement curves reflecting an increased acceleration. Visually traceable boulders in the UAS orthophotos provide independent validation of the methodology applied. Here, we demonstrate that UAS optical flow time series analysis generates a better signal extraction, and thus less noise and a wider observable velocity range—highlighting its applicability for the acceleration of a fast, high-alpine landslide.

Abstract. While optical remote sensing has demonstrated its capabilities for landslide detection and monitoring, spatial and temporal demands for landslide early warning systems (LEWSs) had not been met until recently. We introduce a novel conceptual approach to structure and quantitatively assess lead time for LEWSs. We analysed “time to warning” as a sequence: (i) time to collect, (ii) time to process and (iii) time to evaluate relevant optical data. The difference between the time to warning and “forecasting window” (i.e. time from hazard becoming predictable until event) is the lead time for reactive measures. We tested digital image correlation (DIC) of best-suited spatiotemporal techniques, i.e. 3 m resolution PlanetScope daily imagery and 0.16 m resolution unmanned aerial system (UAS)-derived orthophotos to reveal fast ground displacement and acceleration of a deep-seated, complex alpine mass movement leading to massive debris flow events. The time to warning for the UAS/PlanetScope totals 31/21 h and is comprised of time to (i) collect – 12/14 h, (ii) process – 17/5 h and (iii) evaluate – 2/2 h, which is well below the forecasting window for recent benchmarks and facilitates a lead time for reactive measures. We show optical remote sensing data can support LEWSs with a sufficiently fast processing time, demonstrating the feasibility of optical sensors for LEWSs.

<p>Remote sensing for natural hazard assessment and applications offers data on even vast areas, often difficult and dangerous to access. Today, satellite data providers such as PlanetLabs Inc. and the European Copernicus program provide a sub-weekly acquisition frequency of high resolution multispectral imagery. The availability of this high temporal data density suggests that the detection of short-term changes is possible; however, limitations of this data regarding qualitative, spatiotemporal reliability for the early warning of gravitational mass movements have not been analysed and extensively tested.</p><p>This study analyses the effective detection and monitoring potential of PlanetScope Ortho Tiles (3.125 m, daily revisit rate) and Sentinel-2 (10 m, 5-day revisit) satellite imagery between 2018 and 2020. These results are compared to high accuracy UAS orthoimages (0.16 m, 5 acquisitions from 2018-2020). The analysis is conducted based on digital image correlation (DIC) using COSI-Corr (Caltech), a well-established software and the newly developed IRIS (NHAZCA). The mass wasting processes in a steep, glacially-eroded, high alpine cirque, Sattelkar (2’130-2’730 m asl), Austria, are investigated. It is surrounded by a headwall of granitic gneiss with a cirque infill characterised by massive volumes of glacial and periglacial debris including rockfall deposits. Since 2003 the dynamics of these processes have been increased, and between 2012-2015 rates up to 30 m/a were observed.</p><p>Similar results are returned by the two software tools regarding hot-spot detection and signal-to-noise ratio; nonetheless IRIS results in an overall better detection, including a more delimitable ground motion area, with its iterative reference and secondary image combination. This analysis is supported by field investigations as well as clearly demarcated DIC-results from UAS imagery. Here, COSI-Corr shows limitations in the form of decorrelation and ambiguous velocity vectors due to high ground motion and surface changes for very high resolution of this input data. In contrast, IRIS performs better returning more coherent displacement rates. The results of both DIC tools for satellite images are affected by spatial resolution, data quality and imprecise image co-registration.</p><p>Knowledge of data potential and applicability is of high importance for a reliable and precise detection of gravitational mass movements. UAS data provides trustworthy, relative ground motion rates for moderate velocities and thus the possibility to draw conclusions regarding landslide processes. In contrast satellite data returns results which cannot always be clearly delimited due to spatial resolution, precision, and accuracy. Nevertheless, iterative calculations by IRIS improve the validity of the results.</p>

<p>New remote sensing systems offer an increased spatiotemporal resolution and accuracy. These systems  increase the chance of snow- and cloud-free multispectral images to detect and monitor landslides for early warning issues. Various studies showed the applicability of multispectral remote sensing systems for landslide detection and monitoring. However, a systemic evaluation of the remote sensing systems especially in respect to early warning is still missing. In this study we present a new conceptional approach to evaluate the capability of different systems for early warning issues based on a well suited case study located in the Hohe Tauern Range, Austria.     </p><p>The Sattelkar is a highly dynamic west-facing deglaciated high-alpine cirque in the Großvenedigergruppe, Austria. The abundant rock debris exhibits high movement rates and showed massively enhanced landslide activity after ongoing heavy precipitation in 08/2014, resulting in a 170.000 m³ debris flow event. We estimated time demands for three successive steps consisting of (i) image collection, (ii) processing with motion delineation and (iii) the final evaluation. Digital image correlation, an established tool in landslide remote-sensing research, was used to derive displacement patterns and assess the capabilities of the multispectral images in terms of spatiotemporal resolution and data quality. For our study we used Sentinel-2, RapidEye and PlanetScope images and compared their deduced motion patterns and rates to those from accurate UAV data as well as manually digitized boulder tracks (≥10 m in diameter).</p><p>Within a reasonable amount of processing time, some satellite data revealed similar clustered motions identifiable in the UAV images. However, our analysis also showed identification limitations due positional inaccuracy, image errors and spatiotemporal resolution of the data. On that account, certain processing steps reduce the forecasting window and as a result the lead time, i. e. the remaining time to take action. We postulate that remote sensing data has the ability to support landslide monitoring, but the pre-selection of usable and sound data is essential as it directly influences the forecasting window. Sound knowledge of its different application possibilities enhances overall steps of image collection, processing and final analysis. The critical selection of which data source is best can lead to faster response times for landslide events. This increases the forecasting window, hence the time to take action until a landslide occurs.</p>

<p>With the combination of diverse remote sensing data, one can estimate the detection capabilities of gravitational mass movement dynamics and behaviour. Recent multispectral satellite sensors such as Sentinel-2, RapidEye and PlanetScope offer unprecedented spatiotemporal resolutions, hence reducing data gaps of alpine meteorological constraints. In addition to this data, very high resolution and accurate UAV images cover a broad range of spatial resolutions. The strengths of these remote sensing systems allow the data compilation of vast, difficult and dangerous to access mountain areas. However, the limitations of the spatiotemporal resolution for (i) pre-event landslide detection, (ii) monitoring of already known mass movements and (iii) the capability to measure rapid changes (e.g.  accelerations) for warnings have not been examined extensively. Thus, there is an important need to understand the potential of multispectral images to detect, monitor, and identify rapid changes prior to landslide events to increase the forecasting window.</p><p>Digital image correlation (DIC), as indispensable tool to measure surface displacements, aids in estimating the fitness of different remote sensing images. Here, we present first results of motion delineation by DIC of the Sattelkar, a high-alpine, deglaciated and debris-laden cirque in the Obersulzbach-valley, Austria. We used comprehensive knowledge of the study site to thoroughly understand DIC motion clusters for verification purposes. We then compared three different DIC software tools, COSI-Corr, DIC‑FFT and IMCORR. They revealed similar results for the three satellite systems in terms of hot spot areas as well as noise. Our findings show large motion inaccuracies for Sentinel-2, RapidEye and PlanetScope images due to spatial resolution, poor image co-registration and changing data quality. In contrast, displacement patterns from the three UAV images (7/2018, 7/2019, 9/2019) demonstrate good positional accuracy as well as data usability for this approach. The inherited noise results from decorrelation due to high velocities suggest using an increased temporal image acquisition for further evaluation.</p><p>Reliable, precise results for landslide detection, their ongoing monitoring and the measurement capability for significant changes are necessary for targeted investigations, precautionary measures and the start of the forecasting window. Multispectral UAV images of high positional accuracy and quality are able to provide dependable relative displacement velocities and have the capability to serve as a reliable tool. On the contrary, satellite images showed delusive results, and we recommend reconsidering their deployment in future applications. The knowledge of the most suitable data in terms of accuracy and processing speed is crucial for landslide identification, monitoring and acceleration threshold detection. At present, our prelimiary findings show the capability to detect and monitor relative and mainly slow changes. The detection of rapid changes lacks due to the accuracy, resolution and revisit time of the investigated remote sensing systems.</p>