Potentially mineralisable nitrogen: relationship to crop production and spatial mapping using infrared reflectance spectroscopy.
Matching soil nitrogen (N) supply to crop demand is critical to enable the development of farming systems that increase productivity through greater efficiency of N use, while minimising losses of N to the environment. Crop N uptake is dependent on the amount and distribution (timing and location) of soil N mineralisation and N fertiliser application (Angus 2001). Therefore, gains in efficiency related to crop production will depend, in part, on our ability to estimate N mineralisation (Mulvaney et al. 2001) and consider the spatial distribution of soil N supply within the paddock when defining nutrient management strategies (Mahmoudjafari et al. 1997; Corre et al. 2002; Baxter et al. 2003). Under non-leaching environments, measurements of deep soil, pre-plant mineral N (typically 0.6-1.2 m deep) can be used to adjust fertiliser rates based on soil available N (Angus 1992; Bundy and Meisinger 1994). However, on soils prone to leaching this measurement is not as useful. Here, pre side-dress nitrate testing (Magdoff et al. 1984) has been performed whereby soil sampling is delayed until crop growth stage indicates the period of peak plant N uptake. Both approaches require soil collection to depth at several locations to precisely estimate the spatial distribution of inorganic N within a paddock. Where this has occurred, soil sampling is generally structured around paddock zoning based on additional spatial information such as soil textural changes and previous years' yield monitoring data. This approach relies on an assumption of spatial dependence between sampling points (Giebel et al. 2006). However, previous studies have found that the spatial dependence for nitrate in the soil profile occurred at a distance less than the chosen sampling grid distance (<30 m for 0-0.9 m Dahiya et al. 1985; <50 m for 0-0.9 m Stenger et al. 1998).
Instead we propose the measurement of a simple index of soil N supply capacity in the surface 0-0.1 m layer of soil that can be performed rapidly (potentially within paddock) for minimal cost. Although inorganic N is available for plant uptake throughout the root-zone, most of this inorganic N is derived from soil organic matter mineralisation in the surface soil layer, especially in no-tillage systems where soil organic matter is concentrated in the surface few centimetres and soil inversion is avoided. Mapping the potential soil N supply capacity across a paddock thus provides a means of using only surface soil layer data to adjust N fertiliser application rates accordingly within decision support systems.
Potentially mineralisable N (PMN) as determined by anaerobic incubation (Keeney and Bremner 1966) is often considered to reflect the organic matter pool/s being mineralised (Stockdale and Rees 1994). It is a simple biochemical assay that is affected less by leaching events (Cookson and Murphy 2004) and climatic fluctuations (Murphy et al. 1998) than actual gross N mineralisation rates. Previous studies have thus used PMN as an index to assess within-paddock variation of soil N supply on a scale applicable to farm management practices (Baxter et al. 2003). In their study, Baxter et al. (2003) illustrated that for a 6-ha paddock cropped to winter barley, PMN exhibited sufficient spatial structure to create a map with the intended purpose of zoning for a variable N management strategy. Such an approach implies that PMN can be interpreted with direct relevance to the supply of crop N demand. While this was not reported by Baxter et al. (2003), other studies have found similar indices of soil N supply to be significantly related to crop N uptake (e.g. McTaggart and Smith 1993), although the percentage of yield variability explained is often low (<40%, Walley et al. 2002).
Although simple to perform, the time associated with laboratory analysis of PMN has restricted the widespread applicability of this and similar assays (Picone et al. 2002) as part of an N decision support system. Also, the spatial variation for measurements of soil N supply have been reported in the range 38-104m (Robertson et al. 1993; Cambardella et al. 1994), indicating that the required sampling intensity is likely to be cost-prohibitive for spatial mapping of large areas using standard biochemical analysis of PMN. An alternative to traditional biochemical analysis uses infrared (IR) reflectance spectroscopy (see review Janik et al. 1998) to detect both organic and inorganic components of soil. Once IR spectral libraries have been calibrated, an unknown soil sample can be scanned rapidly (multiple scans of the same sample performed within minutes) to provide predictions for a range of soil properties (Bertrand et al. 2002; Mimmo et al. 2002; Shepherd and Walsh 2002). This technology thus provides the potential for rapid and cost-efficient prediction of PMN for inclusion in within-paddock fertiliser recommendation systems and catchment-scale environmental management strategies. Calibration curves for PMN based on IR spectral libraries across a wide range of soils from different locations have resulted in high correlation between traditional biochemical determination of PMN and IR-predicted PMN (r=0.89, Russell et al. 2002; r=0.86, Shepherd and Walsh 2002). However, the capacity of IR reflectance spectroscopy to accurately zone the spatial distribution of PMN within a paddock has not previously been assessed. Here, we test whether the level of predictive accuracy by IR is sufficient to enable an accurate 'within-paddock' spatial map of PMN to be generated. The objectives of this study were therefore to: (i) determine the relationship between PMN and grain yield in semi-arid soils, and (ii) assess the capacity of IR reflectance spectroscopy to predict the magnitude and spatial distribution of PMN within an agricultural paddock.
Materials and methods
The relationship between PMN and wheat grain yield was assessed across 40 paddocks within a 10km by 50km area located in the central agricultural region of Western Australia. Major soils in this region (Chromosols, Sodosols, Kandosols) are highly weathered acidic soils (pH 1 : 5 in water; mean 5.68, min. 5.17, max. 6.66) with low clay (mean 10.6%, min. 3.2%, max. 39.2%) and soil organic matter carbon contents (mean 1.44%, min. 0.47%, max. 2.53%). Within each paddock, 3 replicate plots were established and soil collected in 0-0.1, 0.1-0.3, 0.3-0.6, and 0.6-0.9m layers 6 weeks after seeding. This sampling time corresponds to when additional fertiliser N could be applied if required; supplementing basal N fertiliser application rates at seeding. Within the paddocks, assessed total N-fertiliser application rates ranged from 0 to 47 kg N/ha.
Soil was sieved (<2 mm) and stored at 5[degrees]C for <7 days before analysis. Traditional biochemical analysis of PMN was determined in triplicate on fresh soil (0-0.1 m) by measuring the accumulation of N[H.sub.4.sup.+]-N during a 7-day anaerobic incubation (10 g dry weight equivalent: 40 mL [H.sub.2]O) at 40[degrees]C (Keeney and Bremner 1966). Inorganic N (N[H.sub.4.sup.+] and N[O.sub.2.sup.-] + N[O.sub.3.sup.-]) concentrations were determined in 0.5 M [K.sub.2]S[O.sub.4] extracts (1:4 soil:solution ratio) for all soil layers. Ammonium-N and N[O.sub.2.sup.-]-N+N[O.sub.3.sup.-]-N concentrations were analysed using a Skalar [San.sup.plus] system[TM], continuous flow colourmetric analyser. At harvest in each location, wheat plants were hand-cut (1[m.sup.2]) and assessed for total aboveground plant biomass and grain yield. Ordinary least square (OLS) regression analysis was used to assess the level of variability in grain yield that could be explained by between-location variation in PMN, applied N-fertiliser, and soil inorganic N.
Within one paddock, soil was collected from a 10-ha area using a 25 m by 25 m structured sampling grid (n = 180). This sampling structure was chosen to create a smaller sampling distance than the previously reported spatial ranges for soil N supply and to provide sufficient values (>100; Webster and Oliver 1992) from which to estimate reliable experimental variograms. Within a 1-m radius around each grid point, a composite bag of 12 soil cores (0-0.1 m) was collected using a 0.05-m-diameter push-in auger. These 'grid' samples were used to develop spatial maps of PMN measured by both biochemical and IR techniques. An additional set of 40 composite bags of soil was collected from 40 random points located within the same farm. These 'calibration' samples were used to develop a site-specific IR calibration curve to predict PMN. Biochemical analysis of PMN was performed as above on sieved (<2 mm) soil. To develop an IR calibration for PMN, subsamples of soil from the 'calibration' samples were first dried (105[degrees]C) and ground (<50 microns) in a ring mill grinder.
All spectra were acquired using a Perkin Elmer Fourier Transform IR spectrometer. Samples were scanned from 7000 to 500 nm at 1 nm resolution and each spectra was the mean of 40 scans. Spectra were converted to Grams format (.spc files) for processing using GRAMS/386 for Windows (Grams/AI, Thermo Galactic, USA). Spectra were calibrated against the traditional biochemical measurement of PMN by partial least-squares (PLS) regression. PLS involved calibration of the 40 "calibration' samples by means of repeated iterations whereby each sample in turn was left out of the calibration so that the performance of the calibration to predict an unknown sample could be evaluated (Haaland and Thomas 1988). The IR spectra of the 180 'grid' samples were acquired in a similar manner and their values for PMN were predicted from the PMN calibration developed from the spectra of the 40 'calibration' soils.
PMN values derived by biochemical analysis were positively skewed and were first [log.sub.10]-transformed to correct for skewness before analysis. The spatial module of S-PLUS 6.2 was used to correct for geometric anisotropy and to develop semivariogram models. The geostatistical analyst extension of ESRIArcMap (Version 8.3) was then used to generate spatial prediction maps of back-transformed PMN data using a spherical ordinary kriging model. A second-order polynomial was used to detrend the data. A neighbourhood search procedure was then applied to limit the number of data points used to predict the values at unsampled locations and the area was divided into 4 sectors to avoid bias in any particular direction. Standard cross-validation procedures were then used to diagnose the goodness-of-fit of the model. Mapped PMN categories were grouped according to quartile ranges.
Across the 40 paddocks PMN explained a greater variation in aboveground plant biomass than applied N fertiliser or the amount of inorganic N in the soil profile 6 weeks after sowing (Table 1). Wheat grain yield varied between 0.38 and 5.46 t/ha (mean 2.50 t/ha), while biochemical analysis of PMN varied between 2.30 and 18.55 mgN/kg (mean 8.62 mg N/kg). The range in PMN between paddocks described 21% (P = 0.003) of the variation in grain yield; a 1.0 mg N/kg change in PMN related to a 0.14t/ha change in grain yield (Table 1). This compared with only 9% (P = 0.055) of the grain yield variability being explained by N-fertiliser application (Table 1). There was no significant relationship between grain yield and the amount of inorganic N in the soil profile 6 weeks after sowing (Table 1), which varied between 3.7 and 42.9 kg N/ha (mean 13.5 kg N/ha).
[FIGURE 1 OMITTED]
The IR calibration curve (n =40) developed for PMN using the 40 'calibration' samples was highly correlated with the biochemical measurement of PMN (Fig. 1; P<0.001). However, the range of biochemical PMN values from the 'calibration' sample set (4.2-16.9mgN/kg) was narrower than that observed in the 180 'grid' samples (3.1-27.4mgN/kg), causing the IR prediction to underestimate PMN above 17mgN/kg in the 'grid' samples (Fig. 2). Thus, within the 10-ha spatial grid, the IR predicted PMN ranged from 3. I to 19.5 mg N/kg; the predicted and actual mean PMN being 10.9 and 11.7 mg N/kg, respectively.
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
For the 'grid' soils, semi-variogram models were determined for the [log.sub.10]-transformed biochemical PMN dataset (nugget 0.05, sill 0.2, range 31.8m) and for the IR-predicted PMN dataset (nugget 3.23, sill 19.70, range 37.3m). The comparison of cross-validation results from the spatial maps (Fig. 3) confirmed that there was high correlation between values of PMN determined by biochemical analysis and IR prediction (Fig. 2a; P< 0.001). The slope and intercept of this relationship could have been further improved (Fig. 2b) when PMN values >17mgN/kg were excluded to account for the lack of IR calibration above 17 mg N/kg. The cross-validation model of the spatial map using the IR-predicted PMN values had a smaller root mean square error (RMSE 2.6) than the model using the biochemical PMN dataset (RMSE 3.2).
Inorganic N application rates to winter wheat for this region typically vary between 0 and 80kgN/ha based on expected yields from anticipated growing season rainfall (200-400 mm). In this study the inorganic N fertiliser application rates were at the low end of this range (0-47 kg N/ha) and reflected the low (210 mm) growing season rainfall. Angus (2001) estimated that N fertiliser contributes ~20% of the total N supply to wheat in Australia. These low input farming systems are thus highly reliant on soil organic matter mineralisation and biological N fixation to meet crop N demand. It can therefore be expected that a stronger relationship will occur between PMN and either plant biomass or grain yield than with inorganic N fertiliser application. However, few attempts have been made to initialise fertiliser decision support systems with site-specific data of soil N supply. Such an approach certainly has an application in the strategic management of nutrients within ecosystems if a suitable soil initialising parameter can be developed which is cost-effective and rapid (Smith 2001).
Infrared technology is now starting to provide the means to rapidly predict soil properties relevant to the initialisation of soil organic matter models (Skjemstad et al. 2004) and N fertiliser recommendation systems. Demand for such technology is likely to increase as fertiliser costs continue to increase and precision agriculture capability becomes commonplace. We conclude that IR is of sufficient accuracy to enable the within-paddock spatial structure of PMN to be estimated. Findings from the semi-variogram models indicate that PMN has a range of ~30 m, indicating a spatial dependence of similar distance to that for nitrate (Dahiya et al. 1985; Stenger et al. 1998). Thus, while the use of IR to predict PMN may not reduce the sampling intensity required to accurately map a paddock, the saving in analytical and time costs will mean that such an intensive mapping exercise is more feasible, particularly as infrared devices become low-cost, robust, and portable for use within the field (Dell et al. 2007).
In this study PMN only explained 21% of the variation in grain yield. Although this is similar to the percentage of grain variation explained by indices of soil N supply in other studies (McTaggart and Smith 1993; Walley et al. 2002) it does highlight the difficulty of predicting yield variation using only I soil variable. Development of robust IR calibration sets that are able to predict a range of soil biological, chemical, and physical parameters that, in combination, better describe crop variability are required in order to manage the spatial variability in the soil resource to achieve greater yields while minimising the impact on the environment.
This research was funded by the Australian Grains Research and Development Corporation.
Manuscript received 28 April 2008, accepted 22 June 2009
Angus JF (1992) Soil N fur wheat yield and grain protein. In 'Transfer of biologically fixed nitrogen to wheat'. (Ed. J. Angus) pp. 109 116. (Grains Research and Development Corporation: Canberra, ACT)
Angus JF (2001) Nitrogen supply and demand in Australian agriculture. Australian Journal of Experimental Agriculture 41, 277-288. doi: 10.1071/EA00141
Baxter SJ, Oliver MA, Gaunt J (2003) A geostatistical analysis of the spatial variation of soil mineral nitrogen and potentially available nitrogen within an arable field. Precision Agriculture 4, 213-226. doi: 10.1023/A:1024565507688
Bertrand I, Janik LJ, Holloway RE, Armstrong RD, McLaughlin MJ (2002) The rapid assessment of concentrations and solid phase associations of macro- and micronutrients in alkaline soils by mid-infrared diffuse reflectance spectroscopy. Australian Journal of Soil Research 40, 1339-1356. doi: 10.1071/SR01104
Bundy LG, Meisinger JJ (1994) Nitrogen availability indices. In 'Methods of soil analysis. Part 2'. SSSA Book Series 5. (Eds RW Weaver, et al.) pp. 951-984. (SSSA: Madison, WI)
Cambardella CA, Moorman TB, Novak JM, Parkin TB, Karlen DL, Turco RF, Konopka AE (1994) Field scale variability of soil properties in Central Iowa soils. Soil Science Society of America Journal 58, 1501-1511.
Cookson WR, Murphy DV (2004) Quantifying the contribution of dissolved organic matter to soil nitrogen cycling using [sup.15]N isotopic pool dilution. Soil Biology & Biochemistry 36, 2097-2100. doi: 10.1016/j.soilbio. 2004.05.013
Corre MD, Schnabel RR, Stout WL (2002) Spatial and seasonal variation of gross nitrogen transformations and microbial biomass in a Northeastern US grassland. Soil Biology & Biochemistry 34, 445-457. doi: 10.1016/ S0038-0717(01)00198-5
Dahiya IS, Anlauf R, Kersebaum KC, Richter J (1985) Spatial variability of some nutrient constituents of an alfisol from loess. 2. Geostatistical analysis. Zeitschrft fur Pflanzenernahrung und Bodenkunde 148, 268-277. doi: 10.1002/jpln.19851480307
Dell JM, Keating AJ, Milne J, Antoszewski J, Musca CA, Faraone L, Murphy D, Samardzic O (2007) Micro-electromechanical systems-based microspectrometers covering wavelengths from 1500nm to 5000nm. In 'Next-generation spectroscopic technologies'. (Eds CD Brown, MA Druy, JP Coates). Proceedings of SPIE 6765, 1-9.
Giebel A, Wendroth O, Reuter HI, Kersebaum K-C, Schwarz J (2006) How representatively can we sample soil mineral nitrogen? Journal of Plant Nutrition and Soil Science 169, 52-59. doi: 10.1002/jpln.200521755
Haaland DM, Thomas VT (1988) Partial least-square methods for spectral analyses. 1. Relation to other quantitative calibration methods and the extraction of qualitative information. Analytical Chemistry 60, 1193-1202. doi: 10.1021/ac00162a020
Janik LJ, Merry RH, Skjemstad JO (1998) Can mid infrared diffuse reflectance analysis replace soil extractions? Australian Journal of Experimental Agriculture 38, 681-696. doi: 10.1071/EA97144
Keeney DR, Bremner JM (1966) Comparison and evaluation of laboratory method of obtaining an index of soil nitrogen availability. Agronomy Journal 58, 498-503.
Magdoff F, Ross D, Amadon J (1984) A soil test for nitrogen availability to com. Soil Science Society of America Journal 48, 1301-1304.
Mahmoudjafari M, Kluitenberg G J, Halvin JL, Sisson JB, Schwab AP (1997) Spatial variability of nitrogen mineralisation at the field scale. Soil Scienee Society of America Journal 61, 1214-1221.
McTaggart IP, Smith KA (1993) Estimation of potentially mineralisable nitrogen in soil by KCl extraction. II. Comparison with soil N uptake in the field. Plant and Soil 157, 175 184. doi: 10.1007/BF00011046
Mimmo T, Reeves JB, McCarty GW, Galletti G (2002) Determination of biochemical measures by mid-infrared diffuse reflectance spectroscopy in soils within a landscape. Soil Science 167, 281-287. doi: 10.1097/ 00010694-200204000-00005
Mulvaney RL, Khan SA, Hoefi RG, Brown HM (2001) A soil organic nitrogen fraction that reduces the need for nitrogen fertilisation. Soil Science Society of America Journal 65, 1164-1172.
Murphy DV, Fillery IRP, Spading GP (1998) Seasonal fluctuations in gross N mineralisation, ammonium consumption and microbial biomass in a Western Australian soil under different land use. Australian Journal q/" Agricultural Research 49, 523 535. doi: 10.1071/A97096
Picone LI, Cabrera ML, Franzluebbers AJ (2002) A rapid method to estimate potentially mineratisable nitrogen in soil. Soil Science Society of America Journal 66, 1843-1847.
Robettson GP, Crum JR, Ellis BG (1993) The spatial variability of soil resources following long-term disturbance. Oecologia 96, 451-456. doi: 10.1007/BF00320501
Russell CA, Angus JF, Batten GD, Dunn BW, Williams RL (2002) The potential of NIR spectroscopy to predict nitrogen mineralisation in rice soils. Plant and Soil 247, 243 252. doi: 10.1023/A:1021532316251
Shepherd KD, Walsh MG (2002) Development of reflectance spectral libraries for characterisation of soil properties. Soil Science Society of America Journal 66, 988-998.
Skjemstad JO, Spouncer LR, Cowie B, Swirl RS (2004) Calibration of the Rothamsted organic carbon turnover model (RothC ver. 26.3), using measurable soil organic carbon pools. Australian Journal of Soil Research 42, 79-88. doi: 10.1071/SR03013
Smith JU (2001) Prediction of soil nitrogen supply by combined modelling and measurement. In 'Proceedings of the SCI Meeting on Assessing Soil Fertility, Modelling, Measurement or both?' London, 20 March 2001 (SCI Agriculture and Environment Group: London)
Stenger R, Priesack E, Beese F (1998) Distribution of inorganic nitrogen in agricultural soils at different dates and scales. Nutrient Cycling in Agroecosystems 50, 291-297. doi: 10.1023/A:1009712916767
Stockdale EA, Rees RM (1994) Relationships between biomass nitrogen and nitrogen extracted by other nitrogen availability methods. Soil Biology & Biochemistry 26, 1213-1220. doi: 10.1016/0038-0717(94)90146-5
Walley F, Yates T, Van Groenigen JW, Van Kessel C (2002) Relationships between soil nitrogen availability indices, yield, and nitrogen accumulation of wheat. Soil Science Society of America Journal 66, 1549-1561.
Webster R, Oliver MA (1992) Sample adequately to estimate variograms of soil properties. Journal of Soil Science 43, 177-192. doi: 10.1111/ j.1365-2389.1992.tb00128.x
D.V. Murphy (A,C) M. Osman (A), C.A. Russell (B), S. Darmawanto (A), and F.C. Hoyle (A)
(A) Soil Biology Group, School of Earth and Environment, The University of Western Australia, Crawley, WA 6009, Australia.
(B) Centre of Excellence in Natural Resource Management, Faculty of Natural and Agricultural Sciences, Albany, WA 6330, Australia.
(C) Corresponding author. Email: email@example.com
Table 1. Ordinary least square (OLS) regression models of plant biomass and grain yield as affected by applied N fertiliser (kg N/ha), soil inorganic N (kg N/ha, 0-0.9 m), and biochemical analysis of potentially mineralisable N (PMN, mgN-kg; 0-0.1 m) across 40 paddocks Values in parentheses are standard errors of the mean (n = 40); ([dagger]) P < 0.1, ** P < 0.01 Parameter N-fertiliser Plant biomass Grain yield Coefficient 0.039 (0.022) ([dagger]) 0.021 (0.011) ([dagger]) Constant 5.28 (0.60) ** 2.04 (0.29) ** [R.sup.2] 0.08 0.09 F statistic 3.14 ([dagger]) 3.93 * Parameter Inorganic-N Plant biomass Grain yield Coefficient 0.027 (0.046) 0.014 (0.022) Constant 5.76 (0.73) ** 2.32 (0.35) ** [R.sup.2] 0.01 0.01 F statistic 0.33 0.37 Parameter PMN Plant biomass Grain yield Coefficient 0.323 (0.090) ** 0.144 (0.045) ** Constant 3.34 (0.84) ** 1.26 (0.42) ** [R.sup.2] 0.25 0.2 F statistic 12.80 ** 10.22 **
|Printer friendly Cite/link Email Feedback|
|Author:||Murphy, D.V.; Osman, M.; Russell, C.A.; Darmawanto, S.; Hoyle, F.C.|
|Publication:||Australian Journal of Soil Research|
|Date:||Nov 1, 2009|
|Previous Article:||Nitrogen and carbon cycling in a New Zealand pumice soil under a manuka (Leptospermum scoparium) and kanuka (Kunzea ericoides) shrubland.|
|Next Article:||Comparison of bicarbonate-extractable soil phosphorus measured by ICP-AES and colourimetry in soils of south-eastern New South Wales.|