Australian grain production depends on applied fertiliser, particularly nitrogen (N) and phosphorus (P), and to a lesser extent potassium (K) and sulfur (S). Despite this dependence, soil testing is used sparingly as a tool to underpin fertiliser decisions. Some grain producers typically conduct soil tests at least once every 3 years on a selection of individual fields, but it is broadly understood that many grain producers use soil testing rarely or not at all. The choice by many grain producers not to support fertiliser decisions by soil testing relates to several factors. One key factor has been a perception that soil test interpretation criteria, previously published separately before collation by K. I. Peverill, L. A. Sparrow, and D. J. Reuter, may be biased or unreliable. The current paper provides an overview of research findings, presented in this special edition of Crop & Pasture Science, describing a national approach to the collation of all available and statistically valid N, P, K, and S response trials for cereal, oilseed, and pulse crops in Australia. It provides an overview of the process adopted to make this single national dataset available to both the grains and fertiliser industries. The process to build adoption has formed an integral component of the approach, as calibration data derived from the national database are being used to underpin soil test interpretation as part of fertiliser recommendations made through Fertcare to grain producers in Australia.

During the past 50 years, 3800 field experiments yielding over 5200 treatment series were conducted in Australia examining yield responses to applied N, P, K, or S fertiliser applications to cereal, oilseed and pulse crops. The experiments all had accompanying soil test data. These data were entered into multiple Microsoft Access^® database templates and then consolidated into a single national online MYSQL^® database. A web application (named the BFDC Interrogator) was also developed to rapidly access the national database (BFDC National Database) and construct soil test calibrations between percentage of the maximum grain yield achieved (hereafter called percentage relative yield) and soil test values recorded for specified ranges of regional or national experiments. Search parameters were applied to define soil test calibrations. These included farming system (dryland or irrigated), year of experiment, soil type, crop type, soil test, depth of soil sampling and soil test units. Other data filters based on site metadata, such as method of nutrient placement, can be applied to enable more definitive calibrations.

The calibrations are used to derive critical soil test values at 80, 90 and 95% relative crop yield with 95% confidence limits. However, the soil test criteria at 90% relative crop yield with 70% confidence limits have been chosen as the single calibration and reliability standard for all crops and soil tests. Corresponding yield increase (t/ha)–soil test relationships for an applied nutrient can also be accessed.

The BFDC National Database and BFDC Interrogator can now be accessed online by trained, registered users. This paper describes the methodologies that underpinned the progressive development of this tool. Through the commitment of the grains and fertiliser industries, it is anticipated that the calibrations will be used to improve decision support systems used to generate fertiliser recommendations for Australian cropping industries.

Comprehensive data on grain yield responsiveness to applications of the major nutrients nitrogen, phosphorus, potassium, sulfur in Australian cropping experiments have been assembled in the Better Fertiliser Decisions for Cropping (BFDC) National Database for scrutiny by the BFDC Interrogator. The database contains the results of individual field experiments on nutrient response that need to be collectively integrated into a model that predicts probable grain yield response from soil tests. The potential degree of grain yield responsiveness (relative yield, RY%) is related to nutrient concentration in the soil (soil test value, STV) across a range of experimental sites and conditions for each nutrient. The RY% is defined as RY = Y₀/Y_max *100, where Y₀ is the yield without applied nutrient, and Y_max is the yield which could be attained through adequate application of the nutrient, given sufficiency of all other nutrients. The raw data for RY and STV are transformed so that a linear regression model can be applied. The BFDC Interrogator uses the arcsine-log calibration curve (ALCC) algorithm to estimate a critical soil test value (CSTV) for a given nutrient. The CSTV is defined as the value that would, on average for the broad agronomic circumstances of the incoming crop, lead to a specified percentage of Y_max (e.g. RY = 90%) without any application of that nutrient. This paper describes the ALCC algorithm, which has been developed to ensure that such estimated CSTVs, with safeguards, are reliable and to as high a precision as is realistic.

More than 1200 wheat and 120 barley experiments conducted in Australia to examine yield responses to applied nitrogen (N) fertiliser are contained in a national database of field crops nutrient research (BFDC National Database). The yield responses are accompanied by various pre-plant soil test data to quantify plant-available N and other indicators of soil fertility status or mineralisable N. A web application (BFDC Interrogator), developed to access the database, enables construction of calibrations between relative crop yield ((Y₀/Y_max) × 100) and N soil test value. In this paper we report the critical soil test values for 90% RY (CV90) and the associated critical ranges (CR90, defined as the 70% confidence interval around that CV90) derived from analysis of various subsets of these winter cereal experiments.

Experimental programs were conducted throughout Australia’s main grain-production regions in different eras, starting from the 1960s in Queensland through to Victoria during 2000s. Improved management practices adopted during the period were reflected in increasing potential yields with research era, increasing from an average Y_max of 2.2 t/ha in Queensland in the 1960s and 1970s, to 3.4 t/ha in South Australia (SA) in the 1980s, to 4.3 t/ha in New South Wales (NSW) in the 1990s, and 4.2 t/ha in Victoria in the 2000s. Various sampling depths (0.1–1.2 m) and methods of quantifying available N (nitrate-N or mineral-N) from pre-planting soil samples were used and provided useful guides to the need for supplementary N. The most regionally consistent relationships were established using nitrate-N (kg/ha) in the top 0.6 m of the soil profile, with regional and seasonal variation in CV90 largely accounted for through impacts on experimental Y_max. The CV90 for nitrate-N within the top 0.6 m of the soil profile for wheat crops increased from 36 to 110 kg nitrate-N/ha as Y_max increased over the range 1 to >5 t/ha. Apparent variation in CV90 with seasonal moisture availability was entirely consistent with impacts on experimental Y_max. Further analyses of wheat trials with available grain protein (∼45% of all experiments) established that grain yield and not grain N content was the major driver of crop N demand and CV90.

Subsets of data explored the impact of crop management practices such as crop rotation or fallow length on both pre-planting profile mineral-N and CV90. Analyses showed that while management practices influenced profile mineral-N at planting and the likelihood and size of yield response to applied N fertiliser, they had no significant impact on CV90. A level of risk is involved with the use of pre-plant testing to determine the need for supplementary N application in all Australian dryland systems. In southern and western regions, where crop performance is based almost entirely on in-crop rainfall, this risk is offset by the management opportunity to split N applications during crop growth in response to changing crop yield potential. In northern cropping systems, where stored soil moisture at sowing is indicative of minimum yield potential, erratic winter rainfall increases uncertainty about actual yield potential as well as reducing the opportunity for effective in-season applications.

The phosphorus (P) status of 535 surface soils from all states of Australia was assessed using the following soil P tests: Colwell-P (0.5 m NaHCO₃), Olsen-P (0.5 m NaHCO₃), BSES-P (0.005 m H₂SO₄), and Mehlich 3-P (0.2 m CH₃COOH   0.25 m NH₄NO₃   0.015 m NH₄F   0.013 m HNO₃   0.001 m EDTA). Results were correlated with soil P assays selected to estimate the following: soil solution P concentration (i.e. 0.01 m CaCl₂ extractable P; Colwell-P/P buffer index); rate of P supply to the soil solution (i.e. P released to FeO-impregnated filter paper); sorbed P (i.e. Colwell-P); mineral P (i.e. fertiliser reaction products and/or soil P minerals estimated as BSES-P minus Colwell-P); the diffusive supply of P (i.e. P diffusing through a thin gel film, DGT-P); and P buffer capacity (i.e. single-point P buffer index corrected for Colwell-P, PBI_Col). Across all soils, Colwell-P and BSES-P were highly correlated with FeO-P (r = 0.76 and 0.58, respectively). Colwell-P was moderately correlated with mineral P (r = 0.24), but not solution P. Olsen-P and Mehlich-P were both highly correlated with FeO-P (r = 0.80 and 0.78, respectively) but, in contrast to Colwell-P and BSES-P, also showed moderate correlations with soil solution P (r = 0.29 and 0.34, respectively) and diffusive P supply (r = 0.31 and 0.49, respectively). Correlation coefficients with mineral P were r = 0.29 for Olsen-P and r = 0.17 for Mehlich-P. Soils were categorised according to their pH, clay activity ratio, content of mineral P and CaCO₃ content, and the relationships between the empirical soil P tests examined for each soil category. Olsen-P and Colwell-P were correlated across all soil categories (r range 0.66–0.90), and a widely applicable linear equation was obtained for converting one soil test to the other. However, the correlations between other soil tests varied markedly between soil categories and it was not possible to develop such widely applicable conversion equations. Multiple step-up linear regressions were used to identify the key soil properties affecting soil solution P, P buffer capacity, and diffusive P supply, respectively. For all soil categories, solution P concentration (measured by CaCl₂-P) increased as rate of P supply (measured as FeO-P) increased and P buffer capacity decreased. As an assay of sorbed P, Colwell-P alone did not significantly (P > 0.05) explain any of the variability in soil solution P, but when used in the index (Colwell-P/P buffer index), it was highly correlated (r = 0.74) with CaCl₂-P. Soil P buffer capacity was dependent on different properties in different soil categories, with 45–65% of the variation in PBI accounted for by various combinations of Mehlich-Al, Mehlich-Fe, total organic C, clay content, clay activity ratio, and CaCO₃ content, depending on soil category. The diffusive supply of P was primarily determined by rate of P supply (measured as FeO-P; r range 0.34–0.49), with significant (P < 0.05) small improvements due to the inclusion of PBI_Col and/or clay content, depending on soil category. For these surface soil samples, key properties of pH, clay activity ratio, clay content, and P buffer capacity varied so widely within individual Australian Soil Orders that soil classification was not useful for inferring intrinsic

The performance of a wide range of soil phosphorus (P) testing methods that included established (Colwell-P, Olsen-P, BSES-P, and CaCl₂-P) and more recently introduced methods (DGT-P and Mehlich 3-P) was evaluated on 164 archived soil samples corresponding to P fertiliser response experiments with wheat (Triticum aestivum) conducted in south-eastern Australia between 1968 and 2008. Soil test calibration relationships were developed for relative grain yield v. soil test using (i) all soils, (ii) Calcarosols, and (iii) all ‘soils other than Calcarosols’. Colwell-P and DGT-P calibration relationships were also derived for Calcarosols and Vertosols containing measureable CaCO₃. The effect of soil P buffer capacity (measured as the single-point P buffer index corrected for Colwell-P, PBI_Col) on critical Colwell-P values was assessed by segregating field sites based on their PBI_Col class: very very low (15–35), very low (36–70), low (71–140), and moderate (141–280). All soil P tests, except Mehlich 3-P, showed moderate correlations with relative grain yield (R-value ≥0.43, P < 0.001) and DGT-P exhibited the largest R-value (0.55). Where soil test calibrations were derived for Calcarosols, Colwell-P had the smallest R-value (0.36), whereas DGT-P had an R-value of 0.66. For ‘soils other than Calcarosols’, R-values >0.45 decreased in the order: DGT-P (r = 0.55), Colwell-P (r = 0.49), CaCl₂-P (r = 0.48), and BSES-P (r = 0.46). These results support the potential of DGT-P as a predictive soil P test, but indicate that Mehlich 3-P has little predictive use in these soils. Colwell-P had tighter critical confidence intervals than any other soil test for all calibrations except for soils classified as Calcarosols. Critical Colwell-P values, and confidence intervals, for the very very low, very low, and low P buffer capacity categories were within the range of other published data that indicate critical Colwell-P value increases as PBI_Col increases. Colwell-P is the current benchmark soil P test used in Australia and for the field trials in this study. With the exception of Calcarosols, no alternative soil P testing method was shown to provide a statistically superior prediction of response by wheat. Although having slightly lower R-values (i.e. <0.1 difference) for some calibration relationships, Colwell-P yielded tighter confidence intervals than did any of the other soil tests. The apparent advantage of DGT-P over Colwell-P on soils classified as Calcarosols was not due to the effects of calcium carbonate content of the analysed surface soils.

Soil testing is the most widely used tool to predict the need for fertiliser phosphorus (P) application to crops. This study examined factors affecting critical soil P concentrations and confidence intervals for wheat and barley grown in Australian soils by interrogating validated data from 1777 wheat and 150 barley field treatment series now held in the BFDC National Database. To narrow confidence intervals associated with estimated critical P concentrations, filters for yield, crop stress, or low pH were applied. Once treatment series with low yield (<1 t/ha), severe crop stress, or pH_CaCl2 <4.3 were screened out, critical concentrations were relatively insensitive to wheat yield (>1 t/ha). There was a clear increase in critical P concentration from early trials when full tillage was common compared with those conducted in 1995–2011, which corresponds to a period of rapid shift towards adoption of minimum tillage. For wheat, critical Colwell-P concentrations associated with 90 or 95% of maximum yield varied among Australian Soil Classification (ASC) Orders and Sub-orders: Calcarosol, Chromosol, Kandosol, Sodosol, Tenosol and Vertosol. Soil type, based on ASC Orders and Sub-orders, produced critical Colwell-P concentrations at 90% of maximum relative yield from 15 mg/kg (Grey Vertosol) to 47 mg/kg (Supracalcic Calcarosols), with other soils having values in the range 19–27 mg/kg. Distinctive differences in critical P concentrations were evident among Sub-orders of Calcarosols, Chromosols, Sodosols, Tenosols, and Vertosols, possibly due to differences in soil properties related to P sorption. However, insufficient data were available to develop a relationship between P buffering index (PBI) and critical P concentration. In general, there was no evidence that critical concentrations for barley would be different from those for wheat on the same soils. Significant knowledge gaps to fill to improve the relevance and reliability of soil P testing for winter cereals were: lack of data for oats; the paucity of treatment series reflecting current cropping practices, especially minimum tillage; and inadequate metadata on soil texture, pH, growing season rainfall, gravel content, and PBI. The critical concentrations determined illustrate the importance of recent experimental data and of soil type, but also provide examples of interrogation pathways into the BFDC National Database to extract locally relevant critical P concentrations for guiding P fertiliser decision-making in wheat and barley.

Australian cropping systems are dominated by winter cereals; however, grain legumes, oilseeds and summer cereals play an important role as break crops. Inputs of phosphorus (P) fertiliser account for a significant proportion of farm expenditure on crop nutrition, so effective fertiliser-use guidelines are essential. A national database (BFDC National Database) of field experiments examining yield responses to P fertiliser application has been established. This paper reports the results of interrogating that database using a web application (BFDC Interrogator) to develop calibration relationships between soil P test (0–10 cm depth; Colwell NaHCO₃ extraction) and relative grain yield. Relationships have been developed for all available data for each crop species, as well as for subsets of those data derived by filtering processes based on experiment quality, presence of abiotic or biotic stressors, P fertiliser placement strategy and subsurface P status.

The available dataset contains >730 entries but is dominated by data for lupin (Lupinus angustifolius; 62% of all P experiments) from the south-west of Western Australia. The number of treatment series able to be analysed for other crop species was quite small (<50–60 treatment series) and available data were sometimes from geographic regions or soil types no longer reflective of current production. There is a need for research to improve information on P fertiliser use for key species of grain legumes [faba bean (Vicia faba), lentil (Lens culinaris), chickpea (Cicer arietinum)], oilseeds [canola (Brassica napus), soybean (Glycine max)] and summer cereals [sorghum (Sorghum bicolor), maize (Zea mays)] in soils and farming systems reflecting current production.

Interrogations highlighted the importance of quantifying subsurface P reserves to predict P fertiliser response, with consistently higher 0–10 cm soil test values required to achieve 90% maximum yield (CV90) when subsurface P was low (<5 mg P/kg). This was recorded for lupin, canola and wheat (Triticum aestivum). Crops grown on soils with subsurface P >5 mg/kg consistently produced higher relative yields than expected on the basis of a 0–10 cm soil test. The lupin dataset illustrated the impact of improving crop yield potentials (through more effective P-fertiliser placement) on critical soil test values. The higher yield potentials arising from placement of P-fertiliser bands deeper in the soil profile resulted in significantly higher CV90 values than for crops grown on the same sites but using less effective (shallower) P placement. This is consistent with deeper bands providing an increased and more accessible volume of profile P enrichment and supports the observation of the importance of crop P supply from soil layers deeper than 0–10 cm.

Soil P requirements for different species were benchmarked against values determined for wheat or barley (Hordeum vulgare) grown in the same regions and/or soil types as a way of extrapolating available data for less researched species. This approach suggested most species had CV90 values and ranges similar to winter cereals, with evidence of different soil P requirements in only peanut (Arachis hypogaea – much lower) and field pea (Pisum sativum – slightly higher). Unfortunately, sorghum data were so limited that benchmarking against wheat was inconclusive.

The Better Fertiliser Decision for Crops (BFDC) National Database holds historic data for 356 potassium (K) fertiliser rate experiments (431 treatment series) for different rain-fed grain crops and soil types across Australia. Bicarbonate-extractable K (Colwell soil-test K) is the most extensively used soil test reported in the database. Data are available for several crop species grown on a range of soil types from all states except Tasmania. Species represented and number of treatment series in the database are: wheat (Triticum aestivum L.), 254; barley (Hordeum vulgare L.), 5; canola (Brassica napus L.), 130; lupin (Lupinus angustifolius L.), 32; sunflower (Helianthus annuus L.), 10; sorghum (Sorghum bicolor L.), 5; and faba bean (Vicia faba L.), 2. About 77% of the available soil-test K (STK) data on wheat, canola, and lupin are from Western Australia. The usual sampling depth of 0–10 cm is recorded for all treatment series within the database, while 68% of experiments have STK information from other soil horizons down the profile, usually in 10-cm increments.

The BFDC Interrogator, a comprehensive data search and calibration support tool developed for use with the BFDC National Database, was used to examine STK–yield relationships for each crop across Australia, with more detailed analysis by state/region and then by soil type if data were available. The BFDC Interrogator was used to determine a critical STK concentration to achieve 90% of the maximum relative yield (90%RY) for each crop species, with a critical range (determined by the 70% confidence limit for the 90%RY) also reported.

The STK for 90%RY for wheat was 40–41 mg/kg on Tenosols and Chromosols, ∼49 mg/kg on Kandosols, and ∼64 mg/kg on Brown Ferrosols. There was some evidence of critical values increasing with increasing crop yield and on soils with no acidity constraints to root growth, with effects presumably driven by increased crop K demand. The STK for 90%RY for canola, grown mainly on Tenosols, was similar to that for wheat, ranging from 43 to 46 mg K/kg, but for lupin, also grown mainly on Tenosols, the STK for 90%RY was a relatively low ∼25 mg K/kg. Data for sunflower were limited and the STK for 90%RY was poorly defined. A comparison of critical STK concentrations for different crops grown on Tenosols suggested that critical ranges for 90%RY of lupin (22–27 mg K/kg) were significantly lower than that for wheat (32–52 mg K/kg) and canola (44–49 mg K/kg). Critical STK values were not determined for sorghum and faba bean.

Accurate definition of the sulfur (S) soil test–crop grain yield increase (response) relationship is required before soil S test measurements can be used to if there are likely to be responses to S fertilisers. An analysis was done using the Better Fertiliser Decision for Crops (BFDC) National Database using a web application (BFDC Interrogator) to develop calibration relationships between soil S tests (KCl-40 and MCP) using a selection of sampling depths and grain relative yields (RY). Critical soil test values (CSTV) and critical soil test ranges (CSTR) were defined at RY 90%. The ability of the KCl-40 extractable S soil test to predict grain yield response to applied S fertiliser was examined for wheat (Triticum aestivum L.) grown in Western Australia (WA), New South Wales (NSW), and Victoria and canola (Brassica napus L.) grown in WA and NSW. A smaller dataset using MCP_i-extractable S was also assessed.

The WA-grown wheat KCl-40 S CSTV, using sampling depth to 30 cm for soil types Chromosols (Coloured), Chromosols (Sesqui-Nodular), Kandosols (Grey and Yellow), Tenosols (Brown and Yellow), and Tenosols (Grey, Sesqui-Nodular), was 2.8 mg kg^–1 with an associated CSTR 2.4–3.2 mg kg^–1 and a correlation coefficient (r) 0.87. Similarly, KCl-40 S CSTV was defined using sampling depth to 10 cm for these selected soil types and for wheat grown on Vertosols in NSW. The accuracy of the KCl-40 S CSTV for canola grown in WA was improved using a sampling to a depth of 30 cm instead of 10 cm for all soil types. The canola KCl-40 S CSTV using sampling depth to 30 cm for these soil types was 7.2 mg kg^–1 with an associated CSTR 6.8–7.5 and an r value 0.70. A similar KCl-40 S CSTV of 7.0 mg kg^–1 was defined using a sampling depth of 10 cm, but the CSTR was higher (6.4–7.7 mg kg^–1) and the r value lower (0.43). A lower KCl-40 S CSTV of 3.9 mg kg^–1 or 31.0 kg ha^–1 using a sampling depth of 60 cm was defined for canola grown in NSW using a limited number of S-rate calibration treatment series. Both MCP_i (r = 0.32) and KCl-40 (r <0.20) soil S test–NSW canola response relationships using a 0–10 cm sampling depth were weak.

The wheat KCl-40 S CSTR of 2.4–3.2 mg kg^–1 can be used widely on soil types where soil sulfate is not leached during the growing season. However, both the WA canola CSTR of 6.4–7.2 mg kg^–1 using a sampling depth of 30 cm and NSW canola CSTR of 25–39 kg ha^–1 or 3.1–4.9 mg kg^–1 using a sampling depth of 60 cm can be considered in regions outside of WA and NSW.

Scientists readily suggest that research and development is not complete until findings and conclusions are reported in the peer-reviewed literature. The authors suggest that industry-specific relevant research and development is actually not complete until the key outputs are extended to primary stakeholder groups. In the case of ‘Making Better Fertiliser Decisions for Cropping Systems in Australia’ (BFDC), this meant training key members of the grains and fertiliser industries, where nitrogen, phosphorus, potassium, and sulfur recommendations are derived from soil tests and provided to farmers.

The BFDC project applied a two-part cascading approach to extension and training to reach the grains and fertiliser industries. The train-the-trainer program (Level 1) was undertaken and targeted at influential and experience persons such as technical leaders and those who influenced crop-nutrition decision support systems (DSS) within interested organisations. Level 2 activities targeted individuals within organisations who had direct discussion with farmers about soil testing, and their training was facilitated by a Level 1 trained colleague.

Development of extension plans, training course structure, and training resources was conducted in parallel with the development of the BFDC National Database and BFDC Interrogator from the commencement of the project. In so doing, it was agreed that controlled access to the information should be established to maintain a consistent standard of use and to provide a platform for gathering feedback to guide future developments (e.g. of the BFDC Interrogator or prioritising future experimental investment).

The BFDC extension approach targeted 100 individuals from the Level 1 audience and 30 individuals from the Level 2 audience through seven train-the-trainer level workshops conducted across Australia. As a result of reaching this audience, it was expected that the cumulative effect of the cascading extension strategy and input of BFDC Interrogator critical nutrient concentrations in commercial soil analysis interpretation software would indirectly and directly influence the crop nutrient management decisions of up to 5000 Australian grain farmers in the year after the commencement of training activities.

Exit survey results, conducted as part of the BFDC train-the-trainer workshops, were aggregated across all Level 1 workshops. These survey results showed that the most significant benefit of the training was the usefulness of the BFDC Interrogator training manual and the additional teaching aids supplied. The importance and significance to the industry of the collation of data through BFDC was also highlighted as a benefit.

Reflecting industry knowledge gaps (e.g. crop × nutrient × geographic region interactions), the lowest ranking survey results focused on the ability of the BFDC National Database to provide usable critical soil test criteria for several situations. Yet despite this reported shortcoming, participants recognised the significance of the structure, tools, skills, and knowledge gained through the training workshop and the importance of the established critical levels.

While face-to-face training enables robust discussion, the ‘time-poor’ nature of roles for agricultural professionals appeared to limit the uptake of training opportunities. Therefore, training materials are being developed into an online course focused particularly on university requirements and the development of agricultural professionals.

Soil testing remains a most valuable tool for assessing the fertiliser requirement of crops. The relationship between soil tests (generally taken from surface soil) and relative yield (RY) response to fertiliser is subject to the influence of environment (e.g. water, temperature) and management (e.g. cultivation, sowing date). As such, the degree of precision is often low when the soil test calibration is based on a wide range of independent experiments on many soil types over many years by many different operators. Hence, the 90% RY target used in soil test interpretation is best described by a critical range (critical concentration and confidence interval) for a given soil test rather than a single critical value. The present Better Fertiliser Decisions for Crops (BFDC) National Database, and the BFDC Interrogator that interacts with the database, provide a great advance over traditional formats and experiment-specific critical values because it allows the use of filters to refine the critical range for specific agronomic conditions. However, as searches become more specific (region, soil type) the quantity of data available to estimate a critical range becomes more vulnerable to data paucity, to outliers, and to clusters of localised experiments. Hence, appropriate training of the users of this database will ensure that the strengths and limitations of the BFDC National Database and BFDC Interrogator are properly understood. Additionally, the lack of standardised metadata for sites within the database makes it generally impossible to isolate the effects on critical values of the specific management or environmental factors listed earlier, which are therefore best determined by specific studies. Finally, the database is dominated (60%) by responses of wheat to nitrogen and phosphorus, meaning that relatively few studies are available for responses by pulses (other than narrow leaf lupins) or oilseeds (other than canola), especially for potassium and sulfur. Moreover, limited data are available for current cropping systems and varieties. However, the identification of these gaps can now be used to focus future research on the crops, nutrients, soils, regions, and management practices where data are lacking. The value of metadata and the need for standardised protocols for nutrition experiments were key lessons.