| dc.description.abstract | Nedbørsinduserte jordskred er en betydelig fare i Norge på grunn av bratte skråninger, mye nedbør, temperaturendringer og snøsmelting. Klimaendringer øker risikoen ytterligere. Prosjektet KlimaDigital, startet av SINTEF i 2018, har hatt som mål å utvikle et varslingssystem for nedbørsinduserte jordskred ved hjelp av sensorer.
I denne oppgaven er sensorverdier fra KlimaDigital-prosjektet brukt til å kalibrere de hydrologiske van Genuchten parameterne til jorda. Dette er gjort ved å bruke Ensemble Kalman Filter-algoritmen (EnKF). Simuleringene er gjort ved å lage en automatisert Plaxis-modell i Python og EnKF fungerer ved å oppdatere de ukjente inputparameterne basert på forskjellen mellom sensordata og Plaxis-output. Formålet er at Plaxis-outputen skal sammenfalle med sensorverdiene, noe som tyder på en virkelighetsnær modell. I tillegg er hensikten at de ukjente parameterne skal konvergere mot estimater med høyere sikkerhet.
Oppgaven konkluderer med at EnKF-algoritmen viser gode tendenser i estimatet av disse parameterne. I midlertidig er nøyaktigheten av de initielle betingelsene og forhåndskunnskapen om parameterne avgjørende for å oppnå mer pålitelige estimater. Oppgaven peker også på den mulige bruken av denne metoden i varslingssystemer for jordskred. Ytterligere for bedring og økt sikkerhet i resultatene er derimot nødvendig for å sikre pålitelighet.
Fremtidig arbeid vil kunne være å implementere sensorverdier for sug, samt kalibrere parameterne med verdier fra faktiske skredhendelser. | |
| dc.description.abstract | Rainfall-induced landslides make out a considerable amount of geohazards in Norway. This is from combination of steep slopes, heavy rainfall during fall, temperature changes, and snow melting in spring and fall. Landslides are dangerous because they can damage critical infrastructure and in the worst case be a threat to human life. Climate change is likely to result in more and intense rainfall events, which is further increasing the landslide risks. Consequently, there is a need to develop strategies to manage landslide risks. One of such strategies are landslide early warning systems that provide a timely that can be used to evacuate people, movable properties, or close roads to reduce consequences in case of a landslide.
In 2018, SINTEF launched the project KlimaDigital with the main objective of creating an early warning system for rainfall induced landslides and debris flow, using monitoring data from sensors installed in landslide prone areas. Sensors were installed on two different slopes in Meråker, Trøndelag to monitor groundwater conditions. The area was selected due to the
high risk for rainfall-induced landslides resulting from the combinations of heavy rainfall and rapid snow-melting events, and topography characterized by steep slopes (Ivan Depina E. O., 2021)
When working with rainfall induced landslide problems, it is crucial to understand groundwater conditions in slopes in response to rainfall infiltration and snow melting. Slopes in the studied area are typically unsaturated and groundwater conditions were monitored with volumetric water content sensors. In this thesis, the volumetric water content sensor data from the Meråker slopes was used to calibrate the hydrological van Genuchten model parameters and the permeability of the soil, using a Plaxis flow-analysis and the Ensemble Kalman Filter method (EnKF). The estimated van Genuchten parameters are defining the Soil Water Characteristics curve (SWCC), that describes relations between suction, permeability, and degree of saturation in the given soil. Calibrated hydrological models provide a basis for more accurate modelling of groundwater conditions (e.g., in response to extreme weather events) and implementation of a reliable landslide early warning system.
Python was used to implement the EnKF algorithm and automate Plaxis analysis of the hydrological model, making it possible to update and run the calculations numerous of times with different hydrological input parameters for each iteration. The Ensemble Kalman filter method works by comparing the Plaxis output values to the real sensor data and updating the parameters based on this difference and the parameter covariance. By iterating through this process, the unknown parameters were updated until the van Genuchten parameters were estimated with better accuracy and the difference between model predictions and sensor values became relatively small.
The study concludes that the EnKF algorithm shows promise in estimating these parameters; however, the accuracy of the initial conditions and prior parameter knowledge are critical for obtaining more reliable estimations. The research also highlights the potential application of this method in early warning systems for rainfall-induced landslides. However, further refinement and increased certainty in the results are necessary to ensure reliability.
Future work could include incorporation of suction sensors alongside the VWC-sensors to enhance the calibration results. It would also be interesting to implement the results in a safety analysis to investigate how the hydrological conditions influence slope stability. | |
