Modelling landslide dynamics in forested landscapes

The research resulting in this thesis covers the geological, geomorphological and landscape ecology related themes of the project 'Podzolisation under Kauri (Agathis australis): for better or worse?' supported by theNetherlands Organisation for Scientific Research (NWO). The general objective of this thesis is to investigate landscape, soil and vegetation dynamics in theWaitakereRangesRegionalParkon the North Island of New Zealand, where also all the fieldwork was carried out. The main core of the thesis consists of the development of a dynamic landscape process model to simulate soil redistribution by shallow landsliding. Resulting spatial patterns of erosion and deposition, changes in landslide susceptibility over time and the relation of spatially explicit landscape attributes with vegetation patterns are further explored.·Chapter 1 is a general introduction elaborating on the geology, climate and socio-economic setting of the study area and explains the main objectives and research questions. The contents and overall structure of the thesis are also illustrated. Following this introductory chapter, the thesis is composed of 5 chapters based on scientific papers published in or submitted to peer reviewed journals. ·Chapter 2 deals with the general tectonic setting of the study area. Quaternary coastal and fluvial terrace morphology and chronology are explored to reconstruct the tectonic history of the south-west coast of the Northland region inNew Zealand. This chapter is situated on the geological timescale (1.8 Ma BP till present) and places the subsequent chapters dealing with the landscape process model and its applications, acting on a timescale of years to decades, in a broader spatio-temporal perspective. Field surveys and the analysis of aerial photography yield an inventory of 13 fluvial and 12 marine terrace levels. Due to poor exposure of clear field evidence in the form of e.g. wave-cut platforms or distinct river sediments, planar landscape morphology forms the main criterion for terrace remnant identification. Based on the record of terrace height spacings, sparse tephra age control and correlation with global paleoclimatic records, an attempt is made to reconstruct the regional Quaternary uplift rates. Because no hard chronostratigraphic marker is present within the fluvial terrace sequence, fluvial terrace levels are linked to the marine sequence by using the mean uplift rates calculated from the marine terraces (0.35 mm yr -1 from 0- 0.1 Ma and 0.26 mm yr -1 from 0.1-0.3 Ma). Both sets of terraces are then correlated with oxygen isotope fluctuations and the astronomically tuned timescale from ODP Site 677 and the Vostok ice core paleoclimatic records. Oldest marine and fluvial terrace levels are estimated 1.21 Ma and 0.242 Ma respectively. Although there seems to be some form of controversy about the uplift history and especially the preservation of terraces in the study area, a general regional uplift, superimposed on glacio-eustatic sea-level changes, is substantiated as the only possible mechanism leading to the maintenance of a considerable relief and active denudation processes inland.·Chapter 3 deals with the development and application of the LAPSUS-LS landscape process model. The model is constructed as a component of the LAPSUS modelling framework ( L andsc A pe P roces S modelling at m U lti dimensions and scale S ; -LS:L and S lide,refers to the process specific model component). LAPSUS-LS delineates the location of shallow landslide initiation sites and simulates the effects on spatial patterns of soil redistribution and resulting landslide hazard for a large watershed within the study area. Processes that need to be incorporated in the model are reviewed followed by the proposed modelling framework. The model predicts the spatial pattern of landslide susceptibility within the simulated catchment and subsequently applies a spatial algorithm for the redistribution of failed material on the basis of a scenario of triggering rainfall events, relative landslide hazard and trajectories with runout criteria for failed slope material. The model forms a spatially explicit method to address the effects of shallow landslide erosion and sedimentation because digital elevation data are adapted between timesteps and on- and off-site effects over the years can be simulated in this way. By visualisation of the modelling results in a GIS environment, the shifting pattern of upslope and downslope (in)stability, triggering of new landslides and the resulting slope retreat by soil material redistribution due to former mass movements is simulated and assessed.·Chapter 4 zooms in on a more theoretical aspect of the LAPSUS-LS model and evaluates digital elevation model (DEM) resolution effects on model results. The focus is on influences of grid size on landslide soil redistribution quantities and resulting spatial patterns and feedback mechanisms. Distributions of slope, specific catchment area and relative hazard for shallow landsliding are analysed for four different DEM resolutions (grid sizes of 10, 25, 50 and 100 m) for a 12 km 2 study catchment in theWaitakereRanges. The effect of DEM resolution proves to be especially pronounced for the boundary conditions determining a valid landslide hazard calculation. For coarse resolutions, the smoothing effect results in a larger area becoming classified as unconditionally stable or unstable. Simple empirical soil redistribution algorithms are applied for scenarios in which all sites with a certain landslide hazard fail and generate debris flow. The lower initial number of failing cells but also the inclusion of slope (limit) in those algorithms becomes apparent with coarser resolutions. For finer resolutions, much larger amounts of soil redistribution are found, which is attributed to the more detailed landscape representation. Looking at spatial patterns of landslide erosion and sedimentation, the size of the area affected by these processes also increases with finer resolutions. In general, landslide erosion occupies larger parts of the area than deposition, although the total amounts of soil material eroded and deposited are the same. Analysis of feedback mechanisms between soil failures over time shows that finer resolutions show higher percentages of the area with an increased or decreased landslide hazard. When the extent of sites with lower and higher hazards are compared, finer grid sizes and higher landslide hazard threshold scenarios tend to increase the total extent of areas becoming more stable relative to the less stable ones. It is concluded that extreme care should be taken when quantifying landslide basin sediment yield by applying simple soil redistribution formulas to DEMs with different resolutions. Rather, quantities should be interpreted as relative amounts. For studying shallow landsliding over a longer timeframe, the 'perfect' DEM resolution may not exist, because no resolution can possibly represent the dimensions of all different slope failures scattered in space and time. It is emphasised that the choice of DEM resolution, possibly restricted by data availability in the first place, should always be adapted to the context of a particular type of analysis.·Chapter 5 and 6 describe two distinct applications of the LAPSUS-LS model: in Chapter 5 , a sediment record is used, in combination with the LAPSUS-LS model, to reconstruct the incidence of high-magnitude/low-frequency landslide events in the upper part of theWaitakereRivercatchment and the history of the Te Henga wetland at the outlet. Sediment stratigraphy and chronology are interpreted by radiocarbon dating, foraminiferal analysis, andprovisionaltephrochronology. Gradual impoundment of the wetland began c. 6000 cal yr BP, coinciding with the start of a gentle sea-level fall, but complete damming and initial sedimentation did not begin until c. 1000 cal yr BP. After damming, four well-defined sediment pulses occurred and these are preserved in the form of distinct clay layers in most of the sediment cores. For interpreting the sediment pulses, the LAPSUS-LS modelisapplied to determine spatially distributed relative landslide hazard, applicable at the catchment scale. An empirical landslide soil redistribution componentisadded to determine sediment delivery ratio and the impact on total catchment sediment yield. Sediment volumesarecalculated from the wetland cores and corresponding landslide scenarios are defined through back-analysis of modelled sediment yield output. In general, at least four major high-magnitude landslide events, both natural and intensified by forest clearance activities, occurred in the catchment upstream of Te Henga wetland during the last c. 1000 years. Their magnitude can be expressed by a range of critical rainfall thresholds representing a LAPSUS-LS scenario.·Chapter 6 is a more ecologically focused application of the model and links digital terrain analysis and landslide modelling with the spatial distribution of mature kauri trees. The use of topographical attributes for the analysis of the spatial distribution and ecological cycle of kauri ( Agathis australis ), a canopy emergent conifer tree from northernNew Zealand, is studied. Several primary and secondary topographic attributes are derived from a DEM for theWaitakereRivercatchment and the contribution of these variables in explaining presence or absence of mature kauri is assessed with logistic regression and Receiver Operating Characteristic (ROC) plots. The topographically based landslide hazard index calculated with the LAPSUS-LS model appears to be very useful in explaining the occurrence and ecological dynamics of kauri. It is shown that the combination of topographic -, soil physical - and hydrological parameters in the calculation of this single landslide hazard index, performs better in explaining presence of mature kauri than using topographic attributes calculated from the DEM properties alone. Moreover, this example demonstrates the possibilities of using terrain attributes for representing geomorphological processes and disturbance mechanisms, often indispensable in explaining a species' ecological cycle and forest stand dynamics. The results of this analysis support the 'temporal stand replacement model', involving disturbance as a dominant ecological process in forest regeneration, as an interpretation of the community dynamics of kauri. Furthermore, a certain threshold maturity stage, in which trees become able to stabilise landslide prone sites and postpone a possible disturbance by this process, together with great longevity are seen as major factors making kauri a 'landscape engineer'.·Synthesising, Chapter 7 reflects on the most important conclusions from the research resulting in this thesis and discusses the achievement of the main objectives and answers to the research questions postulated in Chapter 1. Three general themes are put forward covering the previous chapters. Finally some ideas for future research are suggested.