Testing a soil-landscape model for dry greywacke steeplands on three mountain ranges in the South Island, New Zealand.
Soil-landscape modelling in steeplands
A predictive soil-landscape model for steeplands with uniform geology and simple and repeating landforms has been developed by McIntosh et al. (2000). In such terrain they considered elevation and aspect, through their influence on microclimate, to be the main influences on soil variation. The model was developed on the Benmore Range, South Canterbury, New Zealand. It predicted soil properties on major landscape units identifiable at the 1 : 50 000 scale. These are the management units for which land-use decisions are made on this type of rangeland.
The aim of this study was to test the accuracy of this model and subsequently developed GIS-interpolation techniques for predicting soil properties at randomly selected sites in 3 areas having similar landforms and geology: (1) on the Benmore Range, (2) on parts of the adjacent Kirkliston Range, and (3) parts of a similar mountain range on Molesworth Station, 250 km distant in the South Island, New Zealand.
A number of authors, e.g. Skidmore et al. (1991, 1996), Moore et al. (1993), and McKenzie and Ryan (1999), have used the relationships of geology, landform, vegetation, and climate characteristics to soils, determined in typical areas, as `indicators' of soil map units and/or soil properties present in broadly similar landscapes elsewhere with varying success. Between 40 and 80% of the sample variance of 3 soil properties was explained by variation in landform, climate, and recently available airborne gamma radiometric data for their model projection area by McKenzie and Ryan (1999). McIntosh et al. (2000) postulated that in the dry greywacke steeplands, the prediction of soil distribution at the required scale (1 : 50 000) from traditional indicators was not useful in this geologically uniform area because the dominant ridge and valley landforms provide little indication of the soils on them; and the tussock, Hieracium, and scrub vegetation pattern (Newsome 1987) is largely the result of burning and grazing, and does not reflect soil patterns.
The results of traditional soil survey and mapping in a particular area are difficult to extrapolate to another because individual pedologists may relate soil properties to landscape features (e.g. aspect and elevation) in different ways. Hudson (1992) promoted the publication of soil-landscape models, noting the limited value of soil map units for information transfer. In these steeplands, individual soil properties can have a large variance, making soil map units unsuitable for predicting single soil-factor values. Soil trends tend to be gradual (McIntosh et al. 1981; McIntosh and Hunter 1997; McIntosh et al. 2000); therefore, a mathematical consideration of trends of individual soil properties is likely to be more useful for prediction than an artificial delineation of map units.
The drier greywacke soils of the South Island mountain ranges with <700 mm mean annual rainfall cover some 470 000 ha (National Water and Soil Conservation Organisation 1975-1979; Lynn 1996) (Fig. 1). These rangelands support fragile ecosystems prone to degradation by fire, overgrazing, pests (e.g. rabbits), nutrient decline (O'Connor and Harris 1991; McIntosh 1997), and weed invasion (Hunter 1991). Current resource data available for this terrain are a national spatial soil layer (i.e. a named soil type) incorporated in the New Zealand Land Resource Inventory (NZLRI) (National Water and Soil Conservation Organisation 1975-1979), linked to the national soil database (McDonald et al. 1988) from which `typical' soil characteristics can be derived. These sources provide insufficient detail for on-site farm management or national resource studies (McIntosh and Hunter 1997; Tate et al. 1997; McIntosh et al. 1998, 2000). An understanding of the pattern of soil nutrients and properties throughout this landscape would help predict environmental risks and trends, assist farm management decisions, and identify new land-use opportunities. Variation in leaching, weathering, and organic matter stability in relation to aspect and elevation is implied by several soil studies (Cuff 1973; Tonkin 1984; McIntosh et al. 1981, 1998, 2000). Soil moisture limits both leaching and weathering at low elevations on all aspects. At middle elevations soil moisture still limits leaching and weathering on sunny (N and W) slopes exposed to the desiccating north-west winds that characterise this environment. On the leeward shady aspect (S and E) slopes at this elevation, higher soil moisture levels for longer periods promote more weathering and leaching. At higher elevations the lower soil temperatures limit weathering, and soil leaching conditions are similar on all aspects. However, at these altitudes erosion becomes an important influence on soil properties.
Materials and methods
Description of the soil-landscape model
The soil-landscape model for dry steeplands (McIntosh et al. 2000) comprises the mathematical interpolation of a trend table using as input a digital elevation model (DEM). The trend table was established from a combination of pedological judgement and data from a structured stratified sampling program designed to `capture' the variation in soil properties as driven by elevation, aspect, and land system. Trend tables were developed for 0-7.5 cm %C, pH([H.sub.2]O), and P retention, and A-horizon total C, total N, exchangeable Ca, and A-horizon thickness. Statistical tests showed that aspect and elevation were poor predictors for exchangeable Mg and Na, stones in the A horizon, and profile depth (McIntosh et al. 2000); therefore, trend tables were not developed for these properties. Pedological judgement was used to smooth elevation trends, and to identify boundaries where properties changed sharply (Lilburne et al. 1998). A `fuzzy' boundary that spanned 20 [degrees] (i.e. 60-80 [degrees] and 240-260 [degrees]) was used to distinguish between sunny and shady aspects.
Implementation in a GIS
The distribution of dry greywacke steepland soils for which the predictive model is potentially applicable was extracted from the 1st and 2nd editions of NZLRI (National Water and Soil Conservation Organisation 1975-1979; Lynn 1996) (Fig. 1). The areas are concentrated in the Waitaki District, North Canterbury, inland and northern Marlborough.
On the basis of rainfall and landscape character (assessed from aerial photographs and field observation), the regions were assigned to 1 of the 3 steepland land systems (Glenbrook, Glencairn, Ben Omar) as defined in McIntosh et al. (2000). Elevation and aspect layers were derived from 20 m contour data supplied by Land Information New Zealand. Each of the data layers was clipped to cover the dry steepland area of interest.
Lilburne et al. (1998) describe how triangular irregular networks (TINs) were used to interpolate the trend tables in topographic space (i.e. elevation and aspect). A TIN was generated for each land system. Maps covering all of the dry steeplands were produced for each of the 7 soil properties.
Testing spatial predictions against new data
Additional data from 3 areas were collected to test the accuracy of the spatial predictions. The location, environmental characteristics, and sampling pattern followed in the 3 test areas are summarised in Table 1.
Testing spatial predictions against NZLRI data
Predictions of the McIntosh et al. (2000) model were also tested against values predicted using NZLRI data. Lilburne et al. (1999) evaluated a number of models of A horizon %C over the Benmore Range. The simplest LRI model involved the identification of units mapped on the range, their soil types, slope, and erosion severity descriptions. The mean A horizon %C value from the fundamental data layers of the regional soil legends (Barringer et al. 1998) was modified for the impact of percentage of bare ground and erosion as follows. Up to 20% bare ground was taken as `normal' in this environment and the regional legend A horizon %C value was adopted for those units with erosion severity of 2 (<20% bare ground) or less. For those units with erosion severity 3 (21-40% bare ground), 4 (41-60% bare ground), or 5 (>61% bare ground), 70, 50, and 20% of the adopted value was taken, respectively. Where bare rock was recorded in the soil field, the erosion severity was increased by 1 severity unit. Fig. 1a shows the distribution of A horizon %C using the LRI model.
Sampling pattern and procedure
On the Benmore Range, profiles were sampled by horizon at 25 randomly selected sites on a range of aspects at elevations between 476 and 1525 m. Sites were randomly selected using a GIS function but some were rejected due to difficult access. An additional 16 sites were sampled using the structured stratified sampling pattern (McIntosh et al. 2000), but only on I slope position, using a reduced set of land system-aspect-elevation combinations.
On the Kirkliston Range, we replicated the original structured stratified sampling pattern but again sampled only 1 slope position. Profiles were sampled by horizon on upper to middle backslopes on 4 aspects approximating magnetic N, S, E, and W, at low, medium, and high elevations (c. 650 m, 950 m, and 1260 m), on 2 major spurs. Only medium and high elevation sites were sampled at 3 locations on Molesworth as the valley floors are above 700 m. The positions of all sites were fixed by GPS.
Sampling was undertaken in spring when soils were moist. Profiles were described and all horizons sampled. Twenty bulked 0-7.5 cm samples were also taken at each site, using a 2.54-cm-diameter soil corer.
Analytical procedures and methods
Weight per unit volume of the <2 mm fraction of the A horizon was calculated by oven-drying a sample of known volume at 110 [degrees] C, and estimating stone content in the field. The soil properties measured were A-horizon pH([H.sub.2]O), total C, total N, P retention, exchangeable Ca, Mg, K, Na, A horizon thickness, stone content, and field moisture content. Samples from 0-7.5 cm were analysed for the same chemical soil properties and also for pH(Ca[Cl.sub.2]). Subsoil horizons were analysed for P retention and pH([H.sub.2]O). Soil analyses were by the methods of Blakemore et al. (1987): pH([H.sub.2]O) (method 2A), total C (Leco FP2000), total N (Leco FP2000), P retention (method 5G), exchangeable Ca, Mg, K, Na (method 6A1), and pH(Ca[Cl.sub.2]) (method 2B).
Analysis of model performance
Predictions based on the McIntosh et al. (2000) model were tested against actual observations at the test sites on the 3 ranges. There are a number of measures of the differences between measured data and predicted data. Root mean square error (RMSE) is a conservative measure of error that is often used, although it is sensitive to outliers. Mean absolute error (MAE) indicates the magnitude of the average error. Mean error (ME) estimates the bias, i.e. underestimating or overestimating. Graphs of measured versus predicted, and box plots or histograms of the differences, are also useful for detecting spatial or environmental patterns in model performance (Willmott 1984). All these methods were used. The scale problem of the support of the sampled sites being different from the support of the DEM (100 m) was avoided by using values for aspect and elevation collected in the field (by compass and GPS). This also avoided DEM error affecting predictions (Lilburne et al. 1999).
Comparison of predicted with sampled values
Graphs of measured versus predicted values for 0-7.5 cm %C, pH, and P retention, and A-horizon total C, total N, exchangeable Ca, and A-horizon thickness were plotted. At the 3 test sites the model was most accurate at predicting 0-7.5 cm pH, and least accurate at predicting A-horizon exchangeable Ca (Fig. 2). Graphs of measured versus predicted values also identified the presence of outliers. Outlier sites were not unexpected although every effort was made to sample only stable planar slopes when the stratified sampling pattern was followed. Such criteria could not be followed for random sites. Persistent outliers were present in both the stratified and random samples.
In box plots for the 3 test locations (Fig. 3) the y-axis represents the errors or residuals of the predictions--a positive error indicates underprediction by the model. Each box plot shows the bias, and the magnitude and variability of the error. The model performed best on the Benmore Range but did not consistently perform better at either of the other test locations, and no consistent trends were evident. Prediction error measures (ME and MAE) for the modelled attributes are shown in Table 2.
The model gave a slight overestimation of the 0-7.5 cm %C on the Benmore Range, and an underestimation on the Kirkliston Range and on Molesworth Station, where the biggest error was evident. The mean error of the 0-7.5 cm pH had a neutral bias on the Kirkliston Range but was slightly overestimated on Molesworth Station and underestimated on the Benmore Range. On the Benmore Range, 0-7.5 cm P retention was most accurately predicted even though large outliers were evident in the sample. On both Molesworth Station and the Kirkliston Range 0-7.5 cm P retention was overestimated. A-horizon C values were overestimated on the Kirkliston Range, and underestimated on the Benmore Range, and even more so on Molesworth Station, as was A-horizon N. A-horizon exchangeable Ca was underestimated on both Molesworth Station and the Benmore Range, with extreme outliers evident on the Kirkliston Range, indicating that variation in parent material Ca composition was poorly modelled. A-horizon thickness was slightly underestimated on Molesworth Station and the Benmore Range, and overestimated on the Kirkliston Range.
The mean absolute error (Table 2) for 0-7.5 cm %C, pH, A-horizon total C, and total N increased with distance from the model establishment area, i.e. MAE for Benmore Range < Kirkliston Range < Molesworth Station. For 0-7.5 cm, P retention and A-horizon thickness MAE is similar for both Molesworth Station and the Benmore Range, and greater on the Kirkliston Range.
No consistent patterns were evident in plots of RMSE with respect to aspect or elevation, indicating that there are no systematic errors in the model at the 3 test sites due to aspect and/or elevation. The efficiency of a model is analogous to the coefficient of determination derived in a linear regression analysis (i.e. [R.sup.2]) (Nash and Sutcliffe 1970). Calculating model efficiency indicates that up to 45% of the variability of 0-7.5 cm pH is explained by the model, with less for the other attributes.
Comparison of A horizon %C distribution from existing LRI and modelled data
Figure 1 (a and b) shows the distribution of the A-horizon %C generated from the LRI model and the interpolation of the trend table and the DEM for the Benmore Range. The RMSE of the 2 maps at 38 test sites and the percentage correctly classified into 3 categories (0-2, 2-4, 4+) of %C were 1.6, 44.7%, and 1.09, 50.0% for the LRI and trend table DEM models, respectively. Thus the model is useful for identifying land with environmental risk associated with low soil organic matter.
Although model efficiency is not high, it is comparable to that of McKenzie and Ryan (1999) which includes airborne gamma radiometric data and more terrain attributes. While the model is based on 3 input factors only (aspect, elevation, and land system), a number of other factors are known to influence variability of the soil properties of interest.
Geomorphic activity. The high tempo of geomorphic activity in this environment is evidenced by the presence of steep slopes, high uplift rates, the predominance of coarse colluvium, abundant visual evidence of surficial erosion and deposition, and the presence of buried soils and truncated profiles. Up to 40% non-vegetated bare ground and broken surficial rock or stone pavement and the presence of thin veneers of freshly deposited fine slope-wash material are common on many slopes. Over-thickened A horizons are also a feature, especially at high and middle elevations on E and S aspects. This source of variability could not be included in the model. Satellite imagery was investigated as a potential source of spatial input but it was not of sufficient detail to discriminate bare rock, scree, and eroded slopes in this steep terrain.
Parent material composition. Although the predominantly greywacke sandstone and argillite rocks of the `greywacke steeplands' are generally regarded as uniform, their composition does vary (Dickinson 1971), but remains largely unmapped. Rare conglomerate and limestone bands, thin beds of basic tuff, and small lenses of black limestone are present (Lensen 1962; Mutch 1963; Gair 1967). Differential loess fallout and retention on slopes is another source of variation that was not included in the model.
Management history and vegetation change. The management history of these rangelands over the last 160 years is complex, with major vegetation changes induced by fire and grazing by both domestic stock and pests (O'Connor and Harris 1991; Mark 1992, 1994; Hughes et al. 1995). Overgrazing, plagues of rabbits, and erosion resulted in the properties now amalgamated into Molesworth Station being abandoned and reverting to the Crown in the 1930s and 1940s (McCaskill 1969). Although rabbit numbers have been largely under control since the mid-1950s, and sheep have been replaced by cattle, vegetation recovery has been minimal (Moore 1976), and scrub and weed invasion significant (Stevens and Hughes 1973; Hunter 1991). No comprehensive data exist, at an appropriate scale, of past fire or grazing history for the majority of blocks.
Small-scale variation. Short-range variabilities related to detailed landform character and vegetation (Lilburne et al. 1998; Hewitt and Lilburne 1999), and plant species or associations of species (McIntosh and Allen 1993), were not addressed in this or the primary study (McIntosh et al. 2000), as land-use decisions are made on large blocks.
The surface incorporation of `fresh' slope material or loess would increase expected topsoil pH, decrease expected %P retention values, and dilute topsoil %C and C and N t/ha values. Conversely, the incorporation of pre-weathered materials at the surface, or the development of an A horizon on exposed pre-weathered materials, would be expected to lower topsoil pH, increase %P retention values, and also lower expected topsoil %C and C and N t/ha values. McIntosh and Allen (1993) found that pH was significantly lower and organic C values were significantly higher within Hieracium patches than in the adjacent soil, but found no significant difference in total N. There is a high turnover of rosettes within Hieracium populations as the rosettes are monocarpic (Makepeace 1985). As Hieracium cover has increased significantly over the last 50 years (Anon. 1976; Hunter 1991), variability from this source is expected to be large. McIntosh's (1997) review of work on nutrient changes accompanying botanical changes in the unfertilised South Island tussock grasslands shows a net decline in nutrients in biomass and soils under grazing, or grazing with burning, and measured maximum losses of up to 27 kg N/ha.year. The maldistribution of animal returns in the landscape is also a significant source of soil nutrient variability (Thorrold et al. 1985; Haynes and Williams 1993). The poor prediction of A-horizon exchangeable Ca most likely reflects unmapped variability in parent material composition.
Practical use of the modelled data
Environmental risk identification
The environmental risks associated with low levels of soil organic matter in the South Island high country have been reviewed by Hewitt and McIntosh (1996). They concluded that `total carbon of less than about 2.5% in the topsoil indicates that the soil is in poor condition'. Hewitt and Lilburne (1999) used a 4-fold classification of 0-7.5 cm C: 0-1.9% (very poor soil quality), 2.0-2.5% (marginal soil quality, at risk of decline), 2.6-4.0% (adequate soil quality), and >4.0% (high soil quality) as indicative of soil health on a study site on the Benmore Range. Adopting a simplified 3-fold classification--0-2% (poor), 2-3% (marginal), and >3% (adequate) of 0-7.5 cm carbon--59% of sites in the dry greywacke steeplands are correctly classified by the model; 39% of sites were classified in the adjacent class, and 9% in 2 classes removed. Some 94 000 ha are identified as having poor, and 187 500 ha as having marginal, 0-7.5 cm C values, i.e. 60% of the drier greywacke steeplands have marginal or poor 0-7.5 cm C values. Maps of such terrain are readily generated from the GIS.
Sound land management decisions can be made based on the enhanced knowledge of soil nutrient distribution resulting from the predictive modelling. The identification of sites with more favourable conditions for plant growth and response to applied seed and fertiliser can improve resource utilisation, and assist with species selection. For example, those sites below 1100 m with a 0-7.5 cm pH value of>5.5 and a P retention of <30% are recognised as being suitable for development by oversowing with clovers and topdressing, provided they retain adequate soil moisture (Floate 1992). For 0-7.5 cm pH and P retention, 79% and 70% of sites with the above characteristics, respectively, were correctly predicted. Of the 215 000 ha predicted to have potential for oversowing and topdressing with clovers, 65 000 ha have marginal 0-7.5 cm C values (2-3%) and a pH of 6.0-6.5, and 13 000 ha have 0-7.5 cm C values >3%. Of the soils with pH of 5.5-6.0, 61 000 ha have marginal 0-7.5 cm C values and 76 000 ha are predicted to have 0-7.5 cm C values >3%. Maps of each category of land can be generated from the GIS (e.g. Fig. 4--land suitable for development by oversowing and topdressing).
Improved resource inventories
Refined estimates of the carbon contained in the A horizons of the Benmore Range (0.93 million tonnes) confirms McIntosh et al.'s (1998) preliminary calculation of approximately 1 million tonnes. This value is about one-third of that calculated from previously available information using the NZLRI coverage and the national soils database (Tate et al. 1997, as discussed by McIntosh et al. 1998). A-horizon C for the dry greywacke steeplands for which the model is applicable is calculated to be some 18.6 million tonnes as against 43.1 million tonnes using the previously available information. Thus, the type of modelling outlined here has the potential to greatly improve the reliability of national resource inventories by incorporating variability driven by elevation and aspect, and pedological judgement.
Considering the harsh environmental conditions that operate in these dry greywacke steeplands, parent material variability, management history, and the history of vegetation change not directly included in the model, the model's performance was considered adequate. For example, it explained up to 45% of the variability for 0-7.5 cm pH. The model did not consistently perform better at any one of the test locations, and no consistent patterns were evident. The mean absolute error increased with distance from the model establishment area, for 0-7.5 cm %C, pH, and A-horizon total C and total N.
Predictions were most accurate for properties based on one laboratory measurement (e.g. 0-7.5 cm pH, %C, P retention). For derived values, based on several measurements (e.g. nutrient amounts in units of kg/ha), predictions are less accurate probably because of the cumulative effects of errors from field measurement and variability.
The mathematical model discussed greatly improves predictions of soil properties compared with a previous system based oil map units and a linked soils database. For example, the total dry greywacke steepland area of the South Island is estimated to contain 18.6 million tonnes of A-horizon C, rather than 43.1 million tonnes predicted by earlier models. The model can also be used to predict areas of environmental risk (e.g. areas of depleted topsoil organic matter) and areas that are in principle suitable for pasture improvement by oversowing and topdressing.
Table 1. Location and environmental characteristics of the three test areas Benmore Range Kirkliston Range Location Waitaki District (A) 15 km distant from Benmore Range Latitude c. 44 [degrees] 30'S c. 44 [degrees] 30'S Elevation range 400-1800 m 400-1900 m Uplift rates 0.4 mm/year 0.4 mm/year (Wellman 1979; Williams 1991) Bedrock Predominantly grey- Predominantly grey- wacke and argillite wacke and argillite (Mutch 1963; Gair (Gair 1967) 1967) Soil parent material Colluvium and loess Colluvium and loess Nearest Met. Station Tara Hills 488 m Tara Hills 488 m and elevation (NZ Met. Service 1984) Rainfall at Met. 528 mm (B), evenly 528 mm (B), evenly Station spread throughout spread throughout year year Mean air temperature 9.3 [degrees] C (1.5 9.3 [degrees] C (1.5 at Met. Station [degrees] C July; [degrees] C July; 15.9 [degrees] C 15.9 [degrees] C January) January) Vegetation Tussock grassland, Tussock grassland, Hieracium, scrub Hieracium, scrub Land system (McIntosh Glenbrook, Glencairn, Glencairn et al. 2000) Ben Omar Test sampling pattern Random & stratified Stratified sampling sites within the pattern (2 major model development ridges) (n = 24) area (n = 41) Molesworth Station Location Bonnington and Inland Kaikoura Ranges 250 km distant Latitude c. 42 [degrees] 15'S Elevation range 700-1800 m Uplift rates 2.0 mm/year (Wellman 1979; Williams 1991) Bedrock Predominantly graded bedded greywacke and argillite (Lensen 1962) Soil parent material Colluvium and loess Nearest Met. Station Molesworth Station 669 m and elevation (NZ Met. Service 1984) Rainfall at Met. 669 mm (B), evenly spread Station throughout year Mean air temperature 8.1 [degrees] C (1.5 at Met. Station [degrees] C July; 14.0 [degrees] C January) Vegetation Tussock grassland, Hieracium, scrub Land system (McIntosh Glenbrook et al. 2000) Test sampling pattern Stratified sampling pattern, 3 locations for coverage (n = 16) (A) Range on which the McIntosh et al. (2000) model was developed. (B) Rainfall increases with elevation to values above those stated. Table 2. Mean error (ME) and mean absolute error (MAE) for modelled parameters Location 0-75 cm A horizon %C pH % P C N retention (t/tha) (t/tha) Benmore ME -0.2 0.03 -1.1 4.1 0.4 Range MAE 0.9 0.2 8.0 16.4 1.2 Kirkliston ME 0.4 0.0 -12.8 -12.6 -0.8 Range MAE 1.0 0.3 12.8 22.1 1.3 Molesworth ME 0.7 -0.03 -8.0 9.2 0.7 Station MAE 1.6 0.3 8.0 23.7 1.6 Location A horizon Exch. Ca Thickness (kg/ha) (cm) Benmore ME 319 2.1 Range MAE 601 5.6 Kirkliston ME 19.6 -7.0 Range MAE 786 7.6 Molesworth ME 301 1.6 Station MAE 542 5.0
We thank the farmers of the Benmore Range, Black Forest, Haldon and Molesworth stations for allowing us access for the survey; K. Giddens for the soil analyses; H. Wallace for soil preparation and technical help; M. Kingbury for field assistance; T. Savage for preparing the maps; Dr A. E. Hewitt and M. McLeod, and the anonymous referees for commenting on the manuscript. This research was supported by the New Zealand Foundation for Research, Science and Technology, Contract CO9626.
Anon. (1976) Hieracium--good or bad? Tussock Grasslands and Mountainlands Institute Review 32, 38-54.
Barringer J, Wilde H, Willoughby J, Burgham S, Hewitt A, Gibb R, Newsome P, Rijkse W (1998) Restructuring the New Zealand Land Resource Inventory to meet the changing needs for spatial information in environmental research and management. In `Proceedings SIRC 98-10th Annual Colloquium of the Spatial Information Research Centre'. pp. 25-33. (University of Otago: Dunedin, New Zealand)
Blakemore LC, Searle PL, Daly BK (1987) Methods of chemical analysis of soils. New Zealand Soil Bureau, Scientific Report No. 80.
Cuff JRI (1973) A study of the influence of aspect on the nutrient requirements and soil chemistry of a selection of Hurunui steepland soils in south Canterbury. MAgrSc thesis, Lincoln College, University of Canterbury, Christchurch.
Dickinson WR (1971) Detrital modes of New Zealand greywackes. Sedimentary Geology 5, 37-56.
Floate MJ (Ed.) (1992) `Guide to tussock grassland farming.' (AgResearch: Invermay, New Zealand)
Gair HS (1967) `Sheet 20. Mt Cook. 1st edn geological map of New Zealand, 1 : 250 000.' (Department of Scientific and Industrial Research: Wellington, New Zealand)
Haynes RJ, Williams PH (1993) Nutrient cycling and soil fertility in the grazed pasture ecosystem. Advances in Agronomy 49, 119-199.
Hewitt AE, McIntosh PD (1996) Soil organic matter in the South Island high country. Landcare Research Science Series No. 18.
Hewitt AE, Lilburne LR (1999) Mapping soil carbon using Expector in the South Island high country. In `Proceedings of Third Conference of the Working Group on Pedometrics of the International Union of Soil Science, Sydney'. pp. 19-20.
Hudson BD (1992) The soil survey as paradigm-based science. Soil Science Society of America Journal 56, 836-841.
Hughes P, James G, Woods K (1995) `A review of the Government system for managing the South Island tussock grasslands: with particular reference to tussock burning.' (Parliamentary Commissioner for the Environment: Wellington, New Zealand)
Hunter GG (1991) The distribution of hawkweeds (Hieracium spp.) in the South Island, indicating problem status. New Zealand Mountain Lands Institute Review 48, 21-31.
Lensen GJ (1962) `Sheet 16. Kaikoura. 1st edn geological map of New Zealand, 1 : 250 000.' (Department of Scientific and Industrial Research: Wellington, New Zealand)
Lilburne LR, Hewitt AE, McIntosh PD, Lynn IH (1998) GIS-driven models of soil properties in the high country of the South Island. In `Proceedings SIRC 98-10th Annual Colloquium of the Spatial Information Research Centre'. pp. 173-180. (University of Otago: Dunedin, New Zealand)
Lilburne LR, Hewitt AE, Lynn IH, Benwell GL (1999) Some scale issues in spatial modelling of soil properties. In `Proceedings of Third Conference of the Working Group on Pedometrics of the International Union of Soil Science'. University of Sydney. (Collated by Inakwu O. A. Odeh) pp. 53-55.
Lynn IH (1996) Land use capability classification of the Marlborough Region: a report to accompany the second edition New Zealand land resource inventory. Landcare Research Science Series No. 12.
Makepeace W (1985) Growth, reproduction and production biology of mouse-ear and king devil hawkweed in eastern South Island, New Zealand. New Zealand Journal of Botany 23, 65-78.
Mark AF (1992) Indigenous grasslands of New Zealand. In `Ecosystems of the World, Volume 8B. Natural grasslands, Eastern Hemisphere'. (Ed. RT Coupland) (Elsevier: Amsterdam)
Mark AF (1994) Effects of burning and grazing on sustainable utilisation of upland snow tussock (Chionochloa spp.) Rangelands for pastoralism in South Island, New Zealand. Australian Journal of Botany 42, 149-161.
McCaskill LW (1969) `Molesworth.' (A. H. & A. W. Reed: Wellington, New Zealand)
McDonald WS, Giltrap DJ, McArthur A J (1988) Revised SPG1 database system manual (V1.2). New Zealand Soil Bureau Laboratory Report No. SS16, Department of Scientific and Industrial Research, Wellington, New Zealand.
McIntosh PD (1997) Nutrient changes in tussock grasslands, South Island, New Zealand. Ambio 26, 147-151.
McIntosh PD, Allen BE (1993) Soil pH declines and organic carbon increases under hawkweed (Hieracium pilosella). New Zealand Journal of Ecology 17, 59-60.
McIntosh PD, Hunter GG (1997) Soils and land use issues in the Mackenzie hill country. Landcare Research Science Series No. 19.
McIntosh PD, Blackholm G, Smith J (1981) Soil variation related to landscape and vegetation features in North Otago hill country. New Zealand Journal of Science 24, 225-244.
McIntosh PD, Lynn IH, Lilburne LR, Kingsbury M, Giddens K (1998) Is the soil carbon of dry South Island mountain ranges being accurately assessed? New Zealand Soil News 46, 13-15.
McIntosh PD, Lynn IH, Johnstone PD (2000) Creating and testing a geometric soil-landscape model in dry steeplands using a very low sampling density. Australian Journal of Soil Research 38, 101-112.
McKenzie NJ, Ryan PJ (1999) Spatial prediction of soil properties using environmental correlation. Geoderma 89, 67-94.
Moore LB (1976) The changing vegetation of Molesworth Station, New Zealand 1944 to 1971. DSIR Bulletin No. 217.
Moore ID, Gessler PE, Nielsen GA, Peterson GA (1993) Soil attribute prediction using terrain analysis. Soil Science Society of America Journal 57, 443-452.
Mutch AR (1963) Sheet 23. Oamaru. 1st Edn Geological Map of New Zealand, 1 : 250 000. Department of Scientific and Industrial Research, Wellington, New Zealand.
Nash JE, Sutcliffe JV (1970) River flow forecasting through conceptual models. Part 1--a discussion of principles. Journal of Hydrology 10, 282-290.
National Water and Soil Conservation Organisation (1975-1979) New Zealand Land Resources Inventory worksheets. 1st edn, 1 : 63 360. National Water and Soil Conservation Organisation, Wellington, New Zealand.
Newsome PF J (1987) The vegetative cover of New Zealand. Water and Soil Miscellaneous Publication No. 112. National Water and Soil Conservation Authority, Wellington, New Zealand.
New Zealand Meteorological Service (1984) Summaries of climatological observations to 1980. New Zealand Meteorological Service Miscellaneous Publication No. 177.
O'Connor KF, Harris PS (1991) Biophysical and cultural factors affecting the sustainability of high country pastoral uses. In `Proceedings of the International Conference on sustainable land management'. Napier, Hawkes Bay, New Zealand. (Ed. PR Henriques) pp. 304-313. (Hawkes Bay Regional Council: Napier, New Zealand)
Skidmore AK, Ryan PJ, Dawes W, Short D, O'Loughlin E (1991) Use of an expert system to map forest soils from a geographical information system. International Journal of Geographical Information Systems 5, 431-445.
Skidmore AK, Watford F, Luckananurig P, Ryan PJ (1996) An operational GIS expert system for mapping forest soils. Photogrammetric Engineering and Remote Sensing 62, 501-511.
Stevens EJ, Hughes JG (1973) Distribution of sweet brier, broom and ragwort on Molesworth Station. Tussock Grasslands and Mountain Lands Institute Special Publication No. 9.
Tare KR, Giltrap DJ, Claydon JJ, Newsome PF, Atkinson IAE, Taylor MD, Lee R (1997) Organic stocks in New Zealand's terrestrial ecosystems. Journal of the Royal Society of New Zealand 27, 315-335.
Thorrold BS, O'Connor KF, White JGH (1985) Management influences on sheep behaviour, dung distribution and soil phosphate. Proceedings of the New Zealand Grassland Association 46, 127-134.
Tonkin PJ (1984) Studies of soil development and distribution in the eastern hill country, central South Island, New Zealand. PhD thesis, Lincoln College, University of Canterbury, Christchurch.
Wellman HW (1979) An uplift map for the South Island of New Zealand and a model for the uplift of the Southern Alps. In `The origin of the Southern Alps'. (Eds RI Walcott, MM Cresswell) pp. 13-20. The Royal Society of New Zealand Bulletin No. 181.
Williams PW (1991) Tectonic geomorphology, uplift rates and geomorphic response in New Zealand. Catena 18, 439-452.
Willmott CJ (1984) On the evaluation of model performance in physical geography. In `Spatial statistics and models'. (Eds GL Gaile, CJ Willmott) pp. 443-460. (D. Reidel: Dordrecht, The Netherlands)
Manuscript received 12 April 2001, accepted 24 August 2001
I. H. Lynn (A)(C), L. R. Lilburne (A), and P. D. McIntosh (B)
(A) Landcare Research, Lincoln, Canterbury, New Zealand.
(B) Forest Practices Board, Hobart, Tasmania, Australia.
(C) Corresponding author; email: firstname.lastname@example.org
|Printer friendly Cite/link Email Feedback|
|Author:||Lynn, I.H.; Lilburne, L.R.; McIntosh, P.D.|
|Publication:||Australian Journal of Soil Research|
|Article Type:||Statistical Data Included|
|Date:||Mar 1, 2002|
|Previous Article:||Simulating infiltration and the water balance in cropping systems with APSIM-SWIM.|
|Next Article:||Risk mapping of soil acidification under Stylosanthes in northern Australian rangelands.|