FLAG UPNESS and its application for mapping seasonally wet to waterlogged soils.

Introduction

In this paper the UPNESS index derived in the FLAG model was tested against depth to watertable and groundwater EC to assess how well it represents the subsurface accumulation of water in the Wagga Wagga Catchment. The UPNESS index was then compared with soil landscape mapping in the Kyeamba Catchment, a larger catchment with complex groundwater systems, to assess how well it predicted areas of seasonally wet to waterlogged, saline, and sodic soils.

Wilson and Gallant (2000) reported from Barling et al. (1994) that subsurface flow regimes in a catchment rarely, if ever, reach steady-state. The UPNESS index used in this study is not a dynamic model but the environmental indicators of waterlogging such as vegetation changes do give a steady-state expression of where subsurface processes occur. Many surface water indices exist such as the Compound Topographic Index (CTI) of Beven and Kirkby (1979), and modifications to represent drainage networks better with different algorithms have been made by Mark (1998), and Tarboton et al. (1991). However, Moore et al. (1991) pointed out that source representation of flow is 1dimensional. Work by Costa-Cabral and Burges (1994) introduced a 2-dimensional flow path routing (DEMON) that better represented convergent and divergent flows. These models all imply subsurface flow through soil wetness, and are derived from algorithms of surface flow only. UPNESS, although a static representation of subsurface flow, uses an algorithm that attempts to represent potentiometric head. Conceptually this means that an infiltrated raindrop in the soil profile or shallow groundwater system is linked to every other connected particle of water and that piston flow of water particles occurs through the pressure response. This concept better represents the processes of subsurface flow through a groundwater head than inferring subsurface drainage from a surface catchment-derived wetness index.

Detailed resolution Digital Elevation Models (DEMs) are now available for extensive areas of Australia (Dowling et al. 1997; NSW LIC 1999). This provides new opportunities for hydrologic and pedologic studies over regional or whole catchment scales at resolutions previously available only for local-scale studies.

Criticisms of previous landscape and hydrological modelling studies have been that they are site-specific or too data and computationally intensive. Commonly, models are developed to predict the spatial variation of soils using statistical and process-based frameworks requiring detailed knowledge and datasets of soil topo-sequences, hydrologic processes, and soil physical properties that are relevant to pedogenesis (Gessler et al. 1995). Although the concepts and applications of such models can be useful, the specific data requirements needed to parameterise such models often leave them inaccessible to many users because the data gaps of many parameters lead to generalisation and best guessing. A model that uses only one readily available dataset is the Fuzzy Landscape Analysis GIS (FLAG) model (Laffan 1996; Roberts et al. 1997; Dowling 2000). FLAG was developed to predict dryland groundwater discharge and salinisation in catchments using topographic elevation data alone, incorporating knowledge from other modelling but without separately addressing those processes. FLAG is based on the assumption that many of the factors that lead to waterlogging and salinisation are inter-correlated with landscape position. These include rainfall, vegetation, soils, geology, and geomorphology. This model is a simplification, or integration, of complex pedogenic processes, but in most cases this level of information is adequate to provide insights into water movement within the landscape, and provide a data source for models such as that of Gessler et al. (1995).

Methods

Site descriptions

This study focusses on two catchments in the Murrumbidgee region on the south-western slopes of New South Wales, Australia (Fig. 1). The first, the Wagga Wagga Catchment (33 [km.sup.2]), is prone to urban salinisation, and due to the effects on infrastructure, a dense groundwater-monitoring network has been established across the whole catchment. The catchment geology consists of Ordovician meta-sediments interbedded with a complex of siltstone, shale, and phyllite. The landscape consists of gently undulating to steep hillslopes with some alluvial flats, and an average annual winter-dominant rainfall of approximately 530 mm per year. Land clearing has occurred across 90% of the catchment.

The second catchment used in this study is the Kyeamba Catchment situated 10 km south-west of Wagga Wagga (Fig. 1), covering an area of 602 [km.sup.2]. The dominant geology of the Kyeamba Valley is Ordovician meta-sediments interbedded with a complex of siltstone, shale, phyllite, minor schist, and quartzite with minor intrusions of granite. The landscape of this catchment also consists of gently undulating to steep hillslopes with some alluvial flats, and an average annual winter-dominated rainfall of approximately 660 mm per year. Land clearing has occurred across 95% of the catchment for uses including mixed farming of sheep, cattle, and cereal crops.

The model

The FLAG model derives several landscape position indices from digital elevation data only. Local lowness (LOWNESS) is calculated by smoothing the DEM to calculate the average local elevation, and then calculating the fuzzy set difference between the smoothed and unsmoothed elevations. Plan curvature (CONCAVITY) gives an estimate of landscape concavity. Contributing area (UPNESS) is normally used to derive more complex indices of landscape position (or wetness) through the combination of LOWNESS and UPNESS or CONCAVITY and UPNESS. Only the UPNESS index is considered in this work as one of the aims is to evaluate how well this index represents subsurface processes. Full descriptions of the model can be found in Laffan (1996), Roberts et al. (1997), and Dowling (2000).

UPNESS is an index of surface and subsurface water accumulation inferred from relative height in a landscape. The accumulation at any given point in the UPNESS index is given by the set of points that are connected by a continuous monotonic uphill path. It is assumed that, for subsurface flow, all points that are connected in this way could exert some hydrologic effect on the locations below. For this index, any number of topographic catchment boundaries can be crossed provided the adjacent uphill cells in the next catchment are higher (Dowling 2000; Laffan 2002). Figure 2 illustrates how UPNESS is calculated from a DEM. The values in the cells represent elevations. For cell A1, the UPNESS count is 11, which is derived from the cells marked with a black dot. Cell C4 represents a hilltop with an elevation of 5. Cells B5, C5, D5, and ES, which are effectively behind the hill with respect to AI, are included as contributing to cell A1 because they are connected by the monotonic uphill path through cell A4. Therefore, although cells B5, C5, D5, and ES are outside the topographic or surface water catchment boundary, they have subsurface potential to affect cell A1 hydrologically. This computational method conceptually makes the UPNESS index a representation of a subsurface groundwater head.

The 25-m resolution DEM used in this study was supplied by the NSW Land Information Centre (NSW LIC 1999). Field verification confirmed that the sinks in the DEM were mostly artefacts, occurring primarily in streamlines. For this study the DEM was filled to remove major sinks and reconnect drainage networks using ArcInfo GRID version 7.2.1. (ESRI 1995).

The UPNESS program, written in Fortran 77, was run using an UPNESS threshold of 0.0, specifying that any neighbouring cell (including diagonal cells) with a height difference greater than or equal to zero will be included in the UPNESS count.

The resulting UPNESS grid is linearly normalised between 0 (hilltops) having the least accumulation and 1 (lowest valleys) the most.

Field checking and validation

Groundwater bores

FLAG results for the Wagga Wagga catchment were compared with all groundwater bores in the study area that had complete records on the 26 July 2002, were not dry, and were in the same fractured rock groundwater system of Ordovician recta-sediments. The data were collected by the Wagga Wagga City Council using manual dip measurements as part of their urban salinity monitoring program. Positional errors in bore coordinates where often hundreds of metres. To compensate for this, an average of the UPNESS cells within a 150 by 150 m (6 x 6) window of each bore was compared with the depth to watertable and groundwater EC. UPNESS can have very different values at breaks of slope, and positional accuracy can bias or degrade values taken for bores with inaccurate locations. A 150-m window was large enough to average values of bores at locations with contrasting UPNESS but small enough not to substantially affect the UPNESS values for the bores located away from breaks of slope. Preparation of the groundwater data for comparisons to the CTI index used the same methods as described above. The CTI index (Beven and Kirkby 1979) was calculated using the D[infinity] method, multiple flow path algorithm of Tarboton (1997).

Soil landscapes

FLAG results for the Kyeamba Catchment were compared with 'Soil landscapes of the Wagga Wagga 1 : 100000 map sheet' (Chen and McKane 1996) and field mapping for the 'Soil landscapes of the Tarcutta 1:100000 map sheet" (J. A. Wild, unpublished data).

Soil landscapes are areas of land that 'have recognisable and specifiable topographies and soils' (Northcote 1978) and are justified as a mapping tool because similar causal factors are involved in the formation of both landscapes and soils. In addition, both soil and landscape limitations constrain rural and urban land use, making soil landscape mapping a practical tool for regional and catchment planning. The method of soil landscape mapping included consideration of dominant geomorphic processes, geological parent material, topography, climate, vegetation, and soil type, using various aids such as aerial photo interpretation and satellite imagery. Provisional soil landscape boundaries were then checked by field investigation plus detailed physical and chemical laboratory soil testing. Methods of fieldwork followed McDonald et al. (1990). Soil landscape boundaries in the Kyeamba area were delineated at 1:25000 scale although final mapping boundaries are at I : 100000. This placed a practical limit on the precision of soil landscape mapping at the chosen scale and provided an opportunity to efficiently refine soil landscape boundaries with models such as FLAG.

Areas less than 40 ha were generally not mapped unless considered to be locally significant. The UPNESS index was converted to a discrete map of seasonally wet to waterlogged, saline, and sodic soils. This was achieved by visually determining an UPNESS cut-off threshold by comparing the classified UPNESS to the appropriated classes of the soil landscape map. Summerell et al. (2003) has shown an objective method to determine this cut-off, which will replace the subjective method used for this study. Values above the threshold value of 0.01551 were used to predict soil landscape boundaries of interest to compare with mapped soil landscape boundaries.

Model comparisons

Linear relationships

Freeze and Cherry (1979) summarised groundwater head as being equal to elevation, pressure, and velocity heads. They determined that velocity head is insignificant and that in the saturated zone, pressure head is equal to 0, leaving elevation head as the dominant factor determining groundwater head. The groundwater surface is generally conceptualised as analogous to a smoothed ground surface, making a linear relationship between depth to watertable and UPNESS the simplest model form consistent with process theory and justifying linear regressions for these analyses. The linear relationship between UPNESS and groundwater EC for a single groundwater system was supported by the concept that as accumulating groundwater flow down slope increases, secondary weathering and concentrations of rock and soil minerals will occur, thereby increasing the concentration of salts.

Linkage between shallow watertables and EC, to soil pedogenesis

Fritsch and Fitzpatrick (1994) describe waterlogging, soil salinisation, and soil sodicity as forming in the same locations. The difference between saline or sodic matrices is determined by seasonal hydrological changes caused by the interception of saline watertables, as groundwater rises and fails. They describe the change from soil salinisation to soil sodicity as being a process caused by the lowering of saline watertables allowing fresh water to leach salt away, leaving the clay exchange complex partially saturated with sodium. An overall catchment hydrological change, such as afforestation or even a saline area eroding and allowing drainage of saline water, can drive this process. For this reason, direct spatial comparisons between UPNESS and soil landscape mapping were undertaken.

Results and discussion

Relationships between FLAG UPNESS and bore data Depth to watertable

Initially, all available bores including those located in the alluvial landforms of the Murrumbidgee River flood plain were included to test the relationship between depth to watertable and UPNESS and also CTI. However, due to a weak relationship, alluvial bores were excluded since they were part of a different groundwater system connected to the Murrumbidgee River where preferential flow paths govern groundwater movements. Bores used for analysis were selected from erosional and depositional landscapes where water movement is dominated by hill-slope processes (Fig. 3a, b). Four further exclusions were made where localised conditions influenced processes more dominantly than the hill-slope hydrology. For example, a large groundwater drain that lowers the groundwater level locally, influences bores 11 and 42 and without this drain, higher levels are expected. Bore 17 also had lower water levels than expected and is situated on the boundary of the alluvium. The watertable was within 4 m of the ground surface, and enhanced drainage by the alluvial contact was suspected. Bore 7 had higher water levels than expected but the recharge area around the bore is affected by an open cut gravel quarry that is likely to enhance infiltration locally. Figure 3a and b shows the UPNESS and CTI indices, respectively, and illustrates how the UPNESS index appears very smoothed and independent of the detailed surface flow features as expressed by the CTI index. A linear regression with [r.sup.2] = 0.66 (Fig. 4a) for UPNESS and depth to watertable, and [r.sup.2] = 0.10 between CTI and depth to watertable were found. By optimising the sample window the CTI regression could only be improved to [r.sup.2] = 0.25 (Fig. 4b) (normalising of the CTI index resulted in substantially worse regressions). Given that little improvement was achieved, no further analysis was conducted on the CTI index as UPNESS was demonstrated to give a better representation of groundwater heads in this catchment. exclusion of a further 4 bores. Bores 13 and 47, a nested pair, were fresher than expected. Bore 13 frequently dried up and the deeper bore 47 exhibited large fluctuations in EC (0.9-11 dS/m). This indicates that they are located in a shallow groundwater system with fast response times and small storage capacity. These bores may be frequently flushed, causing the fresher than expected EC values. This also indicates that these bores are not part of the main hill-slope driven groundwater system containing the majority of bores. Similar processes affecting bores 13 and 47 may affect bore 38; however, there was no apparent reason why bore 15 was more saline then expected. Bore 15 was an extreme outlier and was interpreted as a data error. Exclusion of these bores (not related to the main hill-slope groundwater system), also resulted in an improved fit with depth to watertable of [r.sup.2] = 0.71 (Fig. 6).

Comparison of FLAG UPNESS and mapped soil Landscapes

A large UPNESS value (Fig. 7b) represents an area that has high subsurface accumulation and tends to be poorly drained due to low gradient. Such conditions encourage the formation of poorly drained soil types from hydrologic processes described by Fritsch and Fitzpatrick (1994). Figure 7a shows the O'Briens Ck soil landscape unit described by Chen and McKane (1996). Boundaries derived from the UPNESS index (Fig. 7b) generally agree with the soil landscape boundaries (Fig. 7a).

After classifying UPNESS to the cut off value of 0.01551, 89% of the catchment was predicted correctly as both the seasonally wet to waterlogged saline and sodic soils (O'Briens Ck soil landscape unit) and the dry soils. However, the dry soils make up the majority of the area within the catchment and are not of interest for this study. Of the seasonally wet to waterlogged saline and sodic soils, which consist of ~5% of the catchments, 36% were correctly predicted. Differences mainly occurred near streams where the landscape mapping showed larger areas of alluvial flat extending away from the creeks, whereas the UPNESS index was more confined to the streams in these locations. However, in broad, unchannelled or poorly drained flats the UPNESS index matched the soil landscape mapping much more closely and with more detail.

During the process of edge matching soil landscape maps and field checking for the Tarcutta soil landscapes sheet, a discrete soil landscape unit of seasonally waterlogged soils was identified (provisionally named 'Big Springs variant a', or 'bsa', Fig. 8a). The upper boundary of the bsa unit was sharply delineated by the boundary of a surface spring that followed the contour. Positioned between an upper and a lower colluvial/alluvial plain at the break of slope, the spring did not appear to be related to any surface drainage feature such as a streamline. The lower boundary comprised an alluvial plain, the draft 'Gregadoo' unit or 'gr', that had a restricted outlet caused by converging foot slopes (Fig. 8a). Overall, the bsa unit had poorer drainage and more pronounced waterlogging than expected.

The UPNESS index matches the general shape and area of the bsa unit, with 69% of the mapped landscapes predicted correctly. The UPNESS index provides additional detail over the 1:100000 soil landscape mapping, with 3 drier areas depicted within the confines of the predicted waterlogged soils. These 3 areas were marginally higher and consequently better drained than the surrounding area. The difference in elevation was sufficient for cattle yards, silos, and homesteads to be preferentially located on them.

The UPNESS index also differed at the outlet of the bsa unit. In the area marked by transect ([alpha]) in Fig. 8b the UPNESS index depicts the main subsurface accumulation to be more confined to the drainage line, whereas the mapped soil landscapes extend broadly away from the drainage lines (transect ([alpha]) Fig. 8a). This difference is attributed to the methods as the soil landscape mapping integrates interpreted water movement via topography, point samples of soil profile properties such as permeability, and impact of drainage on the rootzone of present day vegetation. In broad valleys, such as upstream at transect ([beta]) (Fig. 8a, b), both soil landscape mapping and UPNESS give similar results since the broad valley is not channelled and thus does not constrain the UPNESS calculations. A deep channel effectively prevents inclusion of parts of one or both sides of the valley even though they are equal in height, as the algorithm prevents paths from crossing the channel due to the continuous monotonic uphill path rule. The channel is lower and therefore always accumulates from both sides.

The realism of the UPNESS algorithm is debatable, as there are some questions relating to how effectively a deeply incised channel drains either side of the valley. It may be more realistic than soil landscape mapping that ignores the existence of the channel in affecting the unit, although this argument is based on a temporal issue. If channel incision occurred in the past it may influence pedogenesis through altering soil hydrology. If incision is very recent then perhaps the approach that does not take subsurface drainage into account may provide a better prediction of current soil conditions. The soil pedogenics will mostly be influenced by hydrologic changes causing sodicity and salinity.

Conclusions

The FLAG UPNESS index used in this study was validated within the Wagga Wagga Catchment (~33 [km.sup.2]) for its ability to predict subsurface water accumulation. Reasonable linear correlations were established between groundwater depth and EC. Once this validation was achieved the model was applied across the Kyeamba Catchment (~602 [km.sup.2]) to see how well it could predict seasonally wet to waterlogged saline and sodic soils. The relative extent of these soils within the catchment was well represented by the model. However, in alluvial or in-filled valley systems, where deep stream incision has occurred, UPNESS predicted a smaller area of the soils than is mapped. In non-incised areas the modelled result showed similar areas with more detail than mapped soil landscapes.

This paper explores only one component of the FLAG model, which contains other indices that may also be useful. Moore et al. (1991) signalled that there is a demand from agencies for simple modelling techniques to assist with day-to-day resource knowledge gathering. The UPNESS index of FLAG has been shown to be useful for this purpose, especially in areas where landscape mapping does not already exist (Murphy et al. 2003). One of the main areas where this modelling can aid soil-landscape mapping is in helping to define more objectively the initial broad soil landscape mapping boundaries.

As conventional soil landscape map boundaries are defined subjectively, validation techniques using a set of field sampling sites in a catchment would provide better comparisons than were used in this study. The field sites could be used to verify how many sites soil landscape mapping and UPNESS methods predict correctly. This is an area identified for further investigations.

Acknowledgments

The authors thank Michial Sutherland from Wagga Wagga City Council for providing the groundwater bore information for the Wagga Wagga Catchment and Mark Mitchell from the Department of Infrastructure, Planning and Natural Resources for discussions on the groundwater processes acting within this catchment. Guy Geeves from the Department of Infrastructure, Planning and Natural Resources, Centre for Natural Resources, provided valuable critical comment on this paper in its initial review stages. Dr Narendra Tuteja and Peter Barker provided continued support for this work.

References

Barling RD, Moore ID, Grayson RB (1994) A quasi-dynamic wetness index for characterizing the spatial distribution of zones of subsurface saturation and soil water content. Water Resources Research 30, 1029-1044.

Beven K J, Kirkby MJ (1979) A physically-based variable contributing area model of basin hydrology. Hydrological Sciences Bulletin 24, 43-69.

Chen XY, McKane DJ (1996) Department of Land and Water Conservation, New South Wales, Australia, Soil Landscapes of the Wagga Wagga 1:100000 Sheet.

Costa-Cabral MC, Burges SJ (1994) Digital elevation model networks (DEMON): A model of flow over hillslopes for computation of contributing and dispersal areas. Water Resources Research 30, 1681-1692.

Dowling TI (2000) FLAG Analysis of catchments in the Wellington region of NSW. CS1RO, Land and Water Consulting Report 12/00, Feb. 2000, Canberra. http://www.clw.csiro.au/publications/ consultancy/

Dowling TI, Roberts DW, Walker J (1997) Predicting salinity risk without the use of a process model. Murray--Darling 1997 Workshop, Groundwater in the balance, Toowoomba, Qld, 26-28 August 1997.

ESRI (1995) 'ARC/INFO User's guide. Version7.0.' (Environmental Systems Research Institute Inc.: California)

Freeze RA, Cherry JA (1979) 'Groundwater'. (Prentice Hall: Englewood Cliffs, NJ)

Fritsch E, Fitzpatrick RW (1994) Interpretation of soil features produced by ancient and modern processes in degraded landscapes I. A new method for constructing conceptual soil-water-landscape models. Australian Journal of Soil Research 32, 889-907.

Gessler PE, Moore ID, McKenzie NJ, Ryan PJ (1995) Soil-landscape modelling and the spatial prediction of soil attributes. International Journal of Geographical Information Systems 9, 421-432.

Laffan SW (1996) Rapid appraisal of groundwater discharge using fuzzy logic and topography. In 'Proceedings of the 3rd International Conference/Workshop on Integrating GIS and Environmental Modelling'. Santa Fe. (National Center for Geographic Information and Analysis)

Laffan SW (2002) Using process models to improve spatial analysis. International Journal of Geographical Information Science 16, 245-257. doi: 10.1080/13658810110099107

Mark DM (1998) Network models in geomorphology, in 'Modelling geomorphological systems'. (Ed. MG Anderson) pp. 73-97. (John Wiley: New York)

McDonald RC, Isbell RE Speight JG, Walker J, Hopkins MS (1990) 'Australian soil and land survey field handbook.' (Inkata Press: Melbourne, Sydney)

Moore ID, Gmyson RB, Ladson A R (1991) Digital terrain modelling: a review of hydrological, geomorphological and biological applications. Hydrological Processes 5, 3-30.

Murphy B, Geeves G, Miller M, Summerell G, Southwell R Rankin M (2003) The application of pedotransfer functions with existing soil maps to predict soil hydraulic properties for catchment-scale hydrologic and salinity modelling. In 'MODSIM'. (Ed. DA Post) pp. 502-507. Townsville, Qld. (Modelling and Simulation of Australia and New Zealand Inc.)

Northcote KH (1978) Soils and land use. In "Atlas of Australian resources'. (Division of National Mapping: Canberra, ACT)

NSW LIC (1999) Statewide digital elevation model data. NSW Land Information Centre, Bathurst, NSW, Australia.

Roberts DW, Dowling TI, Walker J (1997) FLAG: a fuzzy landscape analysis GIS method for dryland salinity assessment. CSIRO, Land and Water Technical Report 8/97, Canberra. http:// www.clw.csiro.au/publications/technical/technical97.html

Summerell GK, Vaze J, Tuteja NK, Grayson RB, Dowling TI (2003) Development of an objective terrain analysis based method for delineating the major landforms of catchments. In 'MODSIM'. (Ed. DA Post) pp. 496 501. Townsville, Qld. (Modelling and Simulation of Australia and New Zealand Inc.)

Tarboton DG (1997) A new method for the determination of flow directions and upslope areas in grid digital elevation models. Water Resources Research 33,309 319.

Tarboton DG, Bras RL, Rodriguez-Iturbe I (1991) On the extraction of channel networks from digital elevation data. Hydrological Processes 5, 81-100.

Wilson JP, Gallant JC (2000) Secondary topographic attributes. In 'Terrain analysis: principles and applications'. (Eds JP Wilson, JC Gallant) pp. 87 131. (John Wiley and Sons: New York)

Manuscript received 24 February 2003, accepted 15 December 2003

G. K. Summerrel (A,B,C,E), T, I. Dowling (C,D), J. A. Wild (A), and G. Bale (A)

(A) Department of Infrastructure, Planning and Natural Resources, Centre for Natural Resources, PO Box 5336, Wagga Wagga, NSW 2650, Australia.

(B) Dept of Civil and Environmental Engineering, University of Melbourne, Parkville, Vic 3052, Australia.

(C) CRC Catchment Hydrology.

(D) CSIRO Land and Water, PO Box 1666, Canberra, ACT 2601, Australia.

(E) Corresponding author; email: [e-mail]