Detecting soil nitrogen supplies by canopy sensing. proof of concept

Transcription

1 Project Report No. 427 January Detecting soil nitrogen supplies by canopy sensing proof of concept by R. Sylvester-Bradley 1, J.J.J. Wiltshire 1, D.R. Kindred 1, D.L.J. Hatley 2 and J. Wilson 3 1 ADAS Boxworth, Boxworth, Cambridge, CB23 4NN 2 ADAS Terrington, Bentinck Farm, Rhoon Road, Terrington St Clement, King s Lynn, Norfolk. PE34 4HZ 3 Soil Essentials Ltd, Hilton of Fern, By Brechin, Angus, Scotland. DD9 6SB This is the final report of a twelve month project, which started in October 26. The project was funded by a contract of 15, from HGCA (Project No. 3285), with 5, each from Soil Essentials Ltd and Yara UK Ltd, making a total of 25,. The Home-Grown Cereals Authority (HGCA) has provided funding for this project but has not conducted the research or written this report. While the authors have worked on the best information available to them, neither HGCA nor the authors shall in any event be liable for any loss, damage or injury howsoever suffered directly or indirectly in relation to the report or the research on which it is based. Reference herein to trade names and proprietary products without stating that they are protected does not imply that they may be regarded as unprotected and thus free for general use. No endorsement of named products is intended nor is it any criticism implied of other alternative, but unnamed, products.

2 ABSTRACT... 1 SUMMARY... 2 TECHNICAL DETAIL INTRODUCTION MATERIALS & METHODS SITES SENSORS DATA COLLECTION AND ANALYSIS RESULTS SOIL MINERAL N OVER-WINTER CROP GROWTH CANOPY REFLECTANCE... 2 CHANGE IN CANOPY REFLECTANCE OTHER FACTORS INFLUENCING REFLECTANCE DISCUSSION ACKNOWLEDGEMENTS REFERENCES ii

3 Abstract This research tested whether sensing the crop can indicate soil nitrogen supplies. Nitrogen fertiliser experiments on cereals were established at four sites in In the following year, commercial cereal crops were grown. In 26-7 at each site, the plot positions as used in the previous year were marked out once again for crop testing using a sensor. Reflectance was measured monthly, December to March, during tillering using a Crop Circle instrument (provided by Soil Essentials Ltd) in the 26-7 plots relating to the 25-6 experiments. This comprised two sensors to measure reflectance at 88 nm (near-infrared, NIR) and 59 nm (orange). A Normalised Difference Vegetation Index () was calculated to give a measure of vegetation cover. Soil mineral N data were obtained from another study at the same sites. Yields of grain and straw and their N concentrations were measured to give total N uptake. Data were interpreted for relationships between canopy reflectance and soil N. Establishment and tillering were early and rapid at Boxworth and Rosemaund, but later and slower at Terrington and High Mowthorpe. The highest level of N applied in 26 had a small effect on soil mineral N at Terrington and Rosemaund, but a large effect at Boxworth. Using total N uptake at harvest as an estimate of soil N supply, the largest response to N supply was also at Boxworth. Over-fertilisation only led to high SMN residues at Boxworth, which had the lowest optimum N requirement in 25-6 : 117 kg/ha, compared to, 22 to 237 kg/ha at other sites). Use of the sensor successfully detected the high SMN residue at Boxworth using the relationship between and soil N supply, especially at soil mineral N values below 15 kg/ha. This relationship was less variable and more useful (shallower slope) when N uptake (rather than soil mineral N) was used to estimate soil N supply. The relationships improved with later assessment of. Change in between assessment dates showed that canopies sometimes declined and sometimes grew during winter. However, change in was less useful for predicting soil N supply than absolute values of. In conclusion at Boxworth, but not at three other sites, differentiated plots in need of fertiliser (soil mineral N up to around 15 kg/ha N) from those with ample soil N supplies. 1

4 Summary Introduction Judging soil N supplies is the most crucial, yet the most uncertain, element of judging fertiliser requirements on the farm. The best current method of judging soil N supplies involves field by field analysis of soil, to determine soil mineral N, to at least 6 cm depth. However, previous work has shown that, across a number of experiments, the crop recovers slightly more soil N than is measured as soil mineral N in February to 9cm depth. Furthermore, sampling and analysis are laborious and expensive, and the results are not regarded with high confidence by the industry. Most farmers therefore use look-up tables (e.g. in Defra booklet RB29), which may give even more imprecise results. The research described here was undertaken to assess whether the new technology of canopy sensing might revitalise an old idea that the crop could be used as a barometer of the soil. If proven, this approach could be applied to most cereal fields with relatively little investment because tractor- or sprayer-mounted canopy sensors are now available at affordable prices (~ 3, each). Methods In winter 26-7, four cereal sites (Boxworth, Terrington, Rosemaund and High Mowthorpe) were identified where there had been N fertiliser experiments in 26 with six N levels under HGCA Project 384, and where Defra was funding analyses of soil mineral N and crop N content (through Defra Project IS223). As part of Defra Project IS223, after sowing a cereal crop in autumn, all plots were re-marked out in the positions of the N fertiliser experiment plots in the previous season (25-26). The treatments are summarised in Table 1. There were three replicates. Crops grown in are listed in Table 2. Plots received no N applications in the 26/27 season. 2

5 Table 1 Sites, wheat varieties for N response plots and N treatments (total N applied) in Site Varieties Total N applied (kg/ha N) Boxworth XI19, 7, 15, 21, 22, 29, 36, 37 Longbow, Norman,, 7, 15, 22, 29, 37 Alchemy, Gladiator Terrington XI19, 7, 14, 21, 28, 35 Longbow, Virtue,, 7, 15, 22, 29, 37 Ambrosia, Gladiator Rosemaund XI19,, 7, 15, 22, 29, 37 Longbow, Hustler, Ambrosia, Robigus High Mowthorpe XI19, Longbow, Norman, Ambrosia, Istabraq, 7, 14, 21, 28, 35 Table 2 Sites and details of crops grown in Site Crop Variety Sowing date Emergence date Boxworth Winter wheat Nijinsky 26 Sep 26 9 Oct 26 Terrington Winter wheat Nijinski 27 Oct Nov 26 Rosemaund Winter barley Saffron 2 Sep 26 1 Oct 26 High Mowthorpe Winter wheat Gladiator 3 Sep Oct 26 At each site two additional plots (2 m wide by ~1 m long) were made in a discard area adjacent to the trial and reflectance data were collected as for the cereal plots from the previous N experiments. One plot was bare soil, and the other was the crop with additional N (2 kg/ha N in November and 2 kg/ha N in January) to give unlimited growth. Also, to give an indication of varying light quality, a mat of green plastic was taken to each site at each measurement date, and reflectance data were collected as for the cereal plots. The Crop Circle instrument, provided by Soil Essentials Ltd, comprised two sensors, each with its own light source (to minimise variation due to variation in ambient light 3

6 conditions). The sensors measured reflectance at 88 nm (near-infrared, NIR) and 59 nm (orange). Reflectance was measured on a monthly basis throughout the period of tillering December to March. Normalised Difference Vegetation Index () was calculated from the sensor data. gives a measure of vegetative cover. Dense canopies strongly absorb red wavelengths and reflect near-infrared wavelengths. is a measure of the difference in reflectance between these wavelength ranges, and high values (to a maximum of 1) indicate dense canopies. was calculated using reflectance intensity data for the two measured wavelengths, as follows. = (NIR-visible/(NIR+visible) In this work we used light of wavelength 59 nm as the visible light for this calculation. Soil mineral N data were obtained from another study using the same site. For plots that had grown varieties other than Xi19 in the previous season, and that had the same previous N applications, plots were paired (paired varieties are on the same lines in Table 1) and one value was obtained for a pair of plots. Reflectance data were also paired in the same way, so that a pair of plots in became one plot in Yields were recorded at crop maturity using a plot combine harvester. Samples of grain were submitted to a laboratory for grain N analysis, and grain uptake was calculated. Grab samples were also taken and straw and chaff were submitted to a laboratory for N analysis, from a sub-set of plots at each site. Nitrogen harvest indices were calculated for each of these plots, and were found to be similar for contrasting N treatments (applied in the previous season). The grand mean N harvest index value for each site was then used to correct grain N uptake, to provide total N uptake data for each plot. These values were used an indicator of soil N supply. Analysis of variance (ANOVA) was used to compare effects of previous treatments on soil mineral N and soil N supply. Data were interpreted for relationships between canopy reflectance and soil N by fitting curves to relationships between and soil mineral N, and and soil N supply. 4

7 Key results Soil mineral N Fig. 1 shows effects of N applied in 26 on soil mineral N in November 26 after establishment of the next crop. There was a small effect of the largest level of N at Terrington, and there were some significant but rather small differences at Rosemaund, but overall, the best site appearing to offer a good test-bed for detecting soil mineral N via canopy sensing was at Boxworth. 4 Boxworth 4 Terrington SMN Nov. 26 (kg/ha, -6cm) SMN Nov. 26 (kg/ha, -6cm) N applied 26 (kg/ha) N applied 26 (kg/ha) SMN Nov. 26 (kg/ha, -6cm) Rosemaund SMN Nov. 26 (kg/ha, -6cm) High Mowthorpe N applied 26 (kg/ha) N applied 26 (kg/ha) Fig. 1 Effect of N applied in spring 26 on soil mineral N to.6 m depth in November 26, for four sites. From ANOVA, Boxworth, P<.1; Terrington, P<.1; Rosemaund, P=.5; High Mowthorpe, NS. 5

8 Over-winter crop growth Digital images of the crops on each sensing date are shown in Fig. 2. Establishment and autumn tillering went well at Boxworth and Rosemaund. The crop at Terrington was sown later and there was little tillering before winter and, therefore, there was little ground cover. At High Mowthorpe, the coldest site, tillering was also slow. Canopy reflectance The relationship between and soil mineral N (-6 cm, measured in November 26) was poor at the three sites (Terrington, Rosemaund and High Mowthorpe) where there were poor responses of soil mineral N in November 26 to N applied earlier in 26. However, at Boxworth, where there was a clear response of soil mineral N to N applied, inspection of the relationships between and N applied suggests that the relationships could be useful, especially at soil mineral N values below 15 kg/ha (Fig. 3). The relationships shown for January, February and March dates were better than the relationship in December. Boxworth Terrington Rosemaund High Mowthorpe Fig. 2 Digital images of the crops at each site. The top row is with nil N applied in 26 and the bottom is with the maximum N application. 6

9 .7 8-Dec Jan SMN, -6 cm (kg/ha) SMN, -6 cm (kg/ha).7 15-Feb Mar SMN, -6 cm (kg/ha) SMN, -6 cm (kg/ha) Fig. 3 Relationships between measured on four dates and soil mineral N (- 6 cm, measured in November 26) at Boxworth. Values of percentage variance accounted for by an exponential curve were 46.4, 67.5, 72.4 and 76.5 for December, January, February and March dates respectively. In Fig. 3, soil mineral N is used to estimate soil N supply. The most precise measure of soil N supply that can practically be taken by field measurement is N uptake by a crop with no N applied (i.e. N recovered from above-ground crop samples, including straw, chaff and grain). Using this approach (Fig. 4), the largest response was again at Boxworth. 7

10 2 Boxworth 2 Terrington SNS - N offtake with nil N applied (kg/ha) 1 SNS - N offtake with nil N applied (kg/ha) 1 N applied 26 (kg/ha) N applied 26 (kg/ha) SMN Nov. 26 (kg/ha, -6cm) Rosemaund N applied 26 (kg/ha) SNS - N offtake with nil N applied (kg/ha) 2 1 High Mowthorpe N applied 26 (kg/ha) Fig. 4 Effect of N applied in spring 26 on soil N supply (N uptake, kg/ha, with nil N applied), for four sites. From ANOVA, Boxworth, P<.1; Terrington, P=.8; Rosemaund, P=.6; High Mowthorpe, not significant. At Boxworth, where there was a clear response of soil N supply to N applied, the relationship between and soil N supply was better than at other sites. Comparison of the graphs in Fig. 5 with those in Fig. 3 shows that there appears to be better (less variation) and more useful (shallower slope) relationships when N uptake (rather than soil mineral N) is used to estimate soil N supply. This is confirmed by statistical analysis, as when N uptake (rather than soil mineral N) was used to estimate soil N supply, values were greater for percentage variance accounted for by the fitted lines. The relationships improved as assessment date (for ) became later. 8

11 Dec SNS - N offtake with nil N applied (kg/ha) Jan SNS - N offtake with nil N applied (kg/ha) Feb SNS - N offtake with nil N applied (kg/ha) Mar SNS - N offtake with nil N applied (kg/ha) Fig. 5 Relationships between measured on four dates and soil N supply (N uptake, kg/ha, with nil N applied) at Boxworth. Values of percentage variance accounted for by an exponential curve were 55.3, 74.2, 81.9 and 85.6 for December, January, February and March dates respectively. Change in canopy reflectance The concept that canopy sensing for changes in canopy size might be used to indicate N supplies was tested by plotting change in since the previous assessment against soil N supply (N uptake with nil N applied). Positive values indicated an increase in canopy size (i.e. more canopy and less soil was visible to the sensor), and negative values indicated a decrease. At Boxworth, for example, canopy size increased from December to January, then decreased by February, and increased again by the March assessment date. Inspection of photographs taken at Boxworth 9

12 (Fig. 4) supports this. The pattern of change differed between sites, and Terrington was the only site that did not show any period of canopy decline. At Boxworth, the site of most interest because of the responses of soil mineral N and soil N supply to previous N application, values for percentage variance accounted for by fitted exponential curves showed a better relationship in February and March (52.9 and 51.7 respectively) than in January (3.5). Relationships at Terrington and High Mowthorpe were poor, but at Rosemaund relationships in January and March were better. However, even the best of these relationships had a lower percentage variance accounted for by a fitted exponential curve than did absolute values of at Boxworth, for January, February and March assessment dates. Using change in per thermal time unit ( C d), rather than per time interval since the last assessment, appeared to improve some relationships, but the mathematical relationship (and so usefulness to predict soil N supply) was unchanged, as change in was divided by the same thermal time value for all treatments within a site. Changes in per unit thermal time were different between sites and between sampling dates, suggesting this would have limited value as a universal factor to use in predictions. Other factors influencing reflectance At Terrington and Rosemaund values from bare soil were approximately constant from date to date, but at Boxworth and High Mowthorpe there were individual high values relative to other dates. The causes of these high values are not known. Values for from the green mat showed some variation: for example, values declined over time at Boxworth. This suggests ambient light conditions were having a small effect on. Values of from a crop with adequate N increased with time at all sites, suggesting increasing canopy sizes as would be expected. However, at Rosemaund this increase was small compared with other sites. Conclusions and implications Over-fertilisation does not always lead to high SMN residues, but when it does (at Boxworth, which had the lowest optimum N requirement in 25-26: 117 kg/ha, compared to, 22 to 237 kg/ha at other sites), then we can detect it with sensors. 1

13 At Boxworth responded to soil mineral N up to around 15 kg/ha N (Fig. 6), but not above. At each of the other three sites was weekly related to soil mineral N, even though the ranges of soil mineral N values were within the range for which there was a response at Boxworth. This suggests that there were other factors, besides N supply, influencing canopy size. The ranges in values for each measurement date were greater at Boxworth than at other sites. At Terrington and High Mowthorpe values were generally low and canopy ground cover was also low. In contrast, values at Rosemaund were high, even on plots with very low soil mineral N, indicating rapid canopy development in autumn before the first measurement. Previous work has shown that remains fairly constant at leaf area indices above 3 (e.g. Scotford and Miller, 24), as values are poorly related to canopy size when there is full ground cover. The data indicate that canopy sensing over winter can indicate soil N supply in crops that have had a sufficient period of canopy growth to allow a measurable response to soil N supply, but are not approaching full canopy cover. The improved relationships when using N uptake rather than soil mineral N support the view that the true estimate of soil N supply (crop N uptake at harvest with nil fertiliser N applied) was determined more precisely than soil mineral N. There is a danger that the errors in autumn soil sampling and soil mineral N analysis may cloud the potential of canopy sensor data to predict true soil N supplies. Changes in did not show advantages over absolute readings, because zero (soil only) and full-canopy readings varied from date to date. Also, change in per thermal time unit did not show any useful consistency between sites. This result suggests that thermal time is not the dominant cause of between-site variation in change. At Boxworth, differentiated plots in need of fertiliser (soil mineral N up to around 15 kg/ha N) from those with ample soil N supplies. The future challenge is to refine the methodology and interpretation so that this can be achieved over a wide range of agronomic and environmental conditions, allowing useful comparison between sites. 11

14 Technical detail Introduction Judging soil N supplies is the most crucial, yet the most uncertain, element of judging fertiliser requirements on the farm. The best current method of judging soil N supplies involves field by field analysis of soil to at least 6 cm depth. Fig. 1 shows that, across a number of experiments, the crop recovers slightly more soil N than is measured as SMN in February to 9cm depth. 2 'Soil' N uptake after February (kg/ha) Soil mineral N in February (kg/ha, -9cm) Fig. 1 Relationship between soil mineral N to 9 cm in February for a wide range of crops (with different soils, previous crops, and sowing dates, over 3 seasons), and uptake of N after February from soil supplies only (i.e. with no fertiliser applied). The dotted line shows 1% recovery. The open circles are for a site where there was a history of using animal manures. However, sampling and analysis is laborious and expensive, and the results are not regarded with high confidence by the industry. Most farmers use look-up tables (e.g. in RB29), which give even more imprecise results (Webb et al. 1998). The HGCA recently commissioned a review of soil N analysis (Knight 26) which analysed relevant issues and recommended various measures to improve reliability. The need 12

15 for improved methods of assessing soil N supply has been highlighted in HGCA Research Review 63 (Richards, 27). The research described here was undertaken to assess whether the new technology of canopy sensing might revitalise an old idea that the crop could be used as a barometer of the soil (Hall 195). If proven, this approach could be applied to most cereal fields with relatively little investment because tractor- or sprayer-mounted canopy sensors are now available at affordable prices (~ 3, each). Previous studies have shown that canopy sensing techniques can detect differences in crops that relate to final grain yield (e.g. Wiltshire et al., 22; Reyniers et al., 24; Delin et al., 25). Sensing technologies have been commercially used to adjust fertiliser N rates within a field from sensing spatial variation (e.g. Yara N Sensor, Crop Circle and satellite images). However, this study has sought to use canopy sensing to directly estimate absolute levels of soil N supply. Initial tests with the Crop Circle canopy sensor in spring 26, following N response experiments in 25, demonstrated that the signal related significantly to SMN differences created by fertiliser treatments in 25 (Fig. 2), even though, from visual inspection, the crop in February appeared totally uniform. It also seemed from these measurements that using change in between different measurement times may provide a better (more consistent across sites) predictor of soil N status than itself. Correcting to thermal time might improve this prediction further, if a rate of change in per degree day could be established that was consistent across sites. 13

16 Boxworth 25-6 R 2 =.79 R 2 = April 28 Feb SMN in November (kg/ha) change (Feb-Apr) x Boxworth 25-6 R 2 = SMN in November (kg/ha) Fig Terrington 25-6 R 2 = April 3-March R 2 = SMN in November (kg/ha) change (Mar-Apr) x.2 Terrington R 2 = SMN in November (kg/ha) Initial measurements using Crop Circle on plots at Boxworth and Terrington in the 25/6 growing season, following N treatments in the previous season. Thus the research described here continued tests of whether canopy sensing between autumn and early spring might be used to indicate N supplies to cereal crops. In particular it was suggested that (i) changes in canopy size, as opposed to the absolute sizes of canopies would be most indicative of soil N supplies, and that (ii) changes in canopy size would be most meaningful if interpreted against thermal time rather than absolute time. This research was opportune because it could exploit existing plots with contrasting soil N, already set up under HGCA & Defra funding. 14

17 Materials & Methods Sites In winter 26-7, four cereal sites (Boxworth, Terrington, Rosemaund and High Mowthorpe) were identified where there had been N fertiliser experiments in 26 with 6 N levels under HGCA Project 384, and where Defra was funding analyses of soil mineral N and crop N content (through Defra Project IS223). A fifth site in eastern Scotland was also initiated, but was aborted before data were recorded because of a period of consistently wet weather and, logistically, there was insufficient time to move sensors between 5 disparate sites within the monthly measurement period. As part of Defra project IS223, after sowing a cereal crop in autumn, all plots were re-marked out in the positions of the N fertiliser experiment plots in the previous season (25-26). The treatments are summarised in Table 1. There were three replicates. Table 1 Sites, wheat varieties for N response plots and total N applied in Site Varieties Total N applied (kg/ha N) Boxworth XI19, 7, 15, 21, 22, 29, 36, 37 Longbow, Norman,, 7, 15, 22, 29, 37 Alchemy, Gladiator Terrington XI19, 7, 14, 21, 28, 35 Longbow, Virtue,, 7, 15, 22, 29, 37 Amb, Gladiator Rosemaund XI19,, 7, 15, 22, 29, 37 Longbow, Hustler, Ambr, Robigus High Mowthorpe XI19, Longbow, Norman, Amb, Istabraq, 7, 14, 21, 28, 35 Crops grown in are listed in Table 2. Plots received no N applications in the 26/27 season. 15

18 Table 2 Sites and details of crops grown in Site Crop Variety Sowing date Emergence date Boxworth Winter wheat Nijinsky 26 Sep 26 9 Oct 26 Terrington Winter wheat Nijinski 27 Oct Nov 26 Rosemaund Winter barley Saffron 2 Sep 26 1 Oct 26 High Mowthorpe Winter wheat Gladiator 3 Sep Oct 26 At each site two additional plots (2m wide by ~1 m long) were made in a discard area adjacent to the trial and reflectance data were collected as for the cereal plots from the previous N experiments. One plot was bare soil, and the other was the crop with additional N (2 kg/ha N in November and 2 kg/ha N in January) to give unlimited growth. Also, to give an indication of varying light quality, a mat of green plastic was taken to each site at each measurement date, and reflectance data were collected as for the cereal plots. Sensors The study was intended to test two canopy sensors: the N-Sensor from Yara UK Ltd. and the Crop Circle from Soil Essentials Ltd. The N-Sensor was designed to be tractormounted, but the measurements of plots required hand-held operation. Considerable effort was made to find a practical method for operating the N-sensor over small plots. A rig was developed to allow both sensors to be carried together by two staff, to give a sensed area no wider than the 2 m width of the plots (Fig. 3). This rig was ready for first use in December 26 and was used at Terrington. However, it proved difficult to achieve the required measurements within the time available (limited by suitable light conditions) under field conditions. After a further attempt to use the N- sensor at Boxworth it was decided that use at other sites was not practical. The work therefore continued with just the Crop Circle. The Crop Circle comprised two sensors, each with its own light source (to minimise variation due to variation in ambient light conditions). The sensors measured reflectance at 88 nm (near-infrared, NIR) and 59 nm (orange). 16

19 Fig. 3 Rig holding the N-Sensor and the Crop Circle canopy sensors. The N- Sensor was later removed because it proved too heavy for protracted operation with two staff. Data collection and analysis Reflectance was measured on a monthly basis throughout the period of tillering December to March. Normalised Difference Vegetation Index () was calculated from the sensor data. gives a measure of vegetative cover. Dense canopies strongly absorb red wavelengths and reflect near-infrared wavelengths. is a measure of the difference in reflectance between these wavelength ranges, and high values (to a maximum of 1) indicate dense canopies. is given by the following equation. = (NIR-visible/(NIR+visible) In this work we used light of wavelength 59 nm as the visible light for this calculation. SMN data were obtained from Defra project IS223 which used the same sites. For plots that had grown varieties other than Xi19 in the previous season, and that had the same previous N applications, plots were paired (paired varieties are on the same lines in Table 1) and one value was obtained for a pair of plots. Reflectance data were also paired in the same way, so that a pair of plots in became one plot in

20 Yields were recorded at crop maturity using a plot combine harvester. Samples of grain were submitted to a laboratory for grain N analysis, and grain uptake was calculated. Grab samples were also taken and straw and chaff were submitted to a laboratory for N analysis, from a sub-set of plots at each site. Nitrogen harvest indices were calculated for each of these plots, and were found to be similar for contrasting N treatments (applied in the previous season). The grand mean N harvest index value for each site was then used to correct grain N uptake, to provide total uptake data for each plot. These values were used as the ultimate indicator of soil N supply (SNS). Analysis of variance (ANOVA) was used to compare effects of previous treatments on SMN and SNS. Data were interpreted for relationships between canopy reflectance and soil N by fitting curves to relationships between and SMN, and and SNS. Single curves were fitted across all varieties for each of the relationships of interest, as there was no improvement from fitting separate curves for different varieties. Results Soil mineral N Fig. 4 shows effects of N applied in 26 on SMN in November 26, after establishment of the next crop at four sites. The only sizable response was at Boxworth. There was a small effect of the largest level of N at Terrington, and there were some significant but rather small differences at Rosemaund, but the best site appearing to offer a good test for detecting SMN via canopy sensing was at Boxworth. Optimum N requirement in the previous season was lower at Boxworth than at other sites (117 kg/ha, compared with, 22, 236 and 237 kg/ha at Terrington, Rosemaund and High Mowthorpe respectively. Thus, over-fertilisation was greater at Boxworth, and SMN residues were greater in high-n plots. Yield (not measured as part of this project) responded to previous N applications in a similar way to SMN at all sites, confirming the validity of these SMN results. 18

21 4 Boxworth 4 Terrington SMN Nov. 26 (kg/ha, -6cm) SMN Nov. 26 (kg/ha, -6cm) N applied 26 (kg/ha) N applied 26 (kg/ha) SMN Nov. 26 (kg/ha, -6cm) Rosemaund SMN Nov. 26 (kg/ha, -6cm) High Mowthorpe N applied 26 (kg/ha) N applied 26 (kg/ha) Fig. 4 Effect of N applied in spring 26 on soil mineral N to.6 m depth in November 26, for four sites. From ANOVA, Boxworth, P<.1; Terrington, P<.1; Rosemaund, P=.5; High Mowthorpe, not significant. Over-winter crop growth Digital images of the crops on each date of sensing are shown in Fig. 5. Crop establishment and autumn tillering went well at Boxworth and Rosemaund. The crop at Terrington was sown later and there was little tillering before winter and, therefore, there was little ground cover. At High Mowthorpe, the coldest site, tillering was also slow. Thus canopy cover was near complete at Boxworth and Rosemaund but small at the other sites. 19

22 Boxworth Terrington Rosemaund High Mowthorpe Fig. 5 Digital images of the crops taken at the four sites when the canopy was sensed. For each site, the top row is with nil N applied in 26 and the bottom is with the maximum N application. Canopy reflectance The relationship between and SMN (-6 cm, measured in November 26) was poor at the three sites (Terrington, Rosemaund and High Mowthorpe, Fig. 6) where there were poor responses of SMN in November 26 to N applied earlier in 26. However, at Boxworth, where there was a clear response of SMN to N applied, inspection of the relationships between and N applied suggests that the relationships could be useful, especially at SMN values below 15 kg/ha (Fig. 6). 2

23 .7 Boxworth.7 Terrington 12-Dec Jan-7 13-Feb Mar Dec Jan Feb Mar SMN, -6 cm (kg/ha) SMN, -6 cm (kg/ha).7 Rosemaund.7 High Mowthorpe 4-Jan Feb-7 5-Mar Apr Dec-6 29-Jan Feb Mar SMN, -6 cm (kg/ha) SMN, -6 cm (kg/ha) Fig. 6 Relationships between (measured on four dates) and SMN (-6 cm, measured in November 26), at four sites. 21

24 The relationships between and SMN at Boxworth, shown in Fig. 6, are plotted individually (for each of four dates) in Fig. 7. The relationships shown for January, February and March dates were better than the relationship in December. The fitted curves and values of percentage variance accounted for by the curves are given in Table Dec Jan SMN, -6 cm (kg/ha) SMN, -6 cm (kg/ha).7 15-Feb Mar SMN, -6 cm (kg/ha) SMN, -6 cm (kg/ha) Fig. 7 Relationships between measured on four dates and SMN (-6 cm, measured in November 26) at Boxworth. See Table 3 for equations of fitted curves and percentage variance accounted for by the fitted curves. In the graphs above, SMN is used to estimate SNS. The most precise approach available for assessing SNS is to measure N uptake by a crop with no N applied (i.e. N recovered from above-ground crop samples, including straw, chaff and grain). Fig. 8 shows effects of N applied in 26 on SNS (N uptake with nil N applied) at four sites. 22

25 As for SMN (Fig. 4), the largest response was at Boxworth. The response of SNS to N applied in the previous season (Fig. 8) was similar to the responses of SMN (Fig. 4) and grain yield (not presented). However, there was more variation in SNS than in SMN at Terrington, Rosemaund and High Mowthorpe, but less at Boxworth. At Rosemaund the slope of the relationship was greater for SNS than for SMN. 2 Boxworth 2 Terrington SNS - N offtake with nil N applied (kg/ha) 1 SNS - N offtake with nil N applied (kg/ha) 1 N applied 26 (kg/ha) N applied 26 (kg/ha) 2 Rosemaund 2 High Mowthorpe SNS - N offtake with nil N applied (kg/ha) 1 SNS - N offtake with nil N applied (kg/ha) 1 N applied 26 (kg/ha) N applied 26 (kg/ha) Fig. 8 Effect of N applied in spring 26 on SNS (N uptake, kg/ha, with nil N applied), for four sites. From ANOVA, Boxworth, P<.1; Terrington, P=.8; Rosemaund, P=.6; High Mowthorpe, not significant. In Fig. 9 is plotted against SNS, rather than SMN as in Fig. 6. At Boxworth, where there was a clear response of SNS to N applied, the relationship was better than at other sites. 23

26 .7 Boxworth.7 Terrington 12-Dec Jan-7 13-Feb Mar Dec-6 23-Jan Feb-7 8-Mar SNS - N offtake with nil N applied (kg/ha) SNS - N offtake with nil N applied (kg/ha).7 Rosemaund.7 High Mowthorpe 4-Jan Feb Mar-7 12-Apr Dec-6 29-Jan Feb Mar SNS - N offtake with nil N applied (kg/ha) SNS - N offtake with nil N applied (kg/ha) Fig. 9 Relationships between (measured on four dates) and SNS (N uptake, kg/ha, with nil N applied), at four sites. The data from Boxworth are shown for each date separately in Fig. 1. Comparison of these graphs with those in Fig. 7 shows that there appears to be better (less 24

27 variation) and more useful (shallower slope) relationships when N uptake (rather than SMN) is used to estimate SNS. This is confirmed by statistical analysis (Table 3), as when N uptake (rather than SMN) was used to estimate SNS, values were greater for percentage variance accounted for by the fitted lines. The relationships improved as assessment date (for ) became later Dec SNS - N offtake with nil N applied (kg/ha) Jan SNS - N offtake with nil N applied (kg/ha) Feb SNS - N offtake with nil N applied (kg/ha) Mar SNS - N offtake with nil N applied (kg/ha) Fig. 1 Relationships between measured on four dates and SNS (N uptake, kg/ha, with nil N applied) at Boxworth. 25

28 Table 3 Parameters of fitted exponential curves, Y=A+B(R X ) (where Y=, and X=SMN or SNS) and percentage variance accounted for by the fitted curve, for the Boxworth site. The curves are presented in Fig. 7 (SMN) and Fig. 1 (SNS). SMN (-6 cm) SNS (N uptake with nil N applied) Date A B R % var. acc. for A B R % var. acc. for 8 Dec Jan Feb Mar Change in canopy reflectance The concept that canopy sensing for changes in canopy size might be used to indicate N supplies was tested by plotting change in since the previous assessment against SNS (N uptake with nil N applied). The mean values, for all N treatments (applied to the previous crop) are shown in Fig. 11. If we assume that is reliably correlated with canopy size, then positive values indicated an increase in canopy size (i.e. more canopy and less soil was visible to the sensor), and negative values indicated a decrease. Thus, Fig. 11 shows that at Boxworth, for example, canopy size increased from December to January, then decreased by February, and increased again by the March assessment date. Inspection of photographs taken at Boxworth (Fig. 5) supports this. The pattern of change differed between sites, and Terrington was the only site that did not show any period of canopy decline. 26

29 Jan Feb Mar Jan Feb Mar Jan Feb Mar Feb Mar Apr Change in Boxworth Terrington Rosemaund High Mowthorpe Fig. 11 Mean values (all plots) of change in since the previous assessment date, for three dates and four sites. The relationship between change in and SNS had a positive slope at some assessments, and a negative slope at others. This is illustrated by the data from Boxworth and Rosemaund (Fig. 12). At Rosemaund change in between December and January became more negative as SNS increased, indicating that canopies of crops with a large SNS declined more than crops with smaller SNS. However, between February and March, the canopy declined more in crops with a small SNS, and declined very little in crops with a high SNS. At Boxworth, the site of most interest because of the responses of SMN and SNS to previous N application (Figs. 4 and 8), the relationships between change in and SNS were positive on each of the three assessment dates. 27

30 Change in Jan-7 15-Feb-7 8-Mar-7 Boxworth SNS - N offtake with nil N applied (kg/ha) Change in Jan-6 21-Feb-7 15-Mar-7 Rosemaund SNS - N offtake with nil N applied (kg/ha) Fig. 12 Relationship between change in since the previous assessment date and SNS (N uptake, kg/ha, with nil N applied) at Boxworth (top graph) (first assessment date, previous to 23 January, was 8 December) and Rosemaund (bottom graph) (first assessment date, previous to 29 January, was 14 December). At Boxworth values for percentage variance accounted for by fitted exponential curves (Table 4) show a better relationship in February and March than in January. Relationships at Terrington and High Mowthorpe were poor, but at Rosemaund relationships in January and March were better (Fig. 12 and Table 4). However, even the best of these relationships had a lower percentage variance accounted for by a fitted exponential curve than did absolute values of at Boxworth, for January, February and March assessment dates (Table 3). 28

31 Table 4 Parameters of fitted exponential curves, y=a+b(r x ) (where y=change in since the last assessment, and x=sns) and percentage variance accounted for by the fitted curves, for three dates at each of four sites. Example curves are presented in Figs. 9 (Rosemaund) and 1 (A) (Boxworth). Site Date A B R % var. acc. for Boxworth 23 Jan Feb Mar Terrington 22 Jan Feb Mar Rosemaund 29 Jan Feb Mar High Mowthorpe 1 Feb Mar Apr Using change in per thermal time unit ( C d), rather than per time interval since the last assessment, appeared to improve some relationships (e.g. Boxworth, 23 January, as shown by comparison of Figs. 12 and 13), but the mathematical relationship (and so usefulness to predict SNS within sites) is unchanged, as change in was divided by the same thermal time value for all treatments within a site. Comparison of change in per thermal time unit ( C d) between sites, for similar time periods and SNS levels, is of interest because thermal time could account for some of the between site variation, and so correcting for thermal time may improve the value of to predict SNS. Values for all sites are plotted in Fig. 13, using the same axis scales to assist comparison. It can be seen from Figure 13 that there is no consistency between sites for values of per C d. For example, estimated values for the first date on each graph, for an SNS value in the range 8 to 1 kg/ha, are.1,.16, -.1 and.1 for Boxworth, Terrington, Rosemaund and High Mowthorpe respectively. 29

32 Change in per C d SNS - N offtake with nil N applied (kg/ha) 23-Jan-7 15-Feb-7 8-Mar-7 Boxworth First assessment date was 8 Dec 7 Change in per C d Change in per C d Change in per C d SNS - N offtake with nil N applied (kg/ha) SNS - N offtake with nil N applied (kg/ha) SNS - N offtake with nil N applied (kg/ha) 22-Jan-7 13-Feb-7 9-Mar-7 29-Jan-6 21-Feb-7 15-Mar-7 1-Feb-7 5-Mar-7 12-Apr-7 Terrington First assessment date was 12 Dec 7 Rosemaund First assessment date was 14 Dec 7 High Mowthorpe First assessment date was 4 Jan 8 Fig. 13 Relationships between change per day degree in since the previous assessment date and SNS (N uptake, kg/ha, with nil N applied). 3

33 Other factors influencing reflectance The possible effects of non-crop factors on the signals received by the sensors were investigated by collecting data from bare soil, a green mat and a crop with adequate N nutrition. Bare soil values of would not be expected to vary between dates if absolute values of accurately signal canopy size. At Terrington and Rosemaund values from soil were approximately constant from date to date (Fig. 14), but at Boxworth in January, and at High Mowthorpe on 1 February, there were high values relative to other dates. The causes of these high values are not known. Descriptive records taken at the time of measurement do not indicate any relationship to soil surface wetness. These changes in between dates at Boxworth and High Mowthorpe were greater than differences between sites. Values for from the green mat were relatively consistent across sites, but tended to decline over time, especially at Boxworth and Terrington (Fig. 14). It is possible that this is related to increasing solar radiation as the solar angle from the horizon increased with advancing date. Values of from a crop with adequate N increased with time at all sites, suggesting increasing canopy sizes as would be expected. However, at Rosemaund this increase was small compared with other sites. 31

34 Boxworth Bare soil Green mat N applied 1-Dec 29-Dec 26-Jan 23-Feb 23-Mar Assessment date.8 Terrington Dec 29-Dec 26-Jan 23-Feb 23-Mar Assessment date.8 Rosemaund Dec 29-Dec 26-Jan 23-Feb 23-Mar Assessment date.8 High Mowthorpe Dec 29-Dec 26-Jan 23-Feb 23-Mar Assessment date Fig. 14 from bare soil, a green surface and a crop with adequate nitrogen, at four sites. 32

35 Discussion This project undertook initial work to test the concept that canopy sensing for changes in canopy size might be used to indicate N supplies to cereal crops. Commercial experience, as well as crop physiology, supports the idea that changes in canopy size (adjusted for temperature) should best indicate soil N supply. Over-fertilisation does not always lead to high SMN residues, but when it does (at Boxworth, which had the lowest optimum N requirement in 25-26: 117 kg/ha, compared to, 22 to 237 kg/ha at other sites), then we can detect it with sensors. At Boxworth, where there was the largest range of SMN, responded to SMN up to around 15 kg/ha N (Fig. 7), but not above. At each of the other three sites was weekly related to SMN, even though the ranges of SMN values were within the range for which there was a response at Boxworth. This suggests that there were other factors, besides N supply, influencing canopy size. The ranges in values for each measurement date were greater at Boxworth than at other sites (see Figs. 6 and 9). At Terrington and High Mowthorpe values were generally low (Fig. 6) and canopy ground cover was also low (Fig. 5). In contrast, values at Rosemaund were high, even on plots with very low SMN (Fig. 6), indicating rapid canopy development in autumn before the first measurement. Previous work has shown that remains fairly constant at leaf area indices above three or until canopy closure as values are poorly related to canopy size when there is full ground cover (e.g. Scotford and Miller, 24a, 24b). It is suggested that the use of (as a measure of canopy size) over winter to indicate soil N supply, may not be useful when sowing date and the subsequent crop environment have allowed development of a large canopy size (i.e. close to full ground cover) before the time of measurement. Conversely, at very low canopy sizes, when the crop has grown little in the autumn, and over-winter growth is restricted by the crop environment (e.g. low temperature), the response of to SNS may be small in relation to values from bare soil, and so difficult to detect. The data indicate that canopy sensing over winter can indicate SNS in crops that have had a sufficient period of canopy growth to allow a measurable response to SNS, but are not approaching full canopy cover. 33

36 The improved relationships between and SNS when using N uptake rather than SMN support the view that the true estimate of SNS (crop N uptake at harvest with nil fertiliser N applied) was determined much more precisely than SMN. There is a danger that the errors in autumn soil sampling and SMN analysis may cloud the potential of canopy sensor data to predict true soil N supplies. It is therefore recommended that, if possible, future studies should continue to assess SNS by nil-n crop uptake as the ultimate calibration measure. Changes in did not show advantages over absolute readings, because zero (soil only) and full-canopy (grass sward) readings varied from date to date. It is also probable that canopy size decreased between some assessment dates, presumably because the rate of green area loss (senescence) exceeded the rate of green area growth. This is strongly supported by the occurrence of positive and negative differences in change between plots with different SNS values (Fig. 12). Change in per thermal time unit did not show any useful consistency between sites. This result suggests that thermal time is not the dominant cause of betweensite variation in change. At Boxworth differentiated plots in need of fertiliser (SMN up to around 15 kg/ha N) from those with ample soil N supplies, and this is an encouraging result. The challenge is to refine the methodology and interpretation so that this differentiation can be achieved over a wide range of agronomic and environmental conditions, allowing useful comparison between sites. For change in to be a useful indicator of SNS, the factors causing canopy change need to be modelled (i.e. accounted for quantitatively using other measurements of crop and environment) to allow the effect of SNS to be separated from other effects. To be useful as an indicator of SNS, canopy sensors either must read the same (for the same canopy) from date to date and place to place, or must be corrected for effects of other factors that influence canopy growth or decline. Variation between dates in values, in the absence of a crop, requires further study. Background factors (e.g. soil colour and wetness, canopy wetness, light conditions) need to be properly accounted for to maximise the opportunity to discriminate between canopy responses to growing conditions (e.g. soil N supply). The variation in values from bare soil indicates that bare soil strips could have some usefulness in helping to account for changes in reflectance from soil. 34

37 Digital camera images usefully complemented sensor information by allowing changes in to be checked against canopy appearance. Such images have the potential to enable separation of leaf greenness from soil-cover. We provisionally conclude that young canopies can signal soil N status where SMN is less than 15 kg/ha (most UK fields), but further work is necessary to confirm the wider applicability of this and study factors causing interference in canopy signals (e.g. soil type and wetness, and light conditions at the time of measurement). This work has been extended for a second year with the following objectives (i) To test how canopies respond to different levels of residual soil N at four sites, using proprietary sensors. (ii) To test the effects of contrasting conditions (e.g. light intensity, canopy wetness, soil colour) and propose ways of allowing for these when interpreting sensor data. (iii) To assemble and interpret the evidence for further LINK-funded research, to develop canopy sensing over-winter as an easy way of assessing soil N supplies. Acknowledgements The support to this project provided by Soil Essentials Ltd (provision of sensors, technical support and data processing) and Yara UK Ltd (provision of sensors and technical support) is gratefully acknowledged. We particularly thank Mark Tucker (Yara UK Ltd) and Clive Blacker (Precision Decisions). The use of experimental plots and data funded under Defra Project IS223 is acknowledged. References Hall, A.D. (195). The analysis of the soil by means of the plant. Journal of Agricultural Science, Volume 1. Delin, S. Linden, B. & Berglund, K. (25). Yield and protein response to fertilizer nitrogen in different parts of a cereal field: potential of site-specific fertilization. European Journal of Agronomy 22(3),

38 Knight, S.M. (26). Soil mineral nitrogen testing: Practice and interpretation. Research Review No. 58, Home-Grown Cereals Authority, London. Reyniers, M. Vrindts, E. & De Baerdemaeker, J. (24). Optical measurement of crop cover for yield prediction of wheat. Biosystems Engineering 89(4), Richards, I. (27). Research needs on nitrogen and phosphate management in cereals and oilseeds. In Research Review No. 63, Home-Grown Cereals Authority, London. Scotford, I.M. & Miller P.C.H. (24). Estimating tiller density and leaf area index of winter wheat using spectral reflectance and ultrasonic techniques. Biosystems Engineering, 89, Scotford, I.M. & Miller P.C.H. (24). Assessment of sensor-based technologies for monitoring crop growth and development in cereals. Project Report No. 331, Home- Grown Cereals Authority, London. Scotford, I. M. & Miller, P. C. H. (24). Combination of spectral reflectance and ultrasonic sensing to monitor the growth of winter wheat. Biosystems Engineering 87(1), Sylvester-Bradley, R., Davies, D.B., Dyer, C.J., Rahn, C. & Johnson, P.A. (1997) The value of nitrogen applied to winter wheat during early development. Nutrient Cycling in Agroecosystems 47, Sylvester-Bradley, R., Stokes, D.T. & Scott, R.K. (21). Dynamics of nitrogen capture without fertiliser: the baseline for fertilising winter wheat in the UK. Journal of Agricultural Science 136, Wiltshire, J., Clark, W., Riding, A., Gay, A., Steven, M., Holmes, G. & Moore, M. (22). Spectral reflectance as a basis for in-field sensing of crop canopies for precision husbandry of winter wheat (SARTAN project). Report to SA LINK. 8 pp. Webb, J., Sylvester-Bradley, R., Shepherd, M.A. & Goodlass, G. (1998) A review of recent UK studies to improve fertilizer-n recommendations for cereals. In Fertilization for sustainable plant production and soil fertility, 11th International World Fertiliser Congress, Gent, Belgium, Vol. III Eds. O. Van Cleemput, S. Haneklaus, G. Hofman, E. Schnug & A. Vermoesen, pp