European Journal of Agronomy Integrating farmer ...

3 downloads 0 Views 954KB Size Report
B., Tozer, P., 2007. Opportunities and constraints for managing within-field spatial variation in Western Australia grain production. Field Crops Res. 104,. 60–67.
Europ. J. Agronomy 32 (2010) 40–50

Contents lists available at ScienceDirect

European Journal of Agronomy journal homepage: www.elsevier.com/locate/eja

Integrating farmer knowledge, precision agriculture tools, and crop simulation modelling to evaluate management options for poor-performing patches in cropping fields Y.M. Oliver a,∗ , M.J. Robertson a , M.T.F. Wong b a b

CSIRO Sustainable Ecosystems, Private Bag 5, PO Wembley, Western Australia 6913, Australia CSIRO Land and Water, Private Bag 5, PO Wembley, Western Australia 6913, Australia

a r t i c l e

i n f o

Article history: Received 4 July 2008 Received in revised form 16 March 2009 Accepted 14 May 2009 Keywords: Precision agriculture APSIM Potential yield Plant available water capacity Subsoil constraint

a b s t r a c t Cropping fields often have poor-performing patches. In an attempt to increase production on poor patches, farmers may apply additional fertiliser or ameliorants without economic or scientific justification. Improved understanding of the extent and causes of poor performance, management options, potential crop yield and economic benefits can give farmers the tools to consider management change. This paper presents an approach to integrating farmer knowledge, precision agriculture tools and crop simulation modelling to evaluate management options for poor-performing patches. We surveyed nine cropping fields in Western Australia and showed that (1) farmers have good understanding of the spatial extent and rank performance of poor-performing areas, when compared to NDVI or yield maps, (2) there is a wide range of physical and chemical soil constraints to crop yield in such patches, some of which can be ameliorated to raise yield potential, and others where crop inputs such as fertiliser can be better matched to low yield potential. Management options for poor-performing patches were evaluated through simulation analysis by removal of constraints to rooting to varying extents, and hence plant available water capacity. These examples show that if the constraint is mis-diagnosed then the potential benefits from amelioration can be overstated. In many cases constraints, often associated with physical limitations such as shallow available rooting depth or light-texture cannot be ameliorated or are uneconomic to ameliorate. In such cases the best intervention may be to lower crop inputs to better match the water-limited yield potential of such poor-performing areas. This research integrated farmer knowledge and spatial data to define yield zones in which targeted soil sampling and crop simulation were then used to determine yield potential and particular constraints to that potential. The economic costs and benefits of differential zone management were examined under a range of husbandry scenarios and, importantly, the sensitivity of economic gain to mis-diagnosis or errors in defining the zones was tested. This approach provided farmers with a robust and credible method for making decisions about spatial management of their fields. Crown Copyright © 2009 Published by Elsevier B.V. All rights reserved.

1. Introduction Cropping fields in the Western Australia (WA) wheatbelt are often large (50–200 ha) with significant spatially heterogeneous soils, crop performance and by inference, profit. Farmers have knowledge of the location of poor-performing patches within their fields, but rarely understand the spatial extent of patches or the basis for poor performance. In an attempt to increase production on poor patches, farmers may apply additional fertiliser or ameliorants without economic or scientific justification. Improved understand-

∗ Corresponding author. Tel.: +61 8 9333 6469; fax: +61 8 9333 6444. E-mail address: [email protected] (Y.M. Oliver).

ing of the extent and causes of poor performance, management options, potential crop yield and economic benefits can give farmers the tools to consider management change. In Western agriculture there is increasing availability of information on spatial variation in soil and crop performance via yield mapping, soil survey and remote sensing (Cook and Bramley, 1998; Corwin and Lesch, 2003; Godwin and Miller, 2003; McBratney et al., 2005). Knowledge of the distribution and identification of crop yield, soil type and plant available water capacity (PAWC) allows exploitation of the spatial variation for site specific management (nutrient, ameliorant, cropping system change, etc.) (Sadler and Russell, 1997; Adams et al., 2000; Wong et al., 2001; Zhang et al., 2002; Koch et al., 2004; Robertson et al., 2007). There have been only a few studies that have assessed the degree to which spatial

1161-0301/$ – see front matter. Crown Copyright © 2009 Published by Elsevier B.V. All rights reserved. doi:10.1016/j.eja.2009.05.002

Y.M. Oliver et al. / Europ. J. Agronomy 32 (2010) 40–50

information supplements farmers’ own knowledge of the location of poor-performing patches (Fleming et al., 1999, 2000; Booltink et al., 2001; Wong et al., 2008). The ability of farmers to define management zones that match crop performance has been tested by matching the ranking of measured yields in these zones (Fleming et al., 2000; Khosla et al., 2002; Hornung et al., 2006). However, there has been little analysis of the cost in financial terms of errors in definition of patch (or zone) boundaries when considered in a zone management context. The zone definition can be inaccurate (due to lack of knowledge or techniques to identify zones) or imprecise (due to logistical considerations relating to location and size of zone which can be managed variably by the farmer and farmer’s machinery). While locating the position and extent of poor-performing patches may be straightforward for many farmers with detailed historical knowledge of their fields, the diagnosis of the causes of yield constraints is more complicated. In Western Australia, causes of poor performance can be linked to low soil plant available water capacity (PAWC) (Tennant and Hall, 2001). PAWC is one of the main drivers of yield potential variation in Mediterranean environments and the relationship is seasonally dependent (Mulla et al., 1992; Morgan et al., 2003; Oliver et al., 2006). PAWC is the difference between the drained upper limit (DUL), or water holding capacity after drainage has ceased, and the crop lower limit (CLL) which is determined by the tightness with which water is held within the soil matrix and the crop’s ability to extract that water to crop rooting depth. Soil type (clay content, structure) affects the DUL as it determines the water holding capacity of the soil, while soil type, soil depth and chemical constraints will affect the ability of the crop to grow roots to depth and extract water. With knowledge of the soil type and crop rooting depth, the PAWC can be estimated. Spatial variation in PAWC has been linked to variation in crop yield in under high fertility conditions, Mediterranean winter dominant rainfall for winter wheat in Western Australian soils (Oliver et al., 2006; Wong and Asseng, 2006), and in mid west USA (Mulla et al., 1992; Morgan et al., 2003). Low PAWC, in turn, has been linked to soil constraints that limit crop production, acting through reduced rooting depth and increased CLL. Soil constraints commonly encountered in WA are compaction and acidity of a plough plan layer at 0.2–0.3 m and acidity to depth (Tennant et al., 1992; Hamza and Anderson, 2003; Davies et al., 2006), water logging (Belford et al., 1992; Tennant et al., 1992), salinity (George et al., 1997; McFarlane and Williamson, 2002), water repellent topsoils (Blackwell, 1993; Harper and Gilkes, 1994) and sodic soils (Cochrane et al., 1994). Once the underlying cause of poor crop performance has been identified it is necessary to consider a range of management responses (including doing nothing), accounting for seasonal variation on the effect of any management intervention, and the economic return. In WA grain-growing systems, management responses may include reducing the level of crop inputs (such as fertiliser or plant density) to match yield potential, application of ameliorants such as lime or gypsum to overcome soil chemo-physical constraints, or deep ripping to modify compacted layers. Simulation modelling has been used to assess the potential payoffs to such interventions (Asseng et al., 1998; Wong and Asseng, 2007). In such cases the effect of a subsoil constraint like compaction, acidity or shallow depth to bedrock is simulated by using a root hospitality factor to adjust the rate of root depth extension according to the severity of the subsoil constraint (Asseng et al., 1998; Wong and Asseng, 2007) or adjustment to the crop lower limit (and therefore PAWC) (Sadras et al., 2003; Hochman et al., 2004, 2007). However, such adjustments in simulation models are often subjective because they depend on the effect of the constraints on root growth and the severity of the constraint.

41

With farmers at workshops and field days in the low-medium rainfall zone (200–400 mm) of the Western Australian wheatbelt we have trialled a four-stage approach to identify the location, causes and management options for poor-performing areas of a field. The approach combines farmer knowledge with precision agriculture’s spatial data and soil diagnosis. Integrating farmer knowledge with other spatial data may be able to reduce cost of data collection and analysis and also improve communication with farmers. Modelling is used to quantify yield potential as well as the yield gains and financial benefits of amelioration, particularly in relation to increasing soil rooting depth and soil PAWC. Accordingly, the aims of this paper are to: (1) describe the process used, (2) analyse the sensitivity of various aspects of the process to variation in assumptions, and (3) highlight where understanding derived through use of scientific tools and analysis can supplement farmer knowledge. 2. Methods 2.1. Locations in Western Australia The studies were conducted with four farmers in two graingrowing regions of the Western Australian wheatbelt, which receive between 300 and 400 mm long-term mean annual rainfall. Kellerberrin is in the central wheatbelt of Western Australia, between 30.8◦ and 32.3◦ S and 116.7◦ and 118.6◦ E, with typically texture contrast (duplex) soils of sand or loams over clay (Luvisols and Lixisols), sands over gravels (Ferralsols) and sand (Arenosols) (Schoknecht, 2002; FAO, 1998). Buntine is in the Northern Agriculture Region which lies between 28.3◦ and 30.7◦ S and 114.7◦ and 116.8◦ E. Soils are deep, well drained sands (Arenosols), yellow sandy loams and loamy sands (Lixisols and Ferralsols) and loams over clay duplex soils (Luvisol and Ferralsol) (Schoknecht, 2002; FAO, 1998). Both regions are in a mixed cropping zone with cropping sequences based on spring wheat (Triticum aestivum L.) in rotation with barley (Hordeum vulgare L.), grain lupins (Lupinus angustifolius L.), canola (Brassica napus L.) and sometimes annual pasture. Typical farm wheat yields in both regions are between 1 and 3 t/ha. In each location, farmers identified fields on their farms that were known to have obvious spatially variable crop yield. In total 9 fields were analysed, one field each from the two farmers in Buntine and seven fields from the two Kellerberrin farmers, and these varied in area from 39 to 244 ha (Table 1). 2.2. The process 1. 2. 3. 4.

Determine the location of different performing areas. Define soil properties and constraints to production. Estimate yield potential with and without constraints. Determine the benefits to management change.

At each stage of the process, the effect of wrongly estimating each component is also determined. 2.2.1. Location of poor-performing patches Farmers were asked to define the boundaries of zones within the field that had distinctly different soil types and performance. Initially soil maps were drawn and then these soil types were allocated to an above-, at- or below-average performance for the field. A farm boundary map or an aerial photograph was often used as the background for these maps (an aerial photograph could be easily obtained from Google Earth® ) while the farmers had a range of prior knowledge of the field. The information from the farmers was entered into GIS as polygon layers and converted to a 25 m grid. Spatial correlations were conducted between farmeridentified zones and those identified via “objective” spatial data

42

Y.M. Oliver et al. / Europ. J. Agronomy 32 (2010) 40–50

Table 1 % agreement between farmer idea and performance where the zones are classed as below average (BA), average (A) and above average (AA). Field

Location

Zoned by

Years used in zone analysis

Field size (ha)

% area where yield class (from spatial data is defined by farmer as) Correct

A1 A2 A3 B1 B2 B3 B4 C1 D1

Kellerberrin Kellerberrin Kellerberrin Kellerberrin Kellerberrin Kellerberrin Kellerberrin Buntine Buntine

NDVI NDVI NDVI NDVI NDVI NDVI NDVI Yield Yield

96, 98, 99, 01, 03, 04 95, 98, 99, 01, 02, 04 95, 97, 99, 01, 02, 04 98, 99, 01, 04 95, 97, 98, 03, 05 96, 00, 01, 02 95, 99, 03, 05 96, 99, 04, 05 96, 99, 01, 03, 05

72 60 62 71 77 39 69 244 134

using the same 25 m grid size and location. The farmers then viewed the spatial data available and commented on how this data matched their knowledge, areas where there was discrepancy and possible reasons for this. This process was used to target areas for additional soil sampling in the areas with discrepancy and could be used to redraw a performance map (however this was not done here). The two Kellerberrin farmers do not have yield monitors and have not viewed any yield maps or NDVI spatial data of their fields prior to drawing the performance maps. Their knowledge, which comes from inspecting the field, was used to draw their soil map and performance maps. For these farmers, the “objective” performance zones were created from a historical series of mid-season normalised difference vegetation index (NDVI, Tucker, 1979) from the Landsat satellite which is supplied at a 25 m grid. The historical NDVI record in Australia extends back to 1993 and allows an analysis of the consistency of ranking, from season-to-season, of below, above and average zones. The usefulness of this technique relies on a strong positive correlation between NDVI and yield, at least at regional scale in WA (Smith et al., 1995). We have found that growers are familiar with NDVI as a measure of crop size and easily relate it to their knowledge of the field. Each field was classed into 3 zones using k-mean analysis (Whelan and McBratney, 2003) for 4 or more years of NDVI images from seasons when cereals were grown. The three zones represented consistently average, belowaverage, and above-average portions of the field. Robertson et al. (2007) showed that a high percentage of the fields they analysed had greater than two-thirds of the field area with NDVI that was consistent from season-to-season in its rank performance. In these situations of consistent performance, NDVI can be used with confidence over a large area of a farm to define management zones (Adams and Maling, 2005). The two Buntine farmers have been yield mapping for 4 years, and have viewed their yield maps but not used them to create variable rate prescription maps. They created their soil map and performance map from their knowledge of the yield maps (but these were not in front of them at the time) and from inspecting the field. The “objective” performance maps were created from 4 to 5 years of yield monitor maps of cereal crops which were obtained from the farmers who had acquired them using a range of yield monitoring equipment (Case AFSTM , RinnexTM , JD OfficeTM ) attached to grain harvesters. The yield map data were cleaned to remove errors from geographic anomalies, exceptionally high yields (>10,000 kg/ha) and errors caused by the harvester dynamics using a grain surge routine (Beck et al., 1999; Robinson and Metternicht, 2005). Smoothed yield maps were generated by kriging the data using techniques outlined by Whelan et al. (2001) and superimposing the surface onto a 25 m grid in Arc GIS. The field was then clustered into 3 zones using the k-mean analysis to define zones as average, below-average and above-average yielding areas (Whelan and McBratney, 2003).

52 44 37 32 42 33 40 57 53

1 class away

2 classes away

BA as A, A as AA

A as BA, AA as A

BA as AA

AA as BA

6 37 18 29 31 23 17 21 32

38 12 27 39 9 44 43 14 6

0 7 6 0 17 0 0 2 8

4 0 12 0 1 0 0 6 1

The percent agreement between farmer-identified rank performance of each zone and that from zoning from NDVI images or zoning from yield maps was determined. Where pixels were misclassified (e.g. high as low, medium as high), the misclassifications were categorised in terms of whether they were one or two classes different. A sensitivity analysis was conducted to quantify the costs involved in incorrectly defining the boundaries between performance zones. Two hypothetical fields were considered: one with a high degree of within-field contrast where equal-sized zones had wheat yield potentials of 500, 1750 and 3000 kg/ha, and another where yield contrast was less at 1500, 1750 and 2000 kg/ha. The consequences of inaccurate zone definition were quantified by assuming varying proportions of the below-average, average and above-average zones had low, medium and high rates of N and P applied, up to a limit of 33% of the maximum size of a zone in the field. N and P nutrient response curves, scaled to yield potential, were used to calculate the yield at any given rate of N and P applied with the soil supply constant across all soils and zones. A partial gross margin used the value of the crop (yield * price of yield at $0.28/kg) minus the cost of fertiliser when nitrogen is $1.2/kg N and phosphorous is $2.5/kg, with all costs and prices in Australian Dollars. The method and equations can be found in more detail in Robertson et al. (2008) and Oliver and Robertson (2009). 2.2.2. Soil properties and constraints Soils in each performance zone were characterised for PAWC and constraints to rooting, with the location of the soil sampling chosen to be in the middle of a large area of similar performance, not near boundaries of performance, patchy performance or fences. At Kellerberrin, one soil pit in each performance zone was dug to 2 m, and soil texture and chemistry analysed in the field and laboratory using standard techniques (Rayment and Higginson, 1992). PAWC was estimated from texture, constraints to depth and observed rooting depth. At Buntine, intact soil cores were extracted to 2 m at various locations in each field where observed yields varied. Cores were classified visually and analysed chemically using standard laboratory methods (Rayment and Higginson, 1992). The drained upper limit and crop lower limit were estimated using methods described by Dalgliesh and Foale (1998) and in conjunction with root depth were used to calculate the PAWC of the soil. All soils were described using the Western Australian soil classification scheme (Schoknecht, 2002) (similar to the farmer’s soil description) and the World reference base for soils (FAO, 1998). Soil chemical and physical properties most commonly used to identify constraints were pH, electrical conductivity (EC), bulk density, soil strength and surface conditions. These were used in conjunction with the presence of roots in cores and patterns of the roots in pits to determine constrained rooting depth. Soils with pH less than 4.5 (in CaCl2 ) have the potential to reduce yield because low pH increases aluminium concentration to toxic levels which

Y.M. Oliver et al. / Europ. J. Agronomy 32 (2010) 40–50

then impairs root growth and so reduces the availability of major plant nutrients (N, P, K, S, Ca, Mg) (McKenzie et al., 2004, p. 16). EC levels greater than 0.15 dS/m for sandy soils (using a soil-to-water ratio of 1:5) are considered moderate but can begin to constrain crop growth (Shaw, 1999). A compaction layer was determined from penetrometer readings or push probe, where the ability of plant roots to penetrate soil is restricted as soil strength increases and ceases entirely at 2.5 MPa (Hamza and Anderson, 2005). Surface conditions (crusting from observation, stability from dispersion test and non-wetting from water drop penetration test) can reduce water infiltration and crop establishment (McKenzie et al., 2004, p. 21). 2.2.3. Yield potential estimates Estimates of water-limited yield potential were provided by simulations from the crop model APSIM (Keating et al., 2003) with soil information collected in each zone. Simulations using APSIM have been shown to account for 82% of observed variation in crop yield (between 0 and 5500 kg/ha) due to soil type and season, including within-field variation when the model was well parameterised with the soil PAWC, start of season soil conditions (water content and nitrogen levels), climate data from the closest bureau of meteorology weather station and the known farmer agronomy i.e. sowing date and nutrient applications (Oliver et al., 2006; Oliver and Robertson, 2009). The model output was used to explain the main drivers of yield variation and to provide insight into the variability observed by the farmers and measured by the spatial data. Various soil constraints were simulated by varying the root extension factor in the relevant soil layers, so that the rate of root depth development through the season and final maximum depth varied. Historical daily climate data (1905–2006) from each location was used to simulate the effect of soil type on yield potential and its variation with season. Management specifications for the long-term simulations included cv. Wyalkatchem wheat sown at 150 plants/m2 between 5th May and 30th June when 15 mm of rain fell over 10 days, however these conditions were not met in 1978, 1985 and 2006 and, as crops were not sown, these years were excluded from the analysis. The median date of sowing was 17th May at Buntine and 18th May at Kellerberrin. Nitrogen was applied at 90 kg N/ha at sowing, with a further 90 kg N/ha applied 49 days after sowing with all other nutrients non-limiting. These high levels of nitrogen application was to ensure yield was not limited by nitrogen over the range of yields (200–5500 k/ha). The soil water was reset to crop lower limit on the 1st January every year, with soil nitrate and surface organic matter reset annually on the 19th April, to remove the impact of the previous crop and season on the following crop. The available nitrogen in the soil was set at 50 kg N/ha as NO3 and 25 kg N/ha as NH4 . 2.2.4. Payoffs to alternative management options Management options for poor-performing patches were evaluated through simulation analysis of four situations: 1. Reduced fertiliser input, so that cost savings occur due to better matching nutrient supply to crop demand. The impact of varying rates of fertiliser N, P and K was assessed by nutrient response curve, the asymptote of which varied with the water-limited potential yield as supplied by APSIM. The method and equations can be found in more detail in Robertson et al. (2008) and Oliver and Robertson (2009) who used them in studies of spatial variation in crop yield and consequences for nutrient management in similar environments to this study. 2. A constraint to rooting near the surface is removed with hospitable conditions for root growth in the rest of the soil profile so that the roots are able to extend throughout the soil profile.

43

3. A constraint near the surface is removed, i.e. an acid band or hard pan in the 0.2–0.3 m layer, but the soil is acidic or compacted throughout the profile so that only a small increase in PAWC or small increase in rooting depth occurs. 4. A constraint to rooting deep in the profile (such as deep compaction) is removed. For these scenarios the poor-performing area was assumed to be a deep sand (Arenosol) with 50 mm PAWC/m with the rooting depth adjusted from 0.2 m for a profile PAWC of 10 mm in 0.1 m increments up to a rooting depth of 2.1 m for a profile PAWC of 105 mm. The APSIM model was used to simulate the potential yield at a range of profile PAWC’s over 100 years using the rainfall data from Kellerberrin and Buntine meteorological stations. These were used to create response curves between yield and PAWC and grouped into the 25th percentile, 50th percentile and 75th percentile years. 3. Results and discussion 3.1. Location of poor-performing patches Farmers were comfortable with defining the boundaries between zones of above-, at- and below-average crop performance. Fig. 1 illustrates two of the fields, one from Kellerberrin (A1) and one from Buntine (C1), with soil and performance maps defined by the farmers and performance defined by NDVI images (Kellerberrin) and yield maps (Buntine) with a 3-zone performance classification. In most cases the farmers converted their soil maps directly to performance zones, highlighting that this is the dominant underlying cause of spatial variation in crop performance, at least in the minds of the farmers. The Kellerberrin farmer described his field (A1) as consisting of soil types of sandy duplex (Arenic Lixisol), deep pale sands (Albic Arenosol), poor (“gutless white”) sand over gravel (Arenic ortihiplithic Lixisol) and red clay duplex (Chromic Luvisol) (Fig. 1a). The farmer associated the poor areas with the poor sand and the above-average area with the red clay duplex (Fig. 1b). When the farmer then viewed the NDVI performance maps, he thought they represented his knowledge of the field well (Fig. 1c) and felt viewing this jogged his memory that the top right hand section of the field is actually an above-average performing area but thought it was a different soil to the above average performance of the red duplex soil. This then allowed him to adjust his “mud map” before considering any zone management. The Buntine farmer divided his field (C1) into 6 soil types, which were sands of varying performance and degrees of soil salinity labelled as better sand, valley sand and salt-affected (Xanthic Ferralsol, Albic Arenosols, Hyposalic Arenosol, respectively), gravels of different depth labelled as shallow gravel and gravel (Plinthic Ferralsol and Ferric Ferralsol, respectively) and a clay soil labelled as heavy soil (Calcic Luvisol) (Fig. 1d). The farmer defined his aboveaverage performing areas as the better sand, the average areas were the shallow gravel and gravel and the below-average areas corresponded to the salt-affected and valley sands and heavy soil (Fig. 1e). The farmer agreed with performance maps as determined from a series of yield maps (Fig. 1f), however had reservations about the patchiness of the zones. The two examples in Fig. 1 illustrate that farmers were able to correctly identify the location of poor-performing zones compared to above average patches, but not their extent. When using a threezone system of below-average, average and above-average yield areas, the Kellerberrin farmer correctly classified 52% of the field (A1), while the Buntine farmer was correct for 57% of the field (C1). From the nine fields tested, the four farmers correctly classified about 43% (range 32–57%) of the variation in the field when com-

44

Y.M. Oliver et al. / Europ. J. Agronomy 32 (2010) 40–50

Fig. 1. The maps for field A1 in Kellerberrin (a, b, c) and field C1 in Buntine (d, e, f) which have been determined from; farmer description of the soils in their field (a, d), farmer defined performance (-below, -at or above-average) of the soils in their field (b, e) and the field zoned into -below, -at or above-average performance by NDVI (c) and yield maps (f).

pared to the NDVI or yield classification (Table 1). Most commonly, the mis-classifications were one class away, i.e. classing average performance as above-average, above-average performance as average, average performance as below-average or below average performance as average. This is mostly due to the yield maps and NDVI images not being spatially contiguous (and are 25 m2 pixel size or 0.0025 ha zones), whereas in our experience a farmer will typically identify areas as continuous management “zones” greater than 1 ha. The minimum size of a zone will depend on the abil-

ity of farmer to differentially manage within a field and is affected by accuracy of GPS and variable controller equipment, as well as the shape and cropping direction of the field (Zhang et al., 2002). The consequence of this is that farmers rarely mis-classify a poorperforming area as above average (2–17% of the field) or the reverse where an above-average area is identified as a poor area (1–12% of the field). Smaller field size did not increase the farmer’s ability to correctly predict zones (Table 1), as even the small fields have large varia-

Y.M. Oliver et al. / Europ. J. Agronomy 32 (2010) 40–50

45

Table 2 Table of constraints, PAWC, yield potential and potential management options for each of the below-average (BA), average (A) and above-average (AA) zones in the nine fields. Zone

Soil and constraint

FAO soil

A1

BA

Arenic ortihiplinthic Lixosol Albic Arenosol Arenic Lixisol Chromic Luvisol

20

1.3

A AA

Poor white sand over gravel Pale sands with acid layer Shallow sandy duplex Red clay

60 110

2.2 2.7

A2

BA A AA

Sand—acidic through whole profile Gravely soil Sand—acidic in 0.1–0.2 m layer

Arenosol Ferric Ferralsol Arenosol

15 45 100

1.0 1.9 3.0

A3

BA A AA

Sand—machinery induced compaction at 0.1–0.2 m Gravely soil Deep sand

Arenosol Ferric Ferralsol Arenosol

20 60 80

1.3 2.3 2.6

B1

BA A AA

Ironstone gravel sand Shallow gravel soil Gravely sand

Ferric Lixisol Leptic ortihiplinthic Lixisol??? Ferric Ferralsol

15 20 40

1.0 1.3 1.6

B2

BA A AA

Sand—subsoil salinity Sand over clay duplex soil Sandy earth

Hyposalic Arenosol Arenic Lixisol Xanthic Ferralsol

25 40 80

1.5 1.8 2.5

B3

BA A AA

Shallow sand over rock Sand over clay duplex Sandy earth

Areni-Leptic Regosols Arenic Lixisol Xanthic Ferralsol

30 60 100

1.6 2.2 3.0

B4

BA A AA

Gravelly soil Sand over clay duplex Deep sand/sandy earth

Ortihiplinthic Ferralsol Arenic Lixisol Xanthic Ferralsol

30 50 100

1.6 2.0 3.0

C1

BA

Poor sand Sand with saline subsoil Gravelly soils Sand—beginnings of acidity and compaction in 0.1–0.2 m layer

Albic Arenosol Hyposalic Arenosol Ferric Ferralsol Xanthic Ferralsol

40

1.8

60 95

2.3 2.6

Shallow sands over rock Acid sand over gravel Sand—acidic in 0.1–0.2 m layer Red calcareous clay

Areni-Leptic Regosols Arenic ortihiplinthic Lixisol Eutric Arenosol Calcic luvisol

25

1.5

50 110

2.1 2.8

A AA D1

BA A AA

tion in yield and soil type (Table 2). No relationship between yield variation and field size was also found for a range of fields across Western Australia (Robertson et al., 2008) indicating the fields have not been fenced to soil type or to reduce the field variability for ease of management. A sensitivity analysis was conducted to quantify the potential financial loss due to incorrect classification of yield potential, under a 3-zone management system for fertiliser N and P application. Low, medium and high fertiliser rates were applied to various proportions of the below-average, average and aboveaverage potential zones. Where the wheat yield potential of each zone is 1500, 1750, and 2000 kg/ha for below-average, average and above-average potential zones, respectively, the income lost from incorrect application of N and P was minor (