- Email: [email protected]
Development of on-line measurement system of bulk density based on on-line measured draught, depth and soil moisture content
Soil & Tillage Research 86 (2006) 218–229 www.elsevier.com/locate/still
Development of on-line measurement system of bulk density based on on-line measured draught, depth and soil moisture content Abdul Mounem Mouazen a,b,*, Herman Ramon a a
Department of Agro-Engineering and -Economics, Faculty of Agricultural and Applied Biological Sciences, Kasteelpark Arenberg 30, B-3001 Heverlee, Belgium b Department of Rural Engineering, Faculty of Agriculture, University of Aleppo, P.O. Box 12214, Aleppo, Syria Received 17 November 2004; received in revised form 16 February 2005; accepted 22 February 2005
Abstract On-line measurement of soil compaction is needed for site specific tillage management. The soil bulk density (r) indicating soil compaction was measured on-line by means of a developed compaction sensor system that comprised several sensors for on-line measurement of the draught (D) of a soil cutting tool (subsoiler), the soil cutting depth (d) and the soil moisture content (w). The subsoiler D was measured with a single shear beam load cell, whereas d was measured with a wheel gauge that consisted of a swinging arm metal wheel and a linear variable differential transducer (LVDT). The soil w was measured with a near infrared fibre-type spectrophotometer sensor. These on-line three measured parameters were used to calculate r, by utilising a hybrid numerical– statistical mathematical model developed in a previous study. Punctual kriging was performed using the variogram estimation and spatial prediction with error (VESPER) 1.6 software to develop the field maps of r, soil w, subsoiler d and D, based on 10 m 10 m grid.To verify theon-linemeasuredr map,this map was compared withthemap measuredby theconventional core samplingmethod. The spherical semivariogram models, providing the best fit for all properties was used for kriging of different maps. Maps developed showed that no clear correlation could be detected between different parameters measured and subsoiler D. However, the D value was smaller at shallow penetration d, whereas large D coincided with large r values at few positions in the field. Maps of r measured with the core sampling and on-line methods were similar, with correlation coefficient (r) and the standard error values of 0.75 and 0.054 Mg m3, respectively. On-line measured r exhibited larger errors at very dry zones. The normal distribution of the r error between the two different measurement methods showed that about 72% of the errors were less than 0.05 Mg m3 in absolute values. However, the overall mean error of on-line measured r was of a small value of 2.3%, which ensures the method accuracy for on-line measurement of r. Measurement under very dry conditions should be minimised, because it can lead to a relatively large error, and hence, compacted zones at dry zones cannot be detected correctly. # 2005 Elsevier B.V. All rights reserved. Keywords: Bulk density; On-line measurement; Sensor; Moisture content; Draught; Depth
* Corresponding author. Tel.: +32 163 285 92; fax: +32 163 285 90. E-mail addresses: [email protected], [email protected] (A.M. Mouazen). 0167-1987/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.still.2005.02.026
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
1. Introduction Soil compaction is one of the main factors limiting plant growth and crop yield. It is well-known that soil compaction is a natural or human created problem that results in increase of bulk density (r) and root penetration resistance, while decrease in void ratio and available nutrients, water and oxygen for plant can very frequently take place. Moreover, it was found by Fleige and Horn (2000) that anthropogenic compaction in traffic rut and plough pan caused a reduction of coarse pores and saturated water conductivity, leading to surface runoff and soil erosion. Alleviation of the negative effect of soil compaction on plant growth, along with the reduction of tillage energy input and increasing the surface plant residue relay on successfully managed site specific tillage operations. With the recent development in precision agricultural technologies, researchers have focused on development of on-line sensors to measure different properties in agricultural soils (Adamchuk et al., 2004). Since online measurement of soil compaction demands simultaneous measurement of different influencing parameters and a proper mathematical modelling technique, development of an on-line soil compaction sensor is still a challenging issue for researchers involved in new technologies and engineering sections of precision agriculture. The most common techniques to determine the degree of soil compaction is the field measurement of r and penetration resistance. Both parameters are measured with traditional methods under static conditions, using core sampling methods and penetrometers, respectively. Under dry soil conditions, these measurements are very difficult and time costly procedures, in addition to the discontinuous data output provided based on fine or coarse measurement grids. Soil compaction was referred to indirectly as on-line measured soil mechanical resistance or draught (D) of different cutting or penetration tools using different load cells (Sprinkle et al., 1970; Upadhyaya et al., 1984; Glancey et al., 1996; Sirjacobs et al., 2002; Verschoore et al., 2003) or strain gauges (Glancey et al., 1989; Adamchuk et al., 2001). Godwin and Miller (2003) stated that there is now, from commercial sources, evidence that the electromagnetic induction (EMI) will distinguish between different levels of soil compaction (Smith,
219
2001). However, EMI is mainly dependent on the clay content, salinity, organic matter, moisture content (w) and r (Schmidhalter et al., 2001), which introduce problems of discrimination of different factors. In a recently published study, Besson et al. (2004) related the soil electrical resistivity to soil r, showing successful discrimination between loose and compacted layers. In addition to the sensitivity of the electrical resistivity to salinity and ambient temperature, the developed electrical resistivity– bulk density relationship depended on the soil type and w. When r is selected for on-line, tractor-based measurement of soil compaction with a soil cutting tool or penetration device, all influencing parameters should be measured simultaneously during measurement, namely D, depth (d) and w. Liu et al. (1996) developed a real-time texture/compaction sensor. They stated that if d and the speed of a soil cutting tine are kept constant, D will be a function of texture, r and w. Based on a combination of the finite element method and multiple linear regression analysis, Mouazen et al. (2003) developed the following model for the calculation of soil compaction indicated as bulk density:
r¼
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi! 2 3 D þ 21:36 w 73:9313 d 1:14 1:6734
(1)
where D is the draught in kN, w the soil moisture content in kg kg1, d the subsoiler depth in m and r is the bulk density in Mg m3. The mathematical model of Eq. (1) stablishes a basis of carrying out an indirect measurement of r, when the independent parameters (D, d and w) can be measured simultaneously on-line. So far, Mouazen et al. (2003) produced r maps by using Eq. (1) and measured D of the subsoiler (cutting tool) with an extended octagonal load cell, similar to the load cell developed by Godwin (1975). However, soil w and subsoiler d were measured using traditional techniques, so that w was measured with oven drying methods and d was measured manually. But, the online measurement of r will only be satisfied, if all the independent parameters (D, d and w) are measured online simultaneously, with values to be substituted into Eq. (1) to obtain r.
220
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
This study aims at introducing an on-line measurement system of r, using on-line, simultaneously measured D, d and w values and a mathematical model. The reliability of an on-line developed r map is investigated by performing a comparison with the r map measured with the traditional core sampling method.
Table 1 Soil texture defined according to the USDA soil classification at a depth of 0–0.02 m Property
Sandy loam (g kg1)
Loam (g kg1)
Sand (>50 mm) Silt (2–50 mm) Clay (<2 mm)
562.77 362.41 74.82
464.24 443.74 92.02
2.2. Sensor system 2. Materials and methods 2.1. Site description The experimental field of 2.3 ha was located in the Zoutleeuw region, southeast of Brussels, Belgium. The soil type is an Arenic Cambisol, according to the FAO classification. The soil texture fractions were determined by a combination of wet sieve and hydrometer tests, using the USDA Soil Classification, as shown in Table 1. The soil texture over the field down to 0.20 m is non-homogeneous, including two textures of sandy loam and loam, as listed in Table 1. Samples were collected from the upper 0.2 m layer during the growing season of wheat based on 50 m 50 m grid, and they were subjected to texture analysis and spectrophotometer sensor calibration.
A standard medium–deep subsoiler used as soil cutting tool was attached to a frame, which was mounted onto the three point hitch of the tractor. The subsoiler consisted of two parts; the chisel of 0.06 m width, and the shank of 0.03 m width.
2.2.1. Sensor for D measurement For simplicity, the extended octagonal load cell used to measure the subsoiler D (Mouazen et al., 2003) was replaced by a commercially available single ended shear beam load cell from CELTRON TECHNOLOGIES Inc. The load cell was inserted within a gap between the subsoiler shank backside and a fixing plate, as shown in Fig. 1. When the shank was pushed backwards by means of the soil resistance
Fig. 1. Set up of r sensor ready for on-line field measurement.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
force opposite to the direction of travel, the load cell is subjected to bending and the output signal is recorded. 2.2.2. Sensor for w measurement The on-line measurement of w is the most difficult parameter needed for this study. A Corona fibre-type visible (VIS) and near infrared (NIR) spectrophotometer from Zeiss company, with a light reflectance measurement range of 306.5–1710.9 nm was used to measure w in the field. The spectrophotometer optical unit was attached to the subsoiler chisel backside to perform the light reflectance measurement from the soil surface of the bottom of the trench opened by the preceded chisel (Fig. 1). More detailed description about the sensor can be found in a previous published study (Mouazen et al., 2004a). 2.2.3. Sensor for d measurement A metal wheel gauge mounted on the frame with a revolute joint was designed to measure the distance variation between the soil surface and the frame. A linear variable differential transducer (LVDT) with a 0.2 m stroke length was used to connect the frame to the axle of a metal wheel of the height sensor, as show in Fig. 2. The performance of the wheel gauge on four different surfaces was investigated in detail by
221
Mouazen et al. (2004b). They found that in comparison with an ultrasonic sensor, the wheelLVDT sensor can be used more accurately than the ultrasonic sensor to measure the subsoiler d, in fields covered with plant residues and stubble. 2.2.4. Sensor electrical system The electrical system, besides the spectrophotometer, consisted of several modules: a basic power supply, travel speed sensor, global positioning system, signal conditioning system, amplifier and data acquisition system. The travel speed was measured using a doppler radar that was mounted on the frame pointing backwards to avoid the effects of stubble or grass movement after the measurement frame passed. The accuracy of the sensor was tested in previous experiments and all errors were smaller than 2.5%. Position, latitude and longitude, were determined with a Trimble AgGPS132 differential global positioning system (DGPS). The antenna was mounted just above the sensor, to obtain a sub-meter accuracy. Using the National Instruments Labview programming language, a custom-built data acquisition software was developed. The speed sensor signal was acquired at a frequency of 200 Hz, while the DGPS information was stored at 1 Hz. The load cell, LVDT together with the speed sensor signal and the DGPS information were logged by use of a Pentium III 800 MHz laptop computer equipped with a National Instruments DAQ700 data acquisition card. An EURORAC data logging system was designed, through which the signals of load cell, LVDT and radar are filtered, amplified and transferred to the Pentium III 800 MHz laptop computer. The spectral signal was acquired and stored at 0.225 Hz. Each spectrum was an average of five successive spectra. They were logged to a separate Pentium IV, 1.7 GHz laptop computer equipped with RS422 cable and PCMCIA acquisition card. 2.3. On-line field experiment
Fig. 2. Wheel gauge consisted of a swinging arm metal wheel and a LVDT sensor.
The on-line field measurement was carried out in the Zoutleeuw field, the same field where soil samples were collected for texture analysis and spectrophotometer sensor calibration. The experiment was carried out after harvesting wheat. After setting up the different sensors, the subsoiler was pulled throughout parallel lines of 10 m apart, as shown in Fig. 3. These
222
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
Fig. 3. Sampling and map design based on a 10 m 10 m grid.
lines were perpendicular in direction to the tramlines, so that 12 and 15 lines were measured in parts A and B, respectively. The subsoiler was driven at a travel speed of 1200 m h1 setting the chisel tip at a d of 0.15 m. After measurement of each line, soil samples were immediately collected from the trench bottom, at each 10 m distance for verification of measured w with the NIR sensor. They were dried in an oven at 105 8C for 24 h to determine w. For the verification of the on-line measured r, samples of soil were taken with a core sampling device, whose cylinder volume was 100 cm3 with 5 cm height. These soil samples were taken before running the subsoiler to eliminate error in measuring r due to soil loosening by the subsoiler. Since the variation of soil compaction of the topsoil was the point of interest of this study, core samples were taken at shallow d by pushing the cylinder bottom within the soil down to a d of about 0.12– 0.15 m. Samples taken after subsoiler running for the determination of w were at very close locations to the samples taken for the determination of r before subsoiler running. 2.4. Geostatistical analysis and map development The variogram estimation and spatial prediction with error (VESPER) 1.6 software, developed by the Australian Centre for Precision Agriculture was used to generate the field maps of D, d, r, w and r error
between the on-line and core sampling methods. All these maps were developed from a 10 m 10 m grid data in order to harmonise the resolution of all maps. The semivariance analysis was performed prior to the development of maps. The global (conventional) kriging option available in VESPER 1.6 software was used, using the semivariance (g(h)) of the entire field area. Utilising the spatial variance structure available in a semivariance, kriging provide the best linear unbiased estimate of an unmeasured value calculated from weighted values measured in a local neighbourhood (Nielsen and Wendroth, 2003). The general equation to calculate the semivariance is written as follows: NðhÞ
gðhÞ ¼
1 X ½Ai ðxi Þ Ai ðxi þ hÞ2 2NðhÞ i¼1
(2)
where h is the distance in m between pair of points N and Ai is the value of a given property at location xi. The semivariogram model is fitted to the data by using the weight nonlinear least-squares method (Jian et al., 1996). Different semivariogram models were evaluated for the different measured properties, such as the spherical, exponential, Gaussian and linear semivariogram models. A semivariogram model is selected when it provides the best fit to g versus h data, determined based on the smaller attained sum of square error. The semivariogram model parameters
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
223
Table 2 Summary of semivariance analysis of different properties Property
Model fit
Nugget (C0)
Sill (C0 + C1)
Range (A1)
Proportion (C1/C0 + C1)
Sum of square error
D d w Core sampling, r On-line, r Error of r
Spherical Spherical Spherical Spherical Spherical Spherical
0.33700 0.00023 0.00016 0.00389 0.00558 0.00237
0.64860 0.00102 0.00069 0.00681 0.00892 0.00473
81.12 80.39 128.50 47.59 75.98 80.72
0.480 0.769 0.761 0.429 0.375 0.498
0.389 10.070 13.780 1.635 2.887 1.146
selected for different properties are given in Table 2, showing that the spherical model provided the best fit for all properties. In order to estimate (krige) the value of A0 at location x0, the following linear function of the known Ai(xi) can be used: A0 ðx0 Þ ¼
n X li Ai ðxi Þ
(3)
i¼1
where n is a set of measurement and li is the weight P factors for the conditions ni¼1 li ¼ 1. These weight factors should have values insuring that the difference between the estimated A0 ðx0 Þ and the true value of A0 (x0) equals zero and the variance of ½A0 ðx0 Þ A0 ðx0 Þ is a minimum, which is obtained when: n X
l j gðxi x j Þ þ m ¼ gðxi x0 Þ
(4)
j¼1
where m is a Lagrangian multiplier. Values of the semivariance of A between sampling locations xi and x j (g(xi x j)) and that between the sampling locations xi and x0 (g(xi x0)) are obtained from the semivariogram model of Eq. (2). The kriging variance (S2k ) can be written as a function of location x0 expressed as follows: S2k ðx0 Þ ¼
n X
l j gðx j x0 Þ þ m
(5)
j¼1
For observation taken in several directions, the lag h distance between xi and x j should be considered a vector instead. The punctual kriging type available in VESPER 1.6 software was selected to develop the different maps. Unlike block kriging that is a technique for averaging observations in a particular domain, punctual kriging utilises the point information in the neighbourhood within a given radius (r0) to calculate values at unsampled locations. The radius adopted for kriging of different maps was 12.
3. Results and discussion After carrying out the simultaneous, on-line measurement of D, d, and w using the system described, data were processed and maps were developed, taking into consideration the identical position for all readings at a given point. Calculation of r was done using D, d, and w values, measured exactly at the same position. 3.1. Measured D Subsoiler D values vary considerably over the field within a variability ranging from 0.86 to 6.37 kN (Table 3), which can be attributed to the variable
Table 3 Sample statistics of different measured properties for map kriging Property
Minimum
Maximum
Mean
Standard deviation
Number of interpolations
D (kN) d (m) w (kg kg1) Core sampling, r (Mg m3) On-line, r (Mg m3) Error of r (Mg m3)
0.856 0.059 0.051 1.207 1.253 0.110
6.370 0.194 0.162 1.701 1.690 0.230
2.950 0.133 0.097 1.519 1.483 0.036
0.791 0.029 0.024 0.081 0.093 0.063
22486 22310 21793 22362 22225 22866
224
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
Fig. 4. Spherical semivariogram model and map of sensor D measured with the single beam load cell based on a 10 m 10 m grid.
distribution and the overlapping effect of the individual independent parameters r, d and w. The spherical model manifests that pairs of measured D remain correlated within a maximum distance of 81.41 m, with a maximum semivariance value (sill) of 0.6486 (Table 2). Examining the spatial variation of r illustrated in Fig. 4 reveals two main zones covering large area of the field with low D and one main zone with large D, near to the border of the field that is parallel to the street (Fig. 3). However, the rest of the field area on the map shows more or less a uniform D distribution with medium values of about 3 kN. 3.2. Measured d In a similar fashion to D variation, measured d with the wheel gauge shows a large range of variation (Table 3), which can be attributed to one of three reasons; variable topography, subsoiler downwards
forces and error of initial setting of subsoiler d at the starting of some measured lines due to tractor wheel resetting on higher or lower soil than the soil level at the subsoiler position. However, the established model of Eq. (1) should be capable to calculate r for any measured d within the range of 0.10–0.37 m, since it was established for variation of this range during the finite element calculation (Mouazen et al., 2003). Similar to D, the spherical model provides the best fit between g(h) and the lag distance h (Fig. 5), with range of 80.39 and a sill of 0.00102 (Table 2). The spatial variation of d illustrates two clearly identified deep zones and one shallow zone (Fig. 5). The deeper soil cutting at the zone that is near to the street (Fig. 3) can contribute to the larger value of D. The area with shallow d coincides remarkably well with the area of registered low D shown in Fig. 4. The very shallow subsoiler d smaller than 0.08 m at this zone makes it possible to interpret the clearly identified low D zone
Fig. 5. Spherical semivariogram model and map of sensor d measured with the wheel gauge based on a 10 m 10 m grid.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
225
Fig. 6. Spherical semivariogram model and map of w measured on-line with the NIR sensor based on 10 m 10 m grid.
as a result of small d values. In general, the correlation between d and D was highest compared to w and r, with a correlation coefficient value (r) of 0.54. 3.3. Measured w Unlike the semivariogram spherical models of D and d, the range (A1) of the spherical model of w was of a larger value, which demands larger lag distance h to be considered (Fig. 6). The measured w over the field shows a clear discrimination between dry and wet zones (Fig. 6), with values ranging between 0.051 and 0.162 kg kg1 (Table 3). A large part of the field had experienced dry soil condition, of which w varied between 0.05 and 0.09 kg kg1. This dry condition is a primary indication of the improper timing for measuring r. The very dry summer of 2003, prevailed not only in Belgium but across Europe, left no chance to capture more water in soil. Under the dry summer conditions, the experiment had to be carried out after wheat harvest and collection of straw bales. In spite of the fact that the dry summer of 2003 led to a decrease in w over a large area of the field, there was still an obvious presence of relatively wet soil zones, particularly at the border of the field near to the street (Figs. 3 and 6). During investigation of the reason for the large compacted area (Fig. 7), particularly in part A shown in Fig. 3, the farmer stated that there was a horse raise 2 years ago on his field. During the raise, the impact of large number of cars and large lorries on soil, accompanied with heavy rain falls created muddy conditions. Therefore, water remained on the soil surface for a long period of time. This might have created a deep compaction, which
could not be eliminated by the conventional tillage the farmer performed over the last two seasons. This deep compaction enhances water accumulation in the upper soil layer due to the impedance of water infiltration, which results in wet zones at that part of the field near to the street (Figs. 3 and 6). Therefore, this part remains relatively wet with w around 0.15 kg kg1 even during the dry summer conditions with large evpotranspiration rates. Almost no correlation was found between D and w with a correlation coefficient r of 0.02, while w was better correlated with d (r = 0.41). Owing to the dry conditions preceding the field experiment that led to a small range of w, it is difficult to establish any correlation between w and D. Under a larger variation range of w, Fielke et al. (1993) reported a D reduction of a cultivator by 30–50% under wet soil conditions in comparison to dry soil conditions. Furthermore, within the upper 0.2 m soil layer, the deeper the soil the larger is w due to the large evaporation from the topsoil. 3.4. Measured r with core sampling method The map of r shows a huge variation over the area of the field, ranging from very loose zones of 1.21 Mg m3 to very compacted zones of 1.70 Mg m3 (Fig. 7). In fact, a large part of the field was extremely compacted, which can also be indirectly attributed to the deep compaction due to the horse raise held on the field two seasons ago. This deep compaction assisted in enhancing the compaction in the upper soil layers due to the bad water infiltration. Accumulation of the heavy rain water on
226
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
Fig. 7. Spherical semivariogram model and of r measured with core sampling method based on a 10 m 10 m grid.
the soil surface is very likely to create compaction in the upper soil. Partial correlation can be established between the loose zone (small r values) shown in the middle of part A (Fig. 7) and the identical zone with low D (Fig. 4). Furthermore, the large D measured at the border of the field with the street can be partially the result of large r values in addition to a deeper penetration. In general, the correlation between r and D was found to be of a value of 0.52. 3.5. Measured r with on-line method By substitution of the measured values of d, D and w in Eq. (1), r was calculated to produce the map shown in Fig. 8. The figure demonstrates a very similar spatial distribution of soil r to the r map developed using the core sampling method (Fig. 7), particularly the highly compacted zones. However, similar to the conventional method of measuring bulk density by core sampling
method, the on-line measured r failed to indicate compaction at a very dry zone, the zone determined within the rectangular in Fig. 8. This conclusion is supported by examining the error map of r between core sampling and on-line measurement methods, shown in Fig. 9. The error map indicates that the largest error zones coincide well with the dry soil zones and no large values of error are recognised outside the dry zones. However, the error ranges between 0.11 and 0.23 Mg m3 (Fig. 9 and Table 3). The linear correlation between r measured with the core sampling and on-line methods is illustrated in Fig. 10. The negative effect of the dry soil conditions during on-line measurement of r affected the correlation between the two methods, resulting in a correlation coefficient r of 0.75 and a standard error of 0.054 Mg m3. The histogram plot of the r error, shown in Fig. 11 between the two measurement methods reveals that about 72% of the errors were less than 0.05 Mg m3 in absolute values. Error distribution skewness in the
Fig. 8. Spherical semivariogram model and map of on-line measured r based on a 10 m 10 m grid.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
227
Fig. 9. Spherical semivariogram model and map of measured r error between the on-line and core sampling methods based on a 10 m 10 m grid.
Fig. 10. Linear correlation of measured r between on-line and core measurement methods.
Fig. 11. Histogram plot of r error between on-line and core measurement methods.
228
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
negative range can be observed in the histogram plot of error (Fig. 11), which can be attributed to the low values of w measured at some parts across the very dry soil zones. The low values of w, substituted into Eq. (1) lead to smaller values of r in comparison to the corresponding values at the same points measured with the core sampling method. However, the overall mean error of on-line measured r was small (2.3%). This ensures that on-line, simultaneous measurement of D, d and w together with the mathematical model implicated comprises a properly accurate method for on-line measurement of r. When on-line measurement of r has to be carried out under very dry conditions a relatively large error should be expected, and hence, compacted zones at dry zones cannot be detected correctly.
4. Conclusions A sensor system composed of several sensors developed for the on-line measurement of r was evaluated. On-line measurements of draught (D) and depth (d) were performed with a single ended shear beam load cell and a metal wheel equipped with a linear variable differential transducer sensors, respectively. The soil moisture content (w) was measured online with a fiber-type visible, near infrared spectrophotometer sensor. The sensor system performed online, simultaneous measurements of D, d and w, whose values were utilised to calculate r by using a previously developed mathematical model, while implement position was recorded with a differential global positioning system (GPS) and radar. No clear correlation could be observed between different soil properties and measured D, which is in line with conclusions drown in a preceding study (Mouazen et al., 2003). This supports the conclusion of that the field measurement is not the best procedure to correlate D with the other influencing parameters (d, w and r), and theoretical modelling is still an alternative option. The poor correlation was particularly clear between D and w, which was attributed to the extremely dry conditions during measurement resulting in a small variation range in w. In general, deeper soil cutting led to larger D and visa versa, whereas large r resulted in larger D. Overlap of the effect of different parameters during on-line measure-
ment of D emphasises that D cannot be considered as a correct parameter to indicate soil compaction, ignoring the other influencing factors. The geostatistical analysis of semivariance manifest that all properties could be simulated by spherical models, providing the best fit between semivariance (g(h)) and lag h. The on-line measured r map was reasonably similar to the measured r using the core sampling method with correlation coefficient r and standard error values of 0.75 and 0.054 Mg m3, respectively. This was particularly distinguishable at very loose and highly compacted zones. Similarity between r maps degraded at the very dry zones, where the compacted zone was not detected properly. The histogram plot of error between the two r maps showed that about 72% of the errors were less than 0.05 Mg m3 in absolute values. The overall mean error of on-line measured r was of a small value of 2.3%. This provides a proof of the method accuracy for the on-line measurement of r, when D, d and w are measured simultaneously and substituted into a mathematical model to calculate r. However, measurement under very dry conditions should be minimised, because it can lead to a relatively large underestimation of r, and hence compacted zones at dry zones cannot be detected correctly.
References Adamchuk, V.I., Morgan, M.T., Sumali, H., 2001. Application of a strain gauge array to estimate soil mechanical impedence onthe-go. Trans. ASAE 44 (6), 1377–1383. Adamchuk, V.I., Hummel, J.W., Morgan, M.T., Upadhyaya, S.K., 2004. On-the-go soil sensors for precision agriculture. Comput. Electron. Agric. 44, 71–91. Besson, A., Cousin, I., Samoue¨lian, A., Boizard, H., Richard, G., 2004. Structure heterogeneity of the soil tilled layer as characterized by 2D electrical resistivity survey. Soil Till. Res. 79 (2), 239–249. Fleige, H., Horn, R., 2000. Field experiments of the effect of soil compaction on soil properties, runoff, interflow and erosion. In: Horn, R., et al. (Eds.), Subsoil Compaction Distribution, Processes and Consequences. Advance in GeoEcology, vol. 32. CATENA VERLAG, Reiskirchen, Germany, pp. 258–268. Fielke, J.M., Riley, T.W., Slattery, M.G., Fitzpatrick, R.W., 1993. Comparison of tillage forces and wear rates of pressed and cast cultivator shares. Soil Till. Res. 25, 317–328. Glancey, J.L., Upadhyaya, S.K., Chancellor, W.J., Rumsey, J.W., 1989. An instrumented chisel for the study of soil-tillage dynamics. Soil Till. Res. 14, 1–24.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229 Glancey, J.L., Upadhyaya, S.K., Chancellor, W.J., Rumsey, J.W., 1996. Prediction of agricultural implement draft using an instrumented analog tillage tool. Soil Till. Res. 37, 47–65. Godwin, R.J., 1975. An extended octagonal ring transducer for use in tillage studies. J. Agric. Eng. Res. 347–367. Godwin, R.J., Miller, P.C.H., 2003. A review of the technologies for mapping within-field variability. Biosyst. Eng. 84 (4), 393–407. Jian, X., Olea, R.A., Yu, Y-S., 1996. Semivariogram modelling by weighted least squares. Comput. Geosci. 22, 387–397. Liu, W., Upadhyaya, S.K., Kataoka, T., Shibusawa, S., 1996. Development of a texture/soil compaction sensor. In: Robert, P.C., Rust, R.H., Larson, W.E. (Eds.), Proceedings of the Third International Conference on Precision Agriculture, American Society of Agronomy, Crop Science Society of America and Soil Science Society of America, Wisconsin, USA, pp. 617–630. Mouazen, A.M., Dumont, K., Maertens, K., Ramon, H., 2003. Twodimensional prediction of spatial variation in topsoil compaction of a sandy loam field based on measured horizontal force of compaction sensor, cutting depth and moisture content. Soil Till. Res. 74 (1), 91–102. Mouazen, A.M., De Baerdemaeker, J., Ramon, H., 2004a. Towards development of on-line soil moisture sensor using a fiber-type NIR spectrophotometer. Soil Till. Res. 80 (1–2), 171–183. Mouazen, A.M., Anthonis, J., Saeys, W., Ramon, H., 2004b. An automatic depth control system for on-line measurement of spatial variation in soil compaction. Part 1. Sensor design for measurement of frame distance variation from soil surface. Biosyst. Eng. 89 (2), 139–150.
229
Nielsen, D.R., Wendroth, O., 2003. Spatial and Temporal Statistics, Sampling Field Soils and their Vegitation. GeoEcology Textbook. CATENA VERLAG, GeoScience Publisher, Reiskirchen, Germany. Schmidhalter, U., Zintel, A., Neudecker, E., 2001. Calibration of electromagnetic induction measurements to survey the spatial variability of soils.. In: Grenier, G., Blackmore, S. (Eds.), Proceedings of the Third European Conference on Precision Agriculture, agro Montpellier, Montpellier, France, pp. 479– 484. Sirjacobs, D., Hanquet, B., Lebeau, F., Destain, M.F., 2002. On-line soil mechanical resistance mapping and correlation with soil physical properties for precision agriculture. Soil Till. Res. 64, 231–242. Smith, J., 2001. Soil Electrical Conductivity Mapping and Interpretation. In-field Mapping of Soil and Crop Factors. SCI, London, UK. Sprinkle, L.W., Langston, T.D., Weber, J.A., Sharon, N.M., 1970. A similitude study with static and dynamic parameters in an artificial soil. Trans. Am. Soc. Agric. Eng. 29, 580–586. Upadhyaya, S.K., Williams, T.H., Kemble, L.J., Collins, N.E., 1984. Energy requirement for chiselling in coastal plain soils. Trans. Am. Soc. Agric. Eng. 27 (6), 1643–1649. Verschoore, R., Pieters, J.G., Seps, T., Spriet, Y., Vangeyte, J., 2003. Development of a sensor for continuous soil resistance measurement. In: Stafford, J.V. (Ed.), Proceedings of the Joint European Conference of ECPA-ECPLF, Wageningen Academic Publishers, Berlin, Germany, pp. 689–695.
Development of on-line measurement system of bulk density based on on-line measured draught, depth and soil moisture content Abdul Mounem Mouazen a,b,*, Herman Ramon a a
Department of Agro-Engineering and -Economics, Faculty of Agricultural and Applied Biological Sciences, Kasteelpark Arenberg 30, B-3001 Heverlee, Belgium b Department of Rural Engineering, Faculty of Agriculture, University of Aleppo, P.O. Box 12214, Aleppo, Syria Received 17 November 2004; received in revised form 16 February 2005; accepted 22 February 2005
Abstract On-line measurement of soil compaction is needed for site specific tillage management. The soil bulk density (r) indicating soil compaction was measured on-line by means of a developed compaction sensor system that comprised several sensors for on-line measurement of the draught (D) of a soil cutting tool (subsoiler), the soil cutting depth (d) and the soil moisture content (w). The subsoiler D was measured with a single shear beam load cell, whereas d was measured with a wheel gauge that consisted of a swinging arm metal wheel and a linear variable differential transducer (LVDT). The soil w was measured with a near infrared fibre-type spectrophotometer sensor. These on-line three measured parameters were used to calculate r, by utilising a hybrid numerical– statistical mathematical model developed in a previous study. Punctual kriging was performed using the variogram estimation and spatial prediction with error (VESPER) 1.6 software to develop the field maps of r, soil w, subsoiler d and D, based on 10 m 10 m grid.To verify theon-linemeasuredr map,this map was compared withthemap measuredby theconventional core samplingmethod. The spherical semivariogram models, providing the best fit for all properties was used for kriging of different maps. Maps developed showed that no clear correlation could be detected between different parameters measured and subsoiler D. However, the D value was smaller at shallow penetration d, whereas large D coincided with large r values at few positions in the field. Maps of r measured with the core sampling and on-line methods were similar, with correlation coefficient (r) and the standard error values of 0.75 and 0.054 Mg m3, respectively. On-line measured r exhibited larger errors at very dry zones. The normal distribution of the r error between the two different measurement methods showed that about 72% of the errors were less than 0.05 Mg m3 in absolute values. However, the overall mean error of on-line measured r was of a small value of 2.3%, which ensures the method accuracy for on-line measurement of r. Measurement under very dry conditions should be minimised, because it can lead to a relatively large error, and hence, compacted zones at dry zones cannot be detected correctly. # 2005 Elsevier B.V. All rights reserved. Keywords: Bulk density; On-line measurement; Sensor; Moisture content; Draught; Depth
* Corresponding author. Tel.: +32 163 285 92; fax: +32 163 285 90. E-mail addresses: [email protected], [email protected] (A.M. Mouazen). 0167-1987/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.still.2005.02.026
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
1. Introduction Soil compaction is one of the main factors limiting plant growth and crop yield. It is well-known that soil compaction is a natural or human created problem that results in increase of bulk density (r) and root penetration resistance, while decrease in void ratio and available nutrients, water and oxygen for plant can very frequently take place. Moreover, it was found by Fleige and Horn (2000) that anthropogenic compaction in traffic rut and plough pan caused a reduction of coarse pores and saturated water conductivity, leading to surface runoff and soil erosion. Alleviation of the negative effect of soil compaction on plant growth, along with the reduction of tillage energy input and increasing the surface plant residue relay on successfully managed site specific tillage operations. With the recent development in precision agricultural technologies, researchers have focused on development of on-line sensors to measure different properties in agricultural soils (Adamchuk et al., 2004). Since online measurement of soil compaction demands simultaneous measurement of different influencing parameters and a proper mathematical modelling technique, development of an on-line soil compaction sensor is still a challenging issue for researchers involved in new technologies and engineering sections of precision agriculture. The most common techniques to determine the degree of soil compaction is the field measurement of r and penetration resistance. Both parameters are measured with traditional methods under static conditions, using core sampling methods and penetrometers, respectively. Under dry soil conditions, these measurements are very difficult and time costly procedures, in addition to the discontinuous data output provided based on fine or coarse measurement grids. Soil compaction was referred to indirectly as on-line measured soil mechanical resistance or draught (D) of different cutting or penetration tools using different load cells (Sprinkle et al., 1970; Upadhyaya et al., 1984; Glancey et al., 1996; Sirjacobs et al., 2002; Verschoore et al., 2003) or strain gauges (Glancey et al., 1989; Adamchuk et al., 2001). Godwin and Miller (2003) stated that there is now, from commercial sources, evidence that the electromagnetic induction (EMI) will distinguish between different levels of soil compaction (Smith,
219
2001). However, EMI is mainly dependent on the clay content, salinity, organic matter, moisture content (w) and r (Schmidhalter et al., 2001), which introduce problems of discrimination of different factors. In a recently published study, Besson et al. (2004) related the soil electrical resistivity to soil r, showing successful discrimination between loose and compacted layers. In addition to the sensitivity of the electrical resistivity to salinity and ambient temperature, the developed electrical resistivity– bulk density relationship depended on the soil type and w. When r is selected for on-line, tractor-based measurement of soil compaction with a soil cutting tool or penetration device, all influencing parameters should be measured simultaneously during measurement, namely D, depth (d) and w. Liu et al. (1996) developed a real-time texture/compaction sensor. They stated that if d and the speed of a soil cutting tine are kept constant, D will be a function of texture, r and w. Based on a combination of the finite element method and multiple linear regression analysis, Mouazen et al. (2003) developed the following model for the calculation of soil compaction indicated as bulk density:
r¼
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi! 2 3 D þ 21:36 w 73:9313 d 1:14 1:6734
(1)
where D is the draught in kN, w the soil moisture content in kg kg1, d the subsoiler depth in m and r is the bulk density in Mg m3. The mathematical model of Eq. (1) stablishes a basis of carrying out an indirect measurement of r, when the independent parameters (D, d and w) can be measured simultaneously on-line. So far, Mouazen et al. (2003) produced r maps by using Eq. (1) and measured D of the subsoiler (cutting tool) with an extended octagonal load cell, similar to the load cell developed by Godwin (1975). However, soil w and subsoiler d were measured using traditional techniques, so that w was measured with oven drying methods and d was measured manually. But, the online measurement of r will only be satisfied, if all the independent parameters (D, d and w) are measured online simultaneously, with values to be substituted into Eq. (1) to obtain r.
220
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
This study aims at introducing an on-line measurement system of r, using on-line, simultaneously measured D, d and w values and a mathematical model. The reliability of an on-line developed r map is investigated by performing a comparison with the r map measured with the traditional core sampling method.
Table 1 Soil texture defined according to the USDA soil classification at a depth of 0–0.02 m Property
Sandy loam (g kg1)
Loam (g kg1)
Sand (>50 mm) Silt (2–50 mm) Clay (<2 mm)
562.77 362.41 74.82
464.24 443.74 92.02
2.2. Sensor system 2. Materials and methods 2.1. Site description The experimental field of 2.3 ha was located in the Zoutleeuw region, southeast of Brussels, Belgium. The soil type is an Arenic Cambisol, according to the FAO classification. The soil texture fractions were determined by a combination of wet sieve and hydrometer tests, using the USDA Soil Classification, as shown in Table 1. The soil texture over the field down to 0.20 m is non-homogeneous, including two textures of sandy loam and loam, as listed in Table 1. Samples were collected from the upper 0.2 m layer during the growing season of wheat based on 50 m 50 m grid, and they were subjected to texture analysis and spectrophotometer sensor calibration.
A standard medium–deep subsoiler used as soil cutting tool was attached to a frame, which was mounted onto the three point hitch of the tractor. The subsoiler consisted of two parts; the chisel of 0.06 m width, and the shank of 0.03 m width.
2.2.1. Sensor for D measurement For simplicity, the extended octagonal load cell used to measure the subsoiler D (Mouazen et al., 2003) was replaced by a commercially available single ended shear beam load cell from CELTRON TECHNOLOGIES Inc. The load cell was inserted within a gap between the subsoiler shank backside and a fixing plate, as shown in Fig. 1. When the shank was pushed backwards by means of the soil resistance
Fig. 1. Set up of r sensor ready for on-line field measurement.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
force opposite to the direction of travel, the load cell is subjected to bending and the output signal is recorded. 2.2.2. Sensor for w measurement The on-line measurement of w is the most difficult parameter needed for this study. A Corona fibre-type visible (VIS) and near infrared (NIR) spectrophotometer from Zeiss company, with a light reflectance measurement range of 306.5–1710.9 nm was used to measure w in the field. The spectrophotometer optical unit was attached to the subsoiler chisel backside to perform the light reflectance measurement from the soil surface of the bottom of the trench opened by the preceded chisel (Fig. 1). More detailed description about the sensor can be found in a previous published study (Mouazen et al., 2004a). 2.2.3. Sensor for d measurement A metal wheel gauge mounted on the frame with a revolute joint was designed to measure the distance variation between the soil surface and the frame. A linear variable differential transducer (LVDT) with a 0.2 m stroke length was used to connect the frame to the axle of a metal wheel of the height sensor, as show in Fig. 2. The performance of the wheel gauge on four different surfaces was investigated in detail by
221
Mouazen et al. (2004b). They found that in comparison with an ultrasonic sensor, the wheelLVDT sensor can be used more accurately than the ultrasonic sensor to measure the subsoiler d, in fields covered with plant residues and stubble. 2.2.4. Sensor electrical system The electrical system, besides the spectrophotometer, consisted of several modules: a basic power supply, travel speed sensor, global positioning system, signal conditioning system, amplifier and data acquisition system. The travel speed was measured using a doppler radar that was mounted on the frame pointing backwards to avoid the effects of stubble or grass movement after the measurement frame passed. The accuracy of the sensor was tested in previous experiments and all errors were smaller than 2.5%. Position, latitude and longitude, were determined with a Trimble AgGPS132 differential global positioning system (DGPS). The antenna was mounted just above the sensor, to obtain a sub-meter accuracy. Using the National Instruments Labview programming language, a custom-built data acquisition software was developed. The speed sensor signal was acquired at a frequency of 200 Hz, while the DGPS information was stored at 1 Hz. The load cell, LVDT together with the speed sensor signal and the DGPS information were logged by use of a Pentium III 800 MHz laptop computer equipped with a National Instruments DAQ700 data acquisition card. An EURORAC data logging system was designed, through which the signals of load cell, LVDT and radar are filtered, amplified and transferred to the Pentium III 800 MHz laptop computer. The spectral signal was acquired and stored at 0.225 Hz. Each spectrum was an average of five successive spectra. They were logged to a separate Pentium IV, 1.7 GHz laptop computer equipped with RS422 cable and PCMCIA acquisition card. 2.3. On-line field experiment
Fig. 2. Wheel gauge consisted of a swinging arm metal wheel and a LVDT sensor.
The on-line field measurement was carried out in the Zoutleeuw field, the same field where soil samples were collected for texture analysis and spectrophotometer sensor calibration. The experiment was carried out after harvesting wheat. After setting up the different sensors, the subsoiler was pulled throughout parallel lines of 10 m apart, as shown in Fig. 3. These
222
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
Fig. 3. Sampling and map design based on a 10 m 10 m grid.
lines were perpendicular in direction to the tramlines, so that 12 and 15 lines were measured in parts A and B, respectively. The subsoiler was driven at a travel speed of 1200 m h1 setting the chisel tip at a d of 0.15 m. After measurement of each line, soil samples were immediately collected from the trench bottom, at each 10 m distance for verification of measured w with the NIR sensor. They were dried in an oven at 105 8C for 24 h to determine w. For the verification of the on-line measured r, samples of soil were taken with a core sampling device, whose cylinder volume was 100 cm3 with 5 cm height. These soil samples were taken before running the subsoiler to eliminate error in measuring r due to soil loosening by the subsoiler. Since the variation of soil compaction of the topsoil was the point of interest of this study, core samples were taken at shallow d by pushing the cylinder bottom within the soil down to a d of about 0.12– 0.15 m. Samples taken after subsoiler running for the determination of w were at very close locations to the samples taken for the determination of r before subsoiler running. 2.4. Geostatistical analysis and map development The variogram estimation and spatial prediction with error (VESPER) 1.6 software, developed by the Australian Centre for Precision Agriculture was used to generate the field maps of D, d, r, w and r error
between the on-line and core sampling methods. All these maps were developed from a 10 m 10 m grid data in order to harmonise the resolution of all maps. The semivariance analysis was performed prior to the development of maps. The global (conventional) kriging option available in VESPER 1.6 software was used, using the semivariance (g(h)) of the entire field area. Utilising the spatial variance structure available in a semivariance, kriging provide the best linear unbiased estimate of an unmeasured value calculated from weighted values measured in a local neighbourhood (Nielsen and Wendroth, 2003). The general equation to calculate the semivariance is written as follows: NðhÞ
gðhÞ ¼
1 X ½Ai ðxi Þ Ai ðxi þ hÞ2 2NðhÞ i¼1
(2)
where h is the distance in m between pair of points N and Ai is the value of a given property at location xi. The semivariogram model is fitted to the data by using the weight nonlinear least-squares method (Jian et al., 1996). Different semivariogram models were evaluated for the different measured properties, such as the spherical, exponential, Gaussian and linear semivariogram models. A semivariogram model is selected when it provides the best fit to g versus h data, determined based on the smaller attained sum of square error. The semivariogram model parameters
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
223
Table 2 Summary of semivariance analysis of different properties Property
Model fit
Nugget (C0)
Sill (C0 + C1)
Range (A1)
Proportion (C1/C0 + C1)
Sum of square error
D d w Core sampling, r On-line, r Error of r
Spherical Spherical Spherical Spherical Spherical Spherical
0.33700 0.00023 0.00016 0.00389 0.00558 0.00237
0.64860 0.00102 0.00069 0.00681 0.00892 0.00473
81.12 80.39 128.50 47.59 75.98 80.72
0.480 0.769 0.761 0.429 0.375 0.498
0.389 10.070 13.780 1.635 2.887 1.146
selected for different properties are given in Table 2, showing that the spherical model provided the best fit for all properties. In order to estimate (krige) the value of A0 at location x0, the following linear function of the known Ai(xi) can be used: A0 ðx0 Þ ¼
n X li Ai ðxi Þ
(3)
i¼1
where n is a set of measurement and li is the weight P factors for the conditions ni¼1 li ¼ 1. These weight factors should have values insuring that the difference between the estimated A0 ðx0 Þ and the true value of A0 (x0) equals zero and the variance of ½A0 ðx0 Þ A0 ðx0 Þ is a minimum, which is obtained when: n X
l j gðxi x j Þ þ m ¼ gðxi x0 Þ
(4)
j¼1
where m is a Lagrangian multiplier. Values of the semivariance of A between sampling locations xi and x j (g(xi x j)) and that between the sampling locations xi and x0 (g(xi x0)) are obtained from the semivariogram model of Eq. (2). The kriging variance (S2k ) can be written as a function of location x0 expressed as follows: S2k ðx0 Þ ¼
n X
l j gðx j x0 Þ þ m
(5)
j¼1
For observation taken in several directions, the lag h distance between xi and x j should be considered a vector instead. The punctual kriging type available in VESPER 1.6 software was selected to develop the different maps. Unlike block kriging that is a technique for averaging observations in a particular domain, punctual kriging utilises the point information in the neighbourhood within a given radius (r0) to calculate values at unsampled locations. The radius adopted for kriging of different maps was 12.
3. Results and discussion After carrying out the simultaneous, on-line measurement of D, d, and w using the system described, data were processed and maps were developed, taking into consideration the identical position for all readings at a given point. Calculation of r was done using D, d, and w values, measured exactly at the same position. 3.1. Measured D Subsoiler D values vary considerably over the field within a variability ranging from 0.86 to 6.37 kN (Table 3), which can be attributed to the variable
Table 3 Sample statistics of different measured properties for map kriging Property
Minimum
Maximum
Mean
Standard deviation
Number of interpolations
D (kN) d (m) w (kg kg1) Core sampling, r (Mg m3) On-line, r (Mg m3) Error of r (Mg m3)
0.856 0.059 0.051 1.207 1.253 0.110
6.370 0.194 0.162 1.701 1.690 0.230
2.950 0.133 0.097 1.519 1.483 0.036
0.791 0.029 0.024 0.081 0.093 0.063
22486 22310 21793 22362 22225 22866
224
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
Fig. 4. Spherical semivariogram model and map of sensor D measured with the single beam load cell based on a 10 m 10 m grid.
distribution and the overlapping effect of the individual independent parameters r, d and w. The spherical model manifests that pairs of measured D remain correlated within a maximum distance of 81.41 m, with a maximum semivariance value (sill) of 0.6486 (Table 2). Examining the spatial variation of r illustrated in Fig. 4 reveals two main zones covering large area of the field with low D and one main zone with large D, near to the border of the field that is parallel to the street (Fig. 3). However, the rest of the field area on the map shows more or less a uniform D distribution with medium values of about 3 kN. 3.2. Measured d In a similar fashion to D variation, measured d with the wheel gauge shows a large range of variation (Table 3), which can be attributed to one of three reasons; variable topography, subsoiler downwards
forces and error of initial setting of subsoiler d at the starting of some measured lines due to tractor wheel resetting on higher or lower soil than the soil level at the subsoiler position. However, the established model of Eq. (1) should be capable to calculate r for any measured d within the range of 0.10–0.37 m, since it was established for variation of this range during the finite element calculation (Mouazen et al., 2003). Similar to D, the spherical model provides the best fit between g(h) and the lag distance h (Fig. 5), with range of 80.39 and a sill of 0.00102 (Table 2). The spatial variation of d illustrates two clearly identified deep zones and one shallow zone (Fig. 5). The deeper soil cutting at the zone that is near to the street (Fig. 3) can contribute to the larger value of D. The area with shallow d coincides remarkably well with the area of registered low D shown in Fig. 4. The very shallow subsoiler d smaller than 0.08 m at this zone makes it possible to interpret the clearly identified low D zone
Fig. 5. Spherical semivariogram model and map of sensor d measured with the wheel gauge based on a 10 m 10 m grid.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
225
Fig. 6. Spherical semivariogram model and map of w measured on-line with the NIR sensor based on 10 m 10 m grid.
as a result of small d values. In general, the correlation between d and D was highest compared to w and r, with a correlation coefficient value (r) of 0.54. 3.3. Measured w Unlike the semivariogram spherical models of D and d, the range (A1) of the spherical model of w was of a larger value, which demands larger lag distance h to be considered (Fig. 6). The measured w over the field shows a clear discrimination between dry and wet zones (Fig. 6), with values ranging between 0.051 and 0.162 kg kg1 (Table 3). A large part of the field had experienced dry soil condition, of which w varied between 0.05 and 0.09 kg kg1. This dry condition is a primary indication of the improper timing for measuring r. The very dry summer of 2003, prevailed not only in Belgium but across Europe, left no chance to capture more water in soil. Under the dry summer conditions, the experiment had to be carried out after wheat harvest and collection of straw bales. In spite of the fact that the dry summer of 2003 led to a decrease in w over a large area of the field, there was still an obvious presence of relatively wet soil zones, particularly at the border of the field near to the street (Figs. 3 and 6). During investigation of the reason for the large compacted area (Fig. 7), particularly in part A shown in Fig. 3, the farmer stated that there was a horse raise 2 years ago on his field. During the raise, the impact of large number of cars and large lorries on soil, accompanied with heavy rain falls created muddy conditions. Therefore, water remained on the soil surface for a long period of time. This might have created a deep compaction, which
could not be eliminated by the conventional tillage the farmer performed over the last two seasons. This deep compaction enhances water accumulation in the upper soil layer due to the impedance of water infiltration, which results in wet zones at that part of the field near to the street (Figs. 3 and 6). Therefore, this part remains relatively wet with w around 0.15 kg kg1 even during the dry summer conditions with large evpotranspiration rates. Almost no correlation was found between D and w with a correlation coefficient r of 0.02, while w was better correlated with d (r = 0.41). Owing to the dry conditions preceding the field experiment that led to a small range of w, it is difficult to establish any correlation between w and D. Under a larger variation range of w, Fielke et al. (1993) reported a D reduction of a cultivator by 30–50% under wet soil conditions in comparison to dry soil conditions. Furthermore, within the upper 0.2 m soil layer, the deeper the soil the larger is w due to the large evaporation from the topsoil. 3.4. Measured r with core sampling method The map of r shows a huge variation over the area of the field, ranging from very loose zones of 1.21 Mg m3 to very compacted zones of 1.70 Mg m3 (Fig. 7). In fact, a large part of the field was extremely compacted, which can also be indirectly attributed to the deep compaction due to the horse raise held on the field two seasons ago. This deep compaction assisted in enhancing the compaction in the upper soil layers due to the bad water infiltration. Accumulation of the heavy rain water on
226
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
Fig. 7. Spherical semivariogram model and of r measured with core sampling method based on a 10 m 10 m grid.
the soil surface is very likely to create compaction in the upper soil. Partial correlation can be established between the loose zone (small r values) shown in the middle of part A (Fig. 7) and the identical zone with low D (Fig. 4). Furthermore, the large D measured at the border of the field with the street can be partially the result of large r values in addition to a deeper penetration. In general, the correlation between r and D was found to be of a value of 0.52. 3.5. Measured r with on-line method By substitution of the measured values of d, D and w in Eq. (1), r was calculated to produce the map shown in Fig. 8. The figure demonstrates a very similar spatial distribution of soil r to the r map developed using the core sampling method (Fig. 7), particularly the highly compacted zones. However, similar to the conventional method of measuring bulk density by core sampling
method, the on-line measured r failed to indicate compaction at a very dry zone, the zone determined within the rectangular in Fig. 8. This conclusion is supported by examining the error map of r between core sampling and on-line measurement methods, shown in Fig. 9. The error map indicates that the largest error zones coincide well with the dry soil zones and no large values of error are recognised outside the dry zones. However, the error ranges between 0.11 and 0.23 Mg m3 (Fig. 9 and Table 3). The linear correlation between r measured with the core sampling and on-line methods is illustrated in Fig. 10. The negative effect of the dry soil conditions during on-line measurement of r affected the correlation between the two methods, resulting in a correlation coefficient r of 0.75 and a standard error of 0.054 Mg m3. The histogram plot of the r error, shown in Fig. 11 between the two measurement methods reveals that about 72% of the errors were less than 0.05 Mg m3 in absolute values. Error distribution skewness in the
Fig. 8. Spherical semivariogram model and map of on-line measured r based on a 10 m 10 m grid.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
227
Fig. 9. Spherical semivariogram model and map of measured r error between the on-line and core sampling methods based on a 10 m 10 m grid.
Fig. 10. Linear correlation of measured r between on-line and core measurement methods.
Fig. 11. Histogram plot of r error between on-line and core measurement methods.
228
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229
negative range can be observed in the histogram plot of error (Fig. 11), which can be attributed to the low values of w measured at some parts across the very dry soil zones. The low values of w, substituted into Eq. (1) lead to smaller values of r in comparison to the corresponding values at the same points measured with the core sampling method. However, the overall mean error of on-line measured r was small (2.3%). This ensures that on-line, simultaneous measurement of D, d and w together with the mathematical model implicated comprises a properly accurate method for on-line measurement of r. When on-line measurement of r has to be carried out under very dry conditions a relatively large error should be expected, and hence, compacted zones at dry zones cannot be detected correctly.
4. Conclusions A sensor system composed of several sensors developed for the on-line measurement of r was evaluated. On-line measurements of draught (D) and depth (d) were performed with a single ended shear beam load cell and a metal wheel equipped with a linear variable differential transducer sensors, respectively. The soil moisture content (w) was measured online with a fiber-type visible, near infrared spectrophotometer sensor. The sensor system performed online, simultaneous measurements of D, d and w, whose values were utilised to calculate r by using a previously developed mathematical model, while implement position was recorded with a differential global positioning system (GPS) and radar. No clear correlation could be observed between different soil properties and measured D, which is in line with conclusions drown in a preceding study (Mouazen et al., 2003). This supports the conclusion of that the field measurement is not the best procedure to correlate D with the other influencing parameters (d, w and r), and theoretical modelling is still an alternative option. The poor correlation was particularly clear between D and w, which was attributed to the extremely dry conditions during measurement resulting in a small variation range in w. In general, deeper soil cutting led to larger D and visa versa, whereas large r resulted in larger D. Overlap of the effect of different parameters during on-line measure-
ment of D emphasises that D cannot be considered as a correct parameter to indicate soil compaction, ignoring the other influencing factors. The geostatistical analysis of semivariance manifest that all properties could be simulated by spherical models, providing the best fit between semivariance (g(h)) and lag h. The on-line measured r map was reasonably similar to the measured r using the core sampling method with correlation coefficient r and standard error values of 0.75 and 0.054 Mg m3, respectively. This was particularly distinguishable at very loose and highly compacted zones. Similarity between r maps degraded at the very dry zones, where the compacted zone was not detected properly. The histogram plot of error between the two r maps showed that about 72% of the errors were less than 0.05 Mg m3 in absolute values. The overall mean error of on-line measured r was of a small value of 2.3%. This provides a proof of the method accuracy for the on-line measurement of r, when D, d and w are measured simultaneously and substituted into a mathematical model to calculate r. However, measurement under very dry conditions should be minimised, because it can lead to a relatively large underestimation of r, and hence compacted zones at dry zones cannot be detected correctly.
References Adamchuk, V.I., Morgan, M.T., Sumali, H., 2001. Application of a strain gauge array to estimate soil mechanical impedence onthe-go. Trans. ASAE 44 (6), 1377–1383. Adamchuk, V.I., Hummel, J.W., Morgan, M.T., Upadhyaya, S.K., 2004. On-the-go soil sensors for precision agriculture. Comput. Electron. Agric. 44, 71–91. Besson, A., Cousin, I., Samoue¨lian, A., Boizard, H., Richard, G., 2004. Structure heterogeneity of the soil tilled layer as characterized by 2D electrical resistivity survey. Soil Till. Res. 79 (2), 239–249. Fleige, H., Horn, R., 2000. Field experiments of the effect of soil compaction on soil properties, runoff, interflow and erosion. In: Horn, R., et al. (Eds.), Subsoil Compaction Distribution, Processes and Consequences. Advance in GeoEcology, vol. 32. CATENA VERLAG, Reiskirchen, Germany, pp. 258–268. Fielke, J.M., Riley, T.W., Slattery, M.G., Fitzpatrick, R.W., 1993. Comparison of tillage forces and wear rates of pressed and cast cultivator shares. Soil Till. Res. 25, 317–328. Glancey, J.L., Upadhyaya, S.K., Chancellor, W.J., Rumsey, J.W., 1989. An instrumented chisel for the study of soil-tillage dynamics. Soil Till. Res. 14, 1–24.
A.M. Mouazen, H. Ramon / Soil & Tillage Research 86 (2006) 218–229 Glancey, J.L., Upadhyaya, S.K., Chancellor, W.J., Rumsey, J.W., 1996. Prediction of agricultural implement draft using an instrumented analog tillage tool. Soil Till. Res. 37, 47–65. Godwin, R.J., 1975. An extended octagonal ring transducer for use in tillage studies. J. Agric. Eng. Res. 347–367. Godwin, R.J., Miller, P.C.H., 2003. A review of the technologies for mapping within-field variability. Biosyst. Eng. 84 (4), 393–407. Jian, X., Olea, R.A., Yu, Y-S., 1996. Semivariogram modelling by weighted least squares. Comput. Geosci. 22, 387–397. Liu, W., Upadhyaya, S.K., Kataoka, T., Shibusawa, S., 1996. Development of a texture/soil compaction sensor. In: Robert, P.C., Rust, R.H., Larson, W.E. (Eds.), Proceedings of the Third International Conference on Precision Agriculture, American Society of Agronomy, Crop Science Society of America and Soil Science Society of America, Wisconsin, USA, pp. 617–630. Mouazen, A.M., Dumont, K., Maertens, K., Ramon, H., 2003. Twodimensional prediction of spatial variation in topsoil compaction of a sandy loam field based on measured horizontal force of compaction sensor, cutting depth and moisture content. Soil Till. Res. 74 (1), 91–102. Mouazen, A.M., De Baerdemaeker, J., Ramon, H., 2004a. Towards development of on-line soil moisture sensor using a fiber-type NIR spectrophotometer. Soil Till. Res. 80 (1–2), 171–183. Mouazen, A.M., Anthonis, J., Saeys, W., Ramon, H., 2004b. An automatic depth control system for on-line measurement of spatial variation in soil compaction. Part 1. Sensor design for measurement of frame distance variation from soil surface. Biosyst. Eng. 89 (2), 139–150.
229
Nielsen, D.R., Wendroth, O., 2003. Spatial and Temporal Statistics, Sampling Field Soils and their Vegitation. GeoEcology Textbook. CATENA VERLAG, GeoScience Publisher, Reiskirchen, Germany. Schmidhalter, U., Zintel, A., Neudecker, E., 2001. Calibration of electromagnetic induction measurements to survey the spatial variability of soils.. In: Grenier, G., Blackmore, S. (Eds.), Proceedings of the Third European Conference on Precision Agriculture, agro Montpellier, Montpellier, France, pp. 479– 484. Sirjacobs, D., Hanquet, B., Lebeau, F., Destain, M.F., 2002. On-line soil mechanical resistance mapping and correlation with soil physical properties for precision agriculture. Soil Till. Res. 64, 231–242. Smith, J., 2001. Soil Electrical Conductivity Mapping and Interpretation. In-field Mapping of Soil and Crop Factors. SCI, London, UK. Sprinkle, L.W., Langston, T.D., Weber, J.A., Sharon, N.M., 1970. A similitude study with static and dynamic parameters in an artificial soil. Trans. Am. Soc. Agric. Eng. 29, 580–586. Upadhyaya, S.K., Williams, T.H., Kemble, L.J., Collins, N.E., 1984. Energy requirement for chiselling in coastal plain soils. Trans. Am. Soc. Agric. Eng. 27 (6), 1643–1649. Verschoore, R., Pieters, J.G., Seps, T., Spriet, Y., Vangeyte, J., 2003. Development of a sensor for continuous soil resistance measurement. In: Stafford, J.V. (Ed.), Proceedings of the Joint European Conference of ECPA-ECPLF, Wageningen Academic Publishers, Berlin, Germany, pp. 689–695.