ANALYSES IN NORTH-EASTERN ICELAND

Arto Vuorela(1) and Jukka Käyhkö(2)

(1)Novosat Ltd

Opastinsilta 12 B, 00520 Helsinki, Finland

Email: arto.vuorela@novogroup.com

(2) Department of Geography

University of Turku, 20014 Turku, Finland

Email: jukka.kayhko@utu.fi

INTRODUCTION

This project concerns the origin and dynamics of the severe land degradation in north-eastern Iceland. The entire project (no. 187) within the ERS-AO3 scheme is titled “Environmental history of the severely eroded north-eastern Icelandic semi-desert – a multi-disciplinary approach utilising remotely sensed data combined with detailed investigations on palaeoecological, sedimentological and cultural aspects“. The sub-project described here aims at modelling the flood routes based on in-house produced InSAR digital elevation model (DEM). The modelling procedure can be divided into three main phases:

•Producing a high-accuracy DEM based on SAR interferometry

•Mapping the potential catastrophic flood routes using the DEM and a hydrological flow model in GIS environment.

•Mapping the lava types, sediment cover on the lava fields, ancient flood routes and, specifically with the aid of ERS-SAR data, quantifying and correcting the effects of moisture and shadow responses in the optical

data interpretation.

This report describes the rationale of the project, the DEM production environment of Novosat and the DEM construction phase of this project. It points out some general limitations and problems of InSAR and aims at assessing potential methods and strategies for resolving them and to improve the accuracy.

STUDY AREA

The study area is located in northern Iceland (Fig. 1). The region is characterised by severe erosion processes, volcanic activity, glaciofluvial processes and aridity due to the Vatnajökull rain-shadow. Most of the area is completely devoid of vegetation, whereas shrub heaths occupy the northern fringes along the coast. The area serves as a type example of an enigmatic region, where it is difficult to assess to what extent ecosystem changes or processes take place naturally, or as consequence of human interference.

Catastrophic floods (jökulhlaups)

Volcanic phenomena are common along the Northern Rift Zone of Iceland. Vigorous volcanic activity occurs beneath the Vatnajökull ice cap to the south of the study area [1] producing recurrently widespread tephra layers as well as catastrophic floods (jökulhlaup Icel.) [2]. Two major volcanic centres lie beneath the ice: the Barðarbunga and the Grimsvötn volcanic centre, both of which exhibit large subglacial caldera depressions. The Grimsvötn volcanic centre is the more active of the two with an eruption frequency close to one eruption per decade. The eruption in September-

Data provided by ESA. Financial support provided by Jenny and Antti Wihuri Foundation and the British Council.November 1996 resulted in a catastrophic flood of nearly 4 km3 of meltwater, which covered the uninhabited Skeiðarársandur region in the south (c.f. Fig. 1). The flood caused substantial material damage by destroying roads and bridges with an estimated cost of 10-15 million USD [3].

Had the October 1996 eruption taken place slightly further north, water would have resulted in jökulhlaup on the northern margin of the Vatnajökull [3]. The fact that traces of the meltwater from the 1996 eruption were detected in the northern rivers illustrates how close meltwater was to draining in a northerly direction. It is widely acknowledged that the Vatnajökull area is entering a period of renewed volcanic activity [3]. It is therefore probable that future eruptions will drain northwards posing a considerable hazard to communities. Based on preliminary investigations, it is hypothesised [4] that ancient catastrophic floods may have triggered some of the present day environmental (erosion) processes in the study area.

Landscape evolution

In addition to interferometry, ERS-SAR data allow determining the aerodynamic roughness of lava surfaces, which is an important parameter in studies of sand transport. Smooth pahoehoe lavas act as significant pathways for aeolian transport, whereas rough aa lavas appear sediment sinks, and barriers for the advancing sand. There are no previous maps of lava surface roughness in Iceland (or elsewhere) and the research group is in the process of developing new techniques in this field. An attempt will be made to divide lava flows into relative age classes by using remote sensing data. In addition, the techniques applied here will allow estimates of the impact of lava flows on drainage systems and vegetation. Thorough mapping of the surface roughness will also allow more accurate sand transport prediction, being of great aid in land reclamation.

The DEM will be used in estimates of eruption volumes. The calculations will be refined with morphological measurements in the field (thickness and width of lava flows). The research group is aware of the many complications in the calculation of volcanic eruption volumes [5], [6]. Therefore, many of our results on the eruptive volumes will be given as order-of-magnitude estimates rather than in absolute figures. These estimates will then be used in assessing environmental stresses and landscape evolution. The digital elevation model will also be used in tracking current and ancient routes of jökulhlaups. The elevation model can also be used in prediction of likely routes of future lava flows and jökulhlaups and is therefore useful in hazard assessment. Finally, the data on the relative and absolute age lava flow fields, volume calculations of eruptions and mapping of the extent of flow fields allow us to construct a general model on the landscape evolution of the area.

Fig. 1. Map of Iceland showing the active volcanic zone, the largest ice caps and the borders of the four ERS tandem pairs applied so far in the DEM construction.MATERIAL AND METHODS

ERS-SAR data

Four ERS-1/2 tandem SLC image pairs were employed in the DEM construction. In the first phase, images with snow cover on the ground (October-February) were ordered (Table 1). Later, a set of snow-free images was ordered, as the coherence of the latter proved to be better. See Discussion for further details on the problems of the material selection. Topographic maps on scales 1:50 000 and 1:100 000 were employed in the combining of the individual models. The 1:50 000 map set did not cover the whole study area. Furthermore, the projection and datum of the 1:100 000 maps were inconsistent with the other data, and were used mainly for elevation assessment.

Table 1. The ERS tandem pairs processed. The highlighted images showed poor coherence and were discarded after pre-processing (see text for details)

SAT ORBIT FRAME SHIFT ACQ.DATE

E124045225909960219

E204372225909960220

EI240452295-2960219

E2043722295-2960220

E1275441305-2961020

E2078711305-2961021

E1275441323-2961020

E2078711323-2961021

E1217692277-1950913

E2020962277-1950914

E1217692295-1950913

E2020962295-1950914

Methods

The interferometric processing was carried out by Novosat Ltd, Helsinki, Finland. The software employed in the process is an in-house product, implemented by Dr. Einar-Arne Herland at the Remote Sensing Group of the Technical Research Centre (VTT) of Finland [7]. The modus operandi is based on the long-term research on SAR data and techniques at VTT, and the method was implemented in collaboration with Novosat. Currently, the software is in operational use for commercial projects, and Novosat is marketing large-area DEM’s based on ERS SLCI data. Comparison procedures against a 25 m raster DEM extracted from a 1:20 000 topographic map (National Land Survey of Finland) have revealed that the InSAR elevation models show typically a vertical accuracy of 5–15 m [8]. In South Finland, a 4 metre RMSE has been accomplished previously.

The software offers highly automated pre-processing of the data. Correlation, slave interpolation, fringe and coherence calculation, fringe filtering, breakpoint and contour generation are carried out automatically. For phase unwrapping, fully automated - but also less reliable - methods were not employed. Instead, unwrapped phase is manually integrated over the total interferogram area, based on branch and cut methods (Fig. 2). Various tools help in this task; e.g. the coherence image is used as a threshold mask and can be viewed in parallel during the work. In addition, the slant and map rectification and advanced joining of the independent models are conducted. The interface has been designed as user-friendly, and the icon-based alternatives offer in-built state-of-the-art InSAR techniques. Due to general limitations of SAR interferometry, Novosat has lately also adapted radargrammetry processes as a complementary alternative for DEM production.

Fig. 2. A colour image, showing the unwrapping of a relatively good-coherence interferogram. Lakes, wetlands and the instrument side of the hills show low coherence. The coherence image can be used as an aid to conduct the unwrapping. The fringe disconnections and the extra fringes on the instrument side of hill can be interpreted and drawn with the blue mask area, remaining wrapped. These hollows can be interpolated later. The yellow mask is the so far unwrapped area. Coherence images can be also used in weighting the DEM mosaicking as well as in thresholding the blue mask for the unwrapping.

Interferometric coherence images are orthorectified and hence, can be used for land cover analyses in combination with optical satellite images. A relative presentation of the DEM reliability can be yielded with the aid of a variance image from the joined models.

RESULTS

After the data selection and the subsequent careful analysis of the available data, a relatively satisfactory elevation model was produced using two winter image pairs and two snow-free pairs (Fig. 3). The final model shows spatially variable variance and some holes on the western edge (c.f. Fig. 3), in location where there were lines missing in the original data. The image pair margins are visible in some parts of the mosaic (see Discussion and Fig. 5), indicating that the match between the individual models is not perfect. Based on mere GCP measurements, an individual model may still tilt about 1:1000, causing a threshold of a few tens of metres. Further adjustments are being carried out to flatten out these thresholds at the individual model margins, by adjusting the tilt of whole individual models. Fig. 4 shows a potential application of the DEM for interpretation of land cover types in combination with a TM image.

Fig. 3. The DEM mosaic of the study region at its current state. The bright area in the south is the Vatnajökull ice cap, which contains elevations well over 1000 metres. At the glacier margin, wet snow and sediment plus melt water streams give rise to poor coherence and hollows in the model. Lake basins such as Askja caldera (the semicircular spot in the middle of the lower half of the figure, surrounded by highlands) appear also as discontinuities. The Upper Pleistocene mountain arc to the right fringes the Volcanic rift zone and acts as a dam to the potential flood water bursting from the south (see also Fig. 4).

Fig. 4. A perspective view towards the northwest with Landsat TM image (432 RGB) draped over the InSAR DEM. Red colour denotes vegetated surface. Askja caldera with snowy rims is visible on the left. The ash (tephra) layer (in pale blue) originated in the 1875 eruption at Askja. Subsequent floods have disintegrated the tephra deposit in the east.

A black lava flow from the 1961 eruption sits on the northern rim of Askja. In the middle of the scene rises a table mountain Herðubreið (1682 m). Perspective images with shading are particularly sensitive in revealing any artefacts in the DEM. With the given illumination angle, the threshold between the two individual models appears as a dark linear band crossing the view from east to west (see text for further details).DISCUSSION

Image pair selection

Season and weather effects

Careful InSAR image selection turned out to be a crucial step in the DEM production. Based on earlier experience from northern Fennoscandia (Lapland) [8], it was assumed that the seasonal snow cover would have a stabilising effect on e.g. the vegetated areas and hence, data from wintertime would show good coherence. Therefore, baseline values and weather conditions during the wintertime ERS-1/2 tandem flights were assessed at first instance, and successive acquisition dates showing as little change as possible in the temperature and wind conditions were selected. A request was also made to ESA to obtain an access to the Interferometric Quick Looks (IQL's) [9] in order to help assessing the data quality before placing the order. There were, however, no IQL’s available for our candidate pairs. Only after the scenes had been received and processed, an offer was made to the authors about testing the IQL system with data stored at the UK-PAF. IQL processing is subject to start there in January 2001.

The quality of the wintertime images proved unsatisfactory and subsequently, severe problems were encountered with the interferometric coherence. Together with some pitfalls in the project funding, these caused delay in the analyses. The first part of the project was carried out in 1998–1999 with four SAR SLCI image pairs requested and received. Only two of the pairs qualified for further processing. The two other image pairs demonstrated too low coherence and were consequently not used (c.f. Table 1). Low coherence is believed to be due to the highly variable weather and snow conditions in the area. The absolute elevation rises from sea level to well over 1000 m in the south, creating a steep spatial gradient in both temperature and humidity. At high altitude, snow may be dry, whereas in the warmer coastal region the water content of the snow may be distinctively higher. High cyclone activity in North Atlantic induces variable climatic conditions in Iceland during winter and therefore, it is practically impossible to find suitable tandem acquisitions with stable weather conditions. Unfortunately, the image pair, which showed the best coherence for winter images, had some bad (missing) lines, resulting to a section of poor coherence in the final DEM (this problem was, however, recognised before placing the order, but could not be avoided due to the initial wintertime constraint). In the year 2000, two snow-free pairs (acquired in September) were ordered and processed, and the models were finally combined.

Based on the advice from ESA [10] and the experience described above, late summer images seem to offer a better option for InSAR in Iceland. This is controversial with the situation in Fennoscandia, where winter images typically show better coherence. It is hypothesised here that one should avoid highly variable weather conditions (e.g. Oceanic subarctic and subantarctic areas during the hemispheric winter) and instead try to find the most stable season for data acquisition. It is obvious that arid areas devoid of vegetation show a good coherence on snow-free pairs, but these images showed a good coherence also for the Icelandic shrubs. Holes in the models cannot be thoroughly avoided as with the global ERS tandem data, the coherence is only rarely good enough throughout the whole scene area for the production of a reliable elevation model. Some of the gaps are so large that interpolation is not feasible.

Number of image pairs employed

Another option for improving the quality of the final DEM may be an employment of a large number of image pairs from the same area. In order to identify and eliminate any atmospheric artefacts, it is recommended to use three to four overlapping pairs from both ascending and descending tracks, as this gives better results than using only one or two pairs.

Ground control points and mosaicking

The horizontal accuracy of the ERS-SAR data is initially about 100 m throughout the world when using the PRC orbit products. Therefore, in principle, a single elevation point will suffice the georeferencing, without a need for additional GCP's. However, by incorporating GCP’s from maps and other databases, the horizontal accuracy can be improved to 20-30 metres (although identifying and measuring representative points on a DEM can prove difficult). Atmospheric artefacts may cause difficulties, as a GCP measured at an artefact may result to a tilt in the model. Hydrological models are most demanding applications and therefore, additional relative points can be measured to smoothen the edges as much as possible. In Fig. 5, the image threshold effect is no more systematic and is therefore presumably due to atmospheric artefacts and varying coherence within an image, rather than a true mismatch in model rectification. Designing a procedure where the GCP’s could be measured directly on the initial images might prove beneficial.

Fig. 5. An example of the effect of atmospheric artefacts at a model border. The threshold between models runs diagonally across the scene from upper left to lower right, indicating an elevation difference of up to several tens of metres. Any systematic incompatibility has already been removed from the mosaic by adjusting the tilt of the models. Thus, the remaining (non-systematic) mismatch is presumably due to atmospheric artefacts and varying coherence rather than a rectification error.

InSAR-based elevation models have a good cost/benefit ratio compared to traditional photogrammetric methods, but one has to acknowledge the fact that the application cannot be employed readily in all regions. In addition to the land cover types known to suffer from limited C-band coherence (e.g. rain forests) and atmospheric artefacts (e.g. deserts), also areas in the Subarctic with steep climatic gradients and variable weather conditions may bring about difficulties in DEM construction.

The next step in the project at hand is fine-tuning the DEM by adding more GCP's and using more image pairs. Thereafter, the DEM will be employed in the modelling of catastrophic flood routes with the aid of GIS techniques. Of specific interest with regard to the modelling exercise will be the high momentum, variable viscosity (due to high sediment load) and the point-source (either single or multi) character of the floodwater bursting from the ice sheet margin. Therefore, a normal hydrological modelling, such as described in e.g. [11] will most probably prove insufficient.

CONCLUSIONS

Based on the case study reported here, the following conclusions can be drawn with regard to the interferometric DEM construction:

1.In the current study region, snow-free image pairs show better coherence than winter pairs, and are therefore

suggested for DEM construction even in moderately vegetated areas. In Boreal forest region, however, the coherence is generally better during the snow period.

2.Areas with steep climatic gradients and highly variable weather conditions (e.g. melting/re-freezing snow) degrade

the image coherence and prove to result in problems in the analyses of the data acquired in ERS tandem flights. 3.The use of only two overlapping image pairs and a limited number and distribution of GCP's failed to produce atotally threshold-free DEM mosaic. Increasing the number of overlapping pairs from 1-2 to 3-4 will improve the total DEM quality. Adjacent models can also be fit to one another, or to a reference DEM, where available.

4.An access to the IQL’s prior to placing a data order would be of great benefit as otherwise, the only way to assess

the applicability of a pair is to purchase and pre-process it. If the IQL’s are not available, it is advisable to order just

a single image pair from a particular track at first hand, to allow preliminary investigation of the coherence. The

availability of IQL's will improve in the near future, as the data from Northern Europe will be processed from January 2001 onwards at UK-PAF.

ACKNOWLEDGEMENTS

This study would not have been possible without the close collaboration of the whole research team: P. Alho (Turku), O. Arnalds (Reykjavík), J. Hendriks (Turku), M. Rossi (Turku), A. Russell (Keele), H. Seppä (Uppsala) and G. Wiggs (Sheffield). The authors thank T. Ikola (Novosat) for fruitful comments on the manuscript and E. Forsvik (Novosat) for invaluable assistance in image processing.

REFERENCES

[1] G. Larsen, M.T.Gudmundsson, and H Björnsson, "Eight centuries of periodic volcanism at the center of the Iceland hotspot revealed by glacier tephrostratigraphy, " Geology, 26, pp. 943-946, 1998.

[2] H. Björnsson, "Jökulhlaups in Iceland: prediction, characteristics and simulation, " Ann. Glaciol., 16, 95-106, 1992.

[3] M.T Guðmundsson, F. Sigmundsson, and H. Björnsson, "Ice-volcano interaction of the 1996 Gjálp subglacial eruption, Vatnajökull, Iceland, " Nature, 3, pp. 954-957, 1997.

[4] J.Käyhkö and P. Worsley, "Sediment distribution and transport processes on Holocene lava fields in north-eastern Iceland, " Supplementi di Geografia Fisica e Dinamica Quaternaria, Supplemento III, Tomo 1, p. 226 (Abstracts of the Fourth International Conference on Geomorphology, Bologna, Italia, 28.08.-03.09.1997).

[5] M.J. Rossi, "Morphology and mechanism of eruption of postglacial shield volcanoes in Iceland. " Bulletin of Volcanology, 57, pp. 530-540, 1996.

[6] M.J. Rossi, "Morphology of the 1984 open-channel lava flow at Krafla volcano, northern Iceland

[7] E.-A. Herland and A. Vuorela, "Operational DEM generation by means of SAR interferometry

[8] A. Vuorela, E.-A. Herland, and A. Saarikoski, "SarDEM – Production Process and Applicability of SarDEM Digital Elevation Models Based on Interferometric ERS SAR Images

[9] "INSI SAR Interferometry Sample Image Browser

[10] D. Massonnet, CNES, pers. communication, 20 January 1999

[11] A. Walker, Using the LANDMAP British Isles 1" IfSAR DEM for Hyrdographical Network Derivation. ERS -ENVISAT SYMPOSIUM, 16 - 20 October, 2000下载本文