Organic C accumulation in soil over 30 years in semiarid southwestern Saskatchewan Effect of crop rotations and fertilizers

Transcription

1 Organic C accumulation in soil over 30 years in semiarid southwestern Saskatchewan Effect of crop rotations and fertilizers C.A. Campbell 1, R. P. Zentner 2, B.-C. Liang 2, G. Roloff 3, E. C. Gregorich 1, and B. Blomert 2 1 Agriculture and Agri-Food Canada, Eastern Cereals and Oilseed Research Centre, Research Centre, Central Experimental Farm, Ottawa, Ontario, Canada K1A 0C5; 2 Agriculture and Agri-Food Canada, Semiarid Prairie Agricultural Research Centre Research Centre, Box 1030, Swift Current, Saskatchewan, Canada S9H 3X2; 3 Departamento de Solos, Universidade Federal do Paraná, Brazil. Received 25 March 1999, accepted 7 September Campbell, C. A., Zentner, R. P., Liang, B.-C., Roloff, G., Gregorich, E. C. and Blomert, B Organic C accumulation in soil over 30 years in semiarid southwestern Saskatchewan Effect of crop rotations and fertilizers. Can. J. Soil Sci. 80: Because crop management has a strong influence on soil C, we analyzed results of a 30-yr crop rotation experiment, initiated in 1967 on a medium textured Orthic Brown Chernozem at Swift Current, Saskatchewan, to determine the influence of cropping frequency, fertilizers and crop types on soil organic C (SOC) changes in the 0- to 15-cm depth. Soil organic C in the 0- to 15-cm and 15- to 30-cm depths were measured in 1976, 1981, 1984, 1990, 1993, and 1996, but results are only presented for the 0- to 15-cm depth since changes in the 15- to 30-cm depth were not significant. We developed an empirical equation to estimate SOC dynamics in the rotations. This equation uses two first order kinetic expressions, one to estimate crop residue decomposition and the other to estimate soil humus C mineralization. Crop residues (including roots) were estimated from straw yields, either measured or calculated from grain yields. The parameter values in our equation were obtained from the scientific literature or were based on various assumptions. Carbon lost by wind and water erosion was estimated using the EPIC model. We found that (i) SOC was increased most by annual cropping with application of adequate fertilizer N and P; (ii) that frequent fallowing resulted in lowest SOC except when fall-seeded crops, such as fall rye (Secale cereale L.), that reduce erosion were included in the rotation, and (iii) the fallow effects are exacerbated when low residue yielding flax (Linum usitatissimum L.) was included in the rotation. Some of the imprecision in SOC values we speculated to be related to variations in soil texture at the test site. In the first 10 yr of the experiment, SOC was low and constant for fallow-spring wheat (Triticum aestivum L.) (F-W) and F-W-W rotations because this land was managed in this manner for the previous 50 yr. However, in rotations that received N + P fertilizer and were cropped annually [continuous wheat (Cont W) and wheat-lentil (Lens culinaris L.)], or that included fall-seeded crops (e.g., F-Rye-W), SOC appeared to increase sharply in this period. In the drought period ( ) SOC was generally constant, but large increases occurred in the wet period (1990 to 1996) in response to high residue inputs. The efficiency of conversion of residue C to SOC for the 30-yr experimental period was about 10 12% for F-W, F-W-W and Cont W (+P) systems, and it was about 17 18% for the well fertilized F-Rye-W, Cont W, and W-Lent systems. The average annual SOC gains (Mg ha 1 yr 1 ) between 1967 and 1996 were 0.11 for F-W (N + P), 0.09 for the mean of the three F-W-W rotations (N + P, + N, + P), 0.23 for F-Rye-W (N + P), 0.32 for Cont W (N + P), 0.12 for Cont W (+ P), and 0.28 for W-Lent (N + P). The corresponding mean estimated (by our equation) annual SOC gains for these rotations, were 0.06, 0.10, 0.16, 0.22, 0.14, and 0.22 Mg ha 1 yr 1, respectively. Because soil C measurements are usually so variable, we recommend that calculations such as ours may be employed to assist in the interpretation of measured C trends and to test if they seem reasonable. Key words: Carbon sequestration, carbon conversion efficiency, eroded carbon, crop residue carbon, cropping frequency, wheat, fall rye, flax Campbell, C. A., Zentner, R. P., Liang, B-C., Roloff, G., Gregorich, E. C. et Blomert, B Accumulation du C organique dans le sol en l espace de 30 ans dans la zone semi-aride du sud-ouest de la Saskatchewan. Effet des rotations des cultures et de la fumure minérale. Can. J. Soil Sci. 80: Étant donné la forte influence reconnue aux systèmes culturaux sur le C du sol, nous avons analysé les résultats d une expérience de 30 ans sur les rotations des cultures, démarrée en 1967 dans un chernozèm brun orthique de texture moyenne, à Swift Current en Saskatchewan. L objet était de déterminer l influence de la fréquence des mises en culture, de la fumure minérale et du choix des cultures sur l évolution du C organique du sol (COS) dans les 15 cm supérieurs du profil. Les mesures du COS dans cette tranche de profondeur et dans celle de 15 à 30 cm avaient en fait étaient prises en 1976, 1981, 1984, 1990, 1993 et 1996, mais nous n avons retenu que les valeurs concernant la tranche supérieure (0 15 cm), la couche inférieure ne révélant pas de changements significatifs. Nous avons construit une équation empirique pour calculer la dynamique du COS dans les rotations. L équation comporte deux expressions de cinétique du premier degré, pour calculer, l une la décomposition des restes de cultures et l autre la minéralisation du C de l humus. Les restes de culture, y compris les racines, étaient évalués d après le rendement en paille, mesuré ou calculé à partir du rendement grainier. Les valeurs paramétrales utilisées dans l équation étaient celles rapportées dans les publications scientifiques ou bien elles reposaient sur diverses hypothèses de départ. Les déperditions de C dues à l érosion éolienne et hydrique étaient calculées selon le modèle ÉPIC. Nous avons constaté que i) les augmentations les plus fortes de COS résultaient de la mise en culture chaque année, avec apports suffisants d engrais N et P, ii) les mises en jachère fréquentes produisaient les valeurs COS les plus basses, sauf quand la rotation incluait des cultures d automne comme le seigle (Secale cereale L.), qui freine l érosion et iii) les effets négatifs de la jachère étaient exacerbés lorsque la rotation 179

2 180 CANADIAN JOURNAL OF SOIL SCIENCE comportait une culture laissant peu de résidus comme le lin (Linum usitatissimum L.). Une partie de l imprécision des valeurs COS a été expliquée hypothétiquement par la variabilité de la granulométrie du sol à l emplacement expérimental. Dans les 10 premières années de l expérience, COS était bas et stable dans les assolements jachère-blé de printemps JB et JBB (Triticum aestivum L.) parce que c étaient les systèmes culturaux utilisés dans ce terrain dans les 50 années précédentes. Cependant, dans les rotations recevant une fumure N et P et sous culture continue : blé en continu (B. cont.) ou blé-lentille (Lens culinaris L.) ou comportant des cultures d automne (par exemple seigle-blé), COS semblait accuser une augmentation importante durant cette période. Dans les années de sécheresse , COS demeurait généralement stable, mais dans les années plus arrosées ( ), on constatait de forts accroissements, résultant de l abondance des restes de cultures. L efficience de la conversion du C des résidus culturaux en COS durant les 30 ans de la période expérimentale était d environ 10 à 12 % pour les rotations JB, JBB et B. cont. (avec fertilisation P) et de 17 à 18 % dans les rotations abondamment fertilisées JSB, B. cont. et blé-lentille. Les gains annuels moyens mesurés de COS en Mg ha 1 an 1 entre 1967 et 1996 étaient de 0,11 pour JB (avec fumure NP), 0,09 en moyenne pour les trois assolements JBB (avec NP, N ou P), 0,23 pour SB (avec NP), 0,32 pour B. cont. (avec NP), 0,12 pour B cont. (avec P) et 0,28 pour B. lent. (avec NP). Les valeurs calculées (avec notre équation) correspondantes étaient, respectivement de 0,06, 0,10, 0,16, 0,22, 0,14 et 0,22 Mg ha 1 an 1. Vu la grande variabilité habituelle des mesures de la teneur en C du sol, nous estimons que des calculs du genre de ceux utilisés ici pourraient faciliter l interprétation et éventuellement la vérification des valeurs mesurées. Mots clés: Séquestration de C, efficience de conversion de C, déperditions de C, érosion, restes de cultures, fréquence des jachères, blé, seigle d automne, lin Soil organic C is an important index of soil quality because of its relationship to crop productivity (Gregorich and Carter 1997; Lal et al. 1997). However, recent renewed interest in SOC is related to its relationship to atmospheric CO 2, which is a major greenhouse gas, and which has implications for global warming (Bruce et al. 1998). Crop management practices are known to result in changes in SOC (Campbell 1978; Gregorich and Carter 997; and Janzen et al. 1998). Because SOC changes are generally directly related to the quantity of crop residues returned to the land (Rasmussen et al. 1980; Campbell et al. 1997), agronomic practices that influence yield and affect the proportion of residues returned to the soil are likely to influence SOC. Thus, the addition of fertilizers, reduction in fallowfrequency (Rasmussen et al. 1980; Janzen et al. 1998), and the inclusion of legumes in crop rotation usually increase SOC (Rosswall and Paustian 1984; Campbell et al. 1997). Frequent summerfallowing, because it keeps the soil more moist for longer periods and it increases the frequency of tillage operations, will increase the rate of residue and soil C decomposition and facilitate erosion, thereby reducing SOC (Janzen et al. 1997; McGill et al. 1988; Havlin et al. 1990). The amount and rate of change in SOC varies with its initial level (McGill et al. 1988; Glendining and Powlson 1991; Janzen et al. 1997). Soils with high initial levels of organic matter are more likely to show decreases, and are more difficult to maintain or increase SOC (Campbell et al. 1997; Janzen et al. 1997). Soil texture also plays an important role in stabilizing organic C as soil C (Bauer and Black 1981; Parton et al. 1987; Hassink and Whitmore 1995; Campbell et al. 1996), often tending to increase it as clay content increases. Long-term agronomic studies in which C is being monitored are essential for determining temporal changes in SOC because on an annual basis the changes are often small and often discernible only after several years or decades (Jenkinson 1991; Janzen et al. 1998). Unfortunately, few such studies were initiated to measure C changes per se; thus, measurements of this factor were often not as accurate as is desirable in order to draw reliable quantitative conclusions. Further, because of inherent spatial variability in the field, many samples must be taken at each sampling time to ensure that small differences can be statistically separated (Campbell et al. 1976). However, the latter is constrained by the number of chemical analyses this would entail and the possible physical destruction of the test plots that would result with extensive sampling. Consequently, in many published studies, including our own, too few samples have been taken per plot at each sampling time, and thus, we must wait many years before we can determine with statistical confidence that certain treatment effects have developed. On two previous occasions we have discussed C changes in the long-term crop rotation experiment that was initiated in 1967 at Swift Current in southwestern Saskatchewan (Biederbeck et al. 1984; Campbell and Zentner 1993). In those analyses we only reported on results of the special plot treatments, which, in each case, involved only a single rotation phase of selected rotations. However, results are available for all phases of each rotation in this study. Further, because we have sampled twice more (in 1993 and 1996) since our last discussion, we decided to re-analyze the complete data set from this study, so as to derive a more accurate interpretation of the trends and management effects on SOC. Our objectives were: (i) to determine the influence of cropping frequency, fertilization, and crop types on the amount of C stored in the surface (0- to 15-cm depth) soil over a 30-yr period, and (ii) to quantify these C changes using an empirical simulation equation that estimates rates of residue decomposition and soil C decomposition simultaneously, using first order kinetic expressions. MATERIALS AND METHODS Experimental Design and Crop Management Details of the design and management of this experiment have been reported (Campbell et al. 1983; Biederbeck et al. 1984; Zentner and Campbell 1988; Campbell and Zentner 1993); therefore, only a review pertinent to this study is presented. The experiment was established in 1966 on the South Farm of the Semiarid Prairie Agricultural Research Centre (SPARC) of Agriculture and Agri-Food Canada at Swift

3 CAMPBELL ET AL. ROTATION AND FERTILIZER EFFECTS ON SOIL CARBON OVER 30 YEARS 181 Table 1. Crop rotations and amount of fertilizer applied to treatments during three decades Average N and P applied (kg ha 1 yr 1 ) y Rotation Rotation phase no. Rotation z Fertilizer criteria cropped N P N P N P 11 F-(W) N and P applied Fallow F-W-(W) N and P applied Fallow Stubble F-W-(W) P applied but no N applied Fallow except that in P fertilizer Stubble F-W-W N applied, no P applied Fallow Stubble a F-(Rye)-W x N and P applied Fallow Stubble b CF-WW-WW N and P applied Fallow Stubble F-Flx-(W) N and P applied Fallow Stubble Cont (W) N and P applied Stubble Cont (W) w (fallow if less than 60 cm moist soil exists at seeding time: N Stubble and P applied) 10 Cont (W) w (fallow if grassy weeds become a problem: N and P applied) Stubble (W)-Lent N and P applied Lentil v Wheat Cont (W) u P applied but no N applied except that in P fertilizer Stubble z Selected plots, indicated in parentheses, were sampled for straw weight at harvest. F = fallow; CF = chemical fallow; W = spring wheat; WW = winter wheat; Rye = fall rye; Flx = flax; Lent = grain lentil; Cont = continuous cropping. y Nitrogen and P were applied based on the prescribed treatment. The N was based on soil NO 3 tests, while the P fertilizer was applied to each crop based on the general recommendations of the soil testing laboratory, University of Saskatchewan. Values are sum of nutrient applied to the crop divided by the number of years the crop was grown on the plot. (eg. For Rot 4a, values in is mean of 10 yr; for it is mean of 8 yr and for mean of 4 yr). x In 1985, rotation 4a was changed to chemical fallow-winter wheat-winter wheat (spring wheat whenever winter wheat failed to survive the winter). But in 1993 it was changed to CF-Rye-W again. For convenience we refer to this system as F-Rye-W. w During the first 12 yr, the criteria necessary for summer fallowing in these two rotations were met on several occasions but the action was not implemented. In 1979, these two rotations were changed to the spring wheat-lentil rotation with N and P applied (i.e., rot. 19). For convenience we refer to this system as W-Lent. v In 1990 and 1991 the lentil was erroneously not fertilized with N or P. u In 1980 and 1982, N was inadvertently applied to this system at rates of 70 and 40 kg N ha 1, respectively. Current, Saskatchewan, on slightly sloping land (<3%) that had been cropped previously to a F-W rotation with minimal fertilizer additions since The soil is a Swinton loam (Ayres et al. 1985), an Orthic Brown Chernozem (Canada Soil Survey Committee, Subcommittee on Soil Classification 1978). At the initiation of the experiment (1967) the top 15 cm of soil was estimated to have an organic N concentration of 1.78 g kg 1 (Biederbeck et al. 1984). Assuming a C/N ratio of 10/1, organic C concentration would be 17.8 g kg 1 or 32.6 Mg ha 1. This value was obtained from measurements taken in 1967 at an adjacent site with a similar cultural history. As seen later, we did not use this estimate. The ph (water paste) of the top 15 cm of soil was 6.5. Twelve crop rotations were established on 81, 0.04-ha plots in a randomized complete-block design with three replicates; however, due to changes over the years, only nine rotations are discussed in this study (Table 1). Two of the well-fertilized continuous wheat rotations (rotations 9 and 10) were intended to be flexible rotations that include summerfallow under specified conditions. Although during the first 12 yr the criteria for summerfallowing these flexible rotations were met on several occasions, through technical oversight the appropriate action was not implemented;

4 182 CANADIAN JOURNAL OF SOIL SCIENCE Table 2. Straw yields z, measured, and estimated from straw/grain regressions y Year Treatment F-(W) (N+P) F-(W)-W (N+P) F-W-(W) F-(W)-W (+P) F-W-(W) F-(W)-W (+N) F-W-(W) F-(Rye)-W x (N+P) F-Rye-(W) F-(Flx)-W (N+P) F-Flx-(W) Cont(W) (N+P) (W)-Lent w (N+P) W-(Lent) w Cont (W) (+P) Year Treatment F-(W) (N+P) F-(W)-W (N+P) F-W-(W) F-(W)-W (+P) F-W-(W) F-(W)-W (+N) F-W-(W) F-(Rye)-W x (N+P) F-Rye-(W) F-(Flx)-W (N+P) F-Flx-(W) Cont(W) (N+P) (W)-Lent w (N+P) W-(Lent) w Cont (W) (+P) Mean Mean Year of of Treatment Phase Rotation F-(W) (N+P) F-(W)-W (N+P) F-W-(W) F-(W)-W (+P) F-W-(W) F-(W)-W (+N) F-W-(W) F-(Rye)-W x (N+P) F-Rye-(W) F-(Flx)-W (N+P) F-Flx-(W) Cont(W) (N+P) (W)-Lent w (N+P) W-(Lent) w Cont (W) (+P) z Straw yields are in kg ha 1 for the rotation phase in parentheses. y The straw yields shown in bold type were calculated from regression equations developed in the study relating straw (Y) to grain (G) yields; for spring wheat, Y = G (r 2 = 0.84***, n = 972); for flax, Y = G (r 2 = 0.70***, n = 32); for grain lentil, Y = G (r 2 = 0.46***, n = 50); for fall rye, Y = G (r 2 = 0.30***, n = 50). x The F-Rye-W system was changed to CF-WW-WW, and to CF-Rye-W in The rotation is called F-Rye-W for convenience. w Wheat-Lentil was Cont W (N+P) from 1967 to 1978; it was changed to W-Lent in It is called W-Lent for convenience. consequently, there were three well-fertilized continuous wheat rotations (rotations 8, 9 and 10) (Table 1). In 1979, two of the latter rotations were changed to a W-Lent (N + P) rotation. Throughout this paper we treat rotations 9, 10 and 19 as one rotation called W-Lent. Rotation 4, which was initially well-fertilized conventional fallow-fall rye-spring wheat [F-Rye-W(N + P)] (4a), was changed to chemical fallow-winter wheat-winter wheat (CF-WW-WW) (N + P)

5 CAMPBELL ET AL. ROTATION AND FERTILIZER EFFECTS ON SOIL CARBON OVER 30 YEARS 183 Table 3. Soil and organic C lost by erosion z from selected rotations estimated by EPIC model Period Rotation Erosion losses F-W (N+P) y Soil lost in period (Mg ha 1 ) Cumulative C lost to end of period (Mg ha 1 ) x F-W-W (N+P) y Soil lost in period (Mg ha 1 ) Cumulative C lost to end of period (Mg ha 1 ) x F-Rye-W (N+P) y Soil lost in period (Mg ha 1 ) Cumulative C lost to end of period (Mg ha 1 ) x Cont W (N+P) y Soil lost in period (Mg ha 1 ) Cumulative C lost to end of period (Mg ha 1 ) x z This includes both wind and water erosion. Wind erosion was usually up to 10% higher than water erosion for fallow-containing systems except those with fall rye where water erosion was twice that of wind erosion. Continuous cropping systems had water erosion four times that of wind erosion. y The weighted mean annual soil loss was 2.3, 1.9, 0.6, and 0.5 Mg ha 1 yr 1 for F-W, F-W-W, F-Rye-W and Cont W, respectively. x Carbon loss obtained by multiplying estimated soil loss and measured %C. (4b) in 1984 and then back to chemical fallow-fall ryespring wheat (CF-Rye-W) (N + P) in Throughout this paper we treat rotations 4a and 4b (Table 1) as one rotation called F-Rye-W. All phases of each rotation were present every year and each rotation was cycled on its assigned plots. The field management (i.e., seedbed preparation, herbicide application, seeding, harvesting, and tillage operations) for crops other than lentil were reported previously by Campbell et al. (1983) and Zentner and Campbell (1988). The management of grain lentil was reported by Campbell et al. (1992). Commercial farm equipment was used to perform cultural and tillage operations. Weed control was achieved by a combination of mechanical tillage (mainly cultivator and rod weeder) and herbicides (as required) using recommended methods and rates. The plots were seeded at the recommended rates of 67 kg ha 1 for spring wheat, 31 kg ha 1 for flax, 63 kg ha 1 for fall rye, 67 kg ha 1 for winter wheat, and 90 kg ha 1 for lentil. Crops generally were seeded in early May, except for rye and winter wheat, which generally were seeded during the first week of September. Recommended cultivars were used each year, but cultivars were changed with time as new ones became available. Fertilizer N, as NH 4 NO 3, was applied in accordance with rotation specifications (Table 1) at rates recommended by the Saskatchewan Advisory Council on soils, based on soil NO 3 -N (0 to 60 cm) from individual plots measured in fall. Phosphorus fertilizer (ammonium phosphate) was applied with the seed in accordance with the general recommendations for the area and crop (Saskatchewan Agriculture 1985). Wheat plots received an annual rate of 9 10 kg P ha 1 for rotations that were designated to receive P (Table 1). In several years during the early period of the experiment, P fertilizer was not required for flax and fall rye production, nor was it recommended for flax (University of Saskatchewan 1975). Crops were harvested at the full-ripe stage. Lentil was desiccated in most years with herbicide to hasten ripening. On all cropped plots, yield determinations were made by cutting a swath 5 m wide and 40 m long through the middle of the plot and harvesting the grain with a conventional combine. Small 2.32-m 2 plots also were hand harvested to determine straw/grain ratios. The straw/grain ratios were used, together with combine harvested grain yields, to estimate straw yields associated with the combine grain yields. For some treatments in which straw measurements were not made, we estimated straw yields using regression equations that relate straw yields to grain yields, which we developed based on data from this experiment (footnotes, Table 2). The straw was distributed on the plots by a paddle-type spreader attachment on the combine. Straw was not incorporated until the following spring or early summer. Monitoring Soil Organic Carbon Two soil samples were taken at random within the central part of each plot. The 0- to 15- and 15- to 30-cm depths of each plot were sampled with a 5-cm-diameter soil probe using a Giddings soil sampling truck. Samples were taken in spring 1976, in fall 1981, and 1984 (some in fall and some in spring 1985), and in fall 1990, 1993, and The two samples per depth in each plot were composited; the soil air dried and sieved (<2 mm). Small pieces of crop residues passing through the 2-mm sieve were regarded as SOM. Soil samples taken prior to 1993 were ground to <1 mm with a Wiley-Thomas mill (Thomas Scientific, Swedesboro, NJ).

6 184 CANADIAN JOURNAL OF SOIL SCIENCE The methods of measuring SOC and inorganic C for the samples taken prior to 1993 are described by Campbell and Zentner (1993). The samples taken in 1993 and 1996 were analyzed differently. Representative subsamples of the <2 mm material were ground with a roller mill (<153 µm) and a 20 mg subsample analyzed for total N, total C and organic C by an automated combustion technique (Carlo Erba TM, Milan, Italy) at 1800 C (Campbell et al. 1996). A reference soil, obtained from the Canadian certified reference materials, supplied by Canada Centre for Mineral and Energy Technology, was included with our samples whenever we conducted C and N analyses. The SOC concentrations were converted to a weight per volume basis using bulk densities of 1.22 and 1.30 Mg m 3 for the 0- to 15-cm and 15- to 30-cm depths, respectively, measured at the start of the experiment (Campbell et al. 1983). We assumed that soil bulk densities were more-orless constant at SOC sampling, because (i) tillage practices were similar to the ones utilized long before the start of the experiment, and (ii) sampling took place at a time of the year (spring before tillage or fall after harvest) in which the soil had lost any loosening effect of tillage. The C values presented are the means and standard error of mean (S x ) of all phases and replicates of each rotation. Estimating SOC Changes from Crop Residue Inputs and Soil C Decomposition Various process-based models (e.g., CENTURY AND EPIC) allow quantitative estimation of SOC changes. However, such models require complex calculations and often, several of the model inputs are not commonly measured. Further, as shown by Campbell et al. (1999b) they do not always provide an accurate estimate of SOC changes. We attempted to model changes in SOC but with the minimum number of variables required to capture the most important features of SOC dynamics, such as the balance between C inputs from crop residues and outputs from soil C decomposition. We used an empirical method of estimating soil C dynamics by treating recent residue additions and pre-existing C separately, similar to the method proposed 50 years ago by Woodruff (1949). This equation is: Fig. 1. Soil organic C in Swift Current rotations (mean of values in 1976, 1981, 1984, 1990, 1993 and 1996 for all rotation phases and reps). ( ) kt kt SOCt = C q e q2e 2 + t r t n r t n + An p e 1 ( + p e ( ) 2( ) 1 2 ) n= 0 where SOC t is the total amount of soil organic C per unit mass of soil remaining in soil after t years (time measured to just before residue addition in the current year, example residue added in yr 10 is not included at t = 10 yr), C 0 is the amount of C in the soil on a mass basis, initially t = 0, q is a proportion of soil C, k is the annual rate of soil carbon decomposition (yr 1 ), A n is the C addition as plant residue (Mg ha 1 ) in year n, p is the proportion of residue C (note p 1 + p 2 = 1), and r is the annual rate of residue decomposition (yr 1 ). The subscripts 1 and 2 refer to different degrees of susceptibility to decomposition with 1 representing the more active, and 2 representing the slower decomposing pool of plant residues and soil humus. One advantage of using Eq. 1 is that values for A (related to residue inputs) are usually available, or can be readily estimated from grain yields, harvest index, and straw/root ratios. These measured yield data integrate climate conditions thus helping to make this approach more robust. We estimated the potential C input from crop residues by assuming the root/straw ratio was 0.59 (Campbell et al. 1977) and the C concentration of tissues was 45% (Millar et al. 1936). We used the coefficients and decomposition rate constants developed by Voroney et al. (1989) for Sceptre clay in semiarid southwestern Saskatchewan, (located 40 km north of Swift Current) to estimate the accumulation of SOC over time after residue input. These coefficients and rate constants were based on a study in which 14 C-labeled wheat straw for a F-W-W-W rotation was incorporated into soil and the C monitored annually for 10 yr: (1) y = 0.72e 1.4t e 0.081t (2) where, y is the proportion of residue C remaining in the soil and t is years since residue application. Although Eq. 2 was developed for a clay textured soil, while our soil is a loam to

7 CAMPBELL ET AL. ROTATION AND FERTILIZER EFFECTS ON SOIL CARBON OVER 30 YEARS 185 Table 4. Example of how C-balance was constructed for F-W (N + P) rotation Mg C ha 1 Item Factor Measured C Eroded to date Sum (1) + (2) Increase due to residue input z Decrease due to soil C decomposition y Change in C to date[item (4) (5)] Estimated SOC on date[item (6) + initial soil C x ] z Calculated from Eq. 2 (Voroney et al. 1989). y Assumes k 1 = 0.02 yr -1, k 2 = yr -1, q 1 = 0.20 and q 2 = x Assumes starting C in 0- to 15-cm depths in 1967 = 30.5 Mg ha 1. clay loam in which C sequestration will be lower than in a clay (Campbell et al. 1996), we substituted its numerical values into Eq. 1 (i.e., P 1 = 0.72, P 2 = 0.28, r 1 = 1.4 yr 1, and r 2 = yr 1 ). To obtain the C present in soil at any time after the start of the study we required an estimate of the initial soil C level (C 0 ) and the rate of soil humus decomposition, i.e., k 1 and k 2. Since we did not measure C 0, we estimated it to be the same as the value for F-W(N + P) taken in 1976 (first measurements taken) because the land had been F-W since This value was 30.5 Mg ha 1 in the 0- to 15-cm depth. Further, because all phases of each rotation were present each year, the calculated C change differed for each phase. Thus, if we averaged over replicates, there were two values for F-W, three values for F-W-W, and one value for Cont W. Thus, to compare calculated to measured results, we averaged the changes over rotation phases. The parameters for the active and slow decomposing pools of soil humus carbon (q 1 and q 2, respectively in Eq. 1) can be estimated from data in the literature, as can values of k 2 (Campbell et al. 1999b). Thus, we estimated the rate constant (k 2 ) for the slow decomposing fraction of soil humus (q 2 ) from the mean residence time (determined by C-dating) of the humic fractions (about yr 1 for Chernozemic soils in Saskatchewan). We assumed that the active soil humus fraction (q 1 ) was about 0.20, equal to the average sum of the light fraction and microbial biomass C during a growing season in this Swift Current soil (Campbell et al. 1999a). Thus q 2 would be 0.80 in this soil. We then estimated k 1 by solving Eq. 1 iteratively so as to match trends in measured SOC as discussed later. Note, however, that estimates of k are empirical because the value of k in Eq. 1 must embody differences not only in inherent susceptibility to decomposition of soil C fractions, but also the effects of soil moisture, temperature, etc., on rate of decomposition. Because fallow conditions favor decomposition, we reasoned that k 1 for F-W > k 1 for F-W-W > k 1 for Cont W. We expect k 2 and r 2 to be relatively constant within a temperate climate, but k 1 and r 1 might vary considerably from site to site, depending on weather, soil texture, tillage, etc. Estimating C Lost by Erosion Because C displaced by erosion is not lost by decomposition, it is necessary to measure or estimate this source of C and add it to the measured C if we are to make an accurate C balance (cf. Table 4). We used the EPIC 5300 model (Williams 1995) to estimate the amount of soil and C lost annually by water and wind erosion from F-W (N + P), F- W-W (N + P), F-Rye-W (N + P), and Cont W (N + P) rotations (Table 3). For water erosion we selected the USLE equation with slope steepness 0.03 mm 3, slope length 50 m, erosivity calculated from daily rainfall, erodibility calculated from Wischmeier s nomograph, and the crop and management factor calculated daily from soil cover by canopy and residue. We assumed conservation practices were absent. The wind erosion was calculated by the Wind Erosion Continuous Simulation (WECS) equation, using daily average wind speed. The power parameter of the modified experimental distribution of wind speed was taken as 0.6. Field length was 1.0 km, field width 0.5 km, field orientation 90 clockwise from north, and vegetative and roughness factor were calculated daily. We summed the annual total soil losses by wind and water for the sampling periods (e.g., 1967 to 1976, 1977 to 1981, etc.), and multiplied these amounts by the percent C measured at the end of each period for each rotation in question to obtain C lost in that period. These values were accumulated over successive periods to estimate C lost up to each sampling date (Table 3). Weather Conditions Daily growing season precipitation, maximum and minimum daily air temperatures, wind speeds, and Class A pan evaporation during the growing season were measured at a meterological station located 1 km from the test site. Statistical Analysis We calculated means and standard error of the means (S x ) for SOC measurements at each sampling time for each treatment. For straw yields we calculated means. RESULTS AND DISCUSSION Effect of Cropping Systems on SOC As found in previous analyses (Biederbeck et al. 1984; Campbell and Zentner 1993), the treatments had no effect on organic C in the 15- to 30-cm depth (data not shown) and therefore we discuss only C in the 0- to 15-cm depth (Fig. 1). Average SOC over the six sampling dates was highest in

8 186 CANADIAN JOURNAL OF SOIL SCIENCE Fig. 2. Measured SOC in the 0- to 15-cm depth over a 30-yr period showing effects of (a) crops in rotation; (b) fertilizer N + P in a F-W-W rotation; and (c) addition of N via fertilizer or pulse crop in a continuous cropping system. (Initial C value in 1967 estimated from F-W (N + P) treatment value for 1976.) (Data are means of all rotation phases and reps.) Fig. 3. Pattern of change in SOC for selected aggrading and degrading systems compared to the pattern of change in residue C inputs for one rotation in each group. (Only two of the latter are shown for sake of simplicity.) the W-Lent (N + P) system, intermediate for F-Rye-W (N + P) and Cont W (N + P), and lowest in Cont W (+P) and the fallow-containing systems with only spring-seeded crops (Fig. 1). The mean difference in measured values, between crop rotations with highest and lowest SOC, was about 5.5 Mg ha 1, accumulated over 30 yr. As shown later, measured SOC data can be quite variable. Nonetheless, the positive effect of the fall-seeded crop in maintaining SOC at a higher level than spring-seeded crops was apparent, while inclusion of flax in the rotation tended to have a negative effect on SOC (Fig. 2a). Both F-W-W (N + P) and F-Rye-W (N + P) produced similar amounts of straw (Table 2), therefore the higher SOC in the Rye-system was partly because the fall-seeded cereal reduces C loss by erosion (Table 3). Second, the length of the fallow period, when conditions are ideal for decomposition of soil humus, is only 12 mo for F-Rye-W compared to 21 months for F- W-W. In contrast to F-Rye-W, flax produces much less straw than wheat (Table 2), its straw does not decompose as readily as cereal straw, and its straw does not anchor well in the soil after tillage and is therefore easily blown from the field. The influence of fertilizer N and P on SOC was small in the 3-yr F-W-W system (Fig 2b), as observed in other studies (Campbell et al. 1998). Because of the high variability in the Cont W (N + P) data, the influence of not fertilizing with N on SOC in Cont W was not readily apparent (Fig. 2c), though the overall means (Fig. 1) suggest an advantage

9 CAMPBELL ET AL. ROTATION AND FERTILIZER EFFECTS ON SOIL CARBON OVER 30 YEARS 187 (a) (c) (b) Fig. 4. Comparison of measured SOC + eroded C and estimates of SOC made using Eq. 1, for various rotations. S x represents the standard error of the mean for the measured values. [(a) = F-W (N + P) and mean of F-W-W (N, P, and N + P); (b) = Cont W (P) and Cont W (N + P); (c) = F-Rye-W (N + P) and W-Lent (N + P)]. [The (?) in Figs. 4a, b, c, identify apparent inconsistencies in the measured data as discussed with regard to Fig. 2 in the text.] to fertilizing with N. The latter results are consistent with those reported elsewhere on the Canadian prairies (Janzen et al. 1997). The W-Lent (N + P) system, with its narrower C/N ratio in crop residues (Campbell et al. 1992), which would enhance its rate of decomposition (Parr and Papendick 1978; Reinertsen et al. 1984; Janzen and Kucey 1988), tended to result in even higher SOC levels than Cont W (N + P) (Figs. 1 and 2c). This suggests that the presence of legumes in rotation with wheat may result in more efficient conversion of residue C to SOC than when monoculture wheat is fertilized. Drinkwater et al. (1998) presents evidence that appears to support this theory. Based on results they obtained with legume-based cropping systems at Rodale Institute in Pennsylvania, they suggest that the use of low C-to-N organic residues to maintain soil fertility, combined with greater temporal diversity in crop sequences, will increase soil C and N. Because we observed a close similarity between soil C trends over the years for F-W and F-W-W systems, we averaged soil C measurements over these four treatments (Fig. 3). Similarly, we averaged values for Cont W (P), W-Lent (N + P) and F-Rye-W (N + P) (Fig. 3). Generally the measured soil C fit two patterns. In the fallow-spring seeded

10 188 CANADIAN JOURNAL OF SOIL SCIENCE Fig. 5. Comparison of measured plus estimated eroded C versus estimates of C gains calculated from Voroney et al. (1989) residue C decomposition model (Eq. 2) with no allowance made for C lost via soil decomposition, for the F-Rye-W (N + P) and W-Lent (N + P) rotations. crop systems, SOC was generally constant until 1990, then increased sharply in response to a 7-yr period of above-average yields (residue input) (Fig. 3, bottom). For the aggrading systems there was an early gradual increase, due to change from over 60 yr of F-W (poor fertility) to continuous cropping (with good fertility); then values leveled off in the droughty 1980s before increasing sharply due to high residue C inputs in the 1990s (Table 2). These results strongly suggest that crop residue inputs play a very significant role in influencing short-term soil C dynamics, perhaps an even more significant role than soil decomposition mechanisms. The W-Lent and F-Rye-W results suggest that soil humus decomposition is slow in the cold semiarid prairie conditions of western Canada. It could be argued that the apparent increase in SOC for the aggrading systems, between 1967 and 1976, is due to our estimate of initial C level. Though possible, especially in view of the noise in the data (Fig. 2), we do not think this is logical. It seems unlikely that all the aggrading systems would have a higher SOC level than the degrading systems in 1976 due only to sampling inconsistencies. Further, if the starting value of these aggrading systems were as high as their 1976 values, our model suggests that all later SOC values would be much higher than we measured. For these reasons, and by improving our precision by pooling more treatments, we believe Fig. 3 provides the most accurate picture of the SOC changes over time in this experiment. Data Inconsistences The measured SOC values often changed in direction and amounts over time, which could not be readily explained (Fig. 2). For example, in Fig. 2(b) it seems unlikely that the system receiving both N and P would have lower values than the two less-well fertilized systems in 1981 and 1984 when input crop residues (Table 2) were higher for the system receiving N + P. Further, it seems unlikely that SOC in F-W-W (N + P) would rise sharply in the period 1984 to 1990 when residues were low due to 3 drought years (1984, 1985 and 1988). Similarly, for Cont W (N + P) the value for 1981 appears to be inordinately high, while values for 1984, 1990 and 1993 appear to be low (Fig. 2c). For W-Lent (N + P), which was actually Cont W (N + P) from 1967 to 1978, the SOC value for 1976 may be high and is much higher than the value for Cont W (N + P) at that time. The noisiness of these measurements is in part due to the few replicates (three) and limited number of subsamples taken (two per plot) when making the measurements. For example, Campbell et al. (1976) has shown that for a similar study in which total N was monitored over 10 yr, the minimum difference required for significance (P < 0.05) between two population means when the number of replicates is 3 is about 3.5 times what it would be if 10 replicates were used. Nonetheless, we suspect that one of the main contributors to the noise in the data was related to textural differences within the experimental plots. O Halloran (1986) reports that soil texture varied from loam to silty clay loam at this site. He assessed the spatial variability in three rotations and stated texture varied significantly within and between treatments and replicates, with sand being the most variable size fraction. Using covariance analysis O Halloran showed that up to 90% of the variability in soil P could be attributed to changes in silt plus clay content. Various workers have shown that there can be a direct relationship between SOC and clay content (Parton et al. 1987; Campbell et al. 1996) or clay plus silt content (Hassink and Whitmore 1995). We therefore suggest that when monitoring changes in SOC we should routinely measure soil texture as well and try to use covariance analysis (as done by O Halloran) to differentiate the textural from treatment effects. Another possible contributor to imprecisions in SOC data over time was the fact that we did not monitor bulk density at each sampling. However, our recent measurements (data not shown) show no marked changes in bulk density in these plots compared to 1967, nor has organic matter had any influence on bulk density in this soil (Campbell et al. 1999a). Estimating SOC Changes Using Equation 1 We used iterative calculations with Eq. 1 and the two curves (Fig. 3) to estimate approximate values for k 1. This process indicated the k 1 value for F-W was about 0.02 yr 1 and for the continuous cropping systems was about yr 1. The value for F-W-W will be between that for the latter two extremes, but closer to F-W; for simplicity we chose a value of k 1 = 0.01 for F-W-W. The estimated C lost by erosion over the 30-yr period (Table 3) was added to the measured SOC values and then these sums were compared to the estimated soil C values derived with Eq. 1 (Table 4 and Fig. 4).

11 CAMPBELL ET AL. ROTATION AND FERTILIZER EFFECTS ON SOIL CARBON OVER 30 YEARS 189 Equation 1 estimated the trends in SOC consistently well though it tended to underestimate the measured + erosion values for W-Lent (N + P). Trends estimated by Eq. 1 showed a flat SOC response in the 1980s when droughts in 1984, 1985 and 1988 resulted in very low yields. This was followed by a consistent trend to higher SOC values in the wet 1990s. Note that if we estimate C gains using Eq. 2, and assume that no soil C mineralization occurred, then the estimates fit the measured data for F-Rye-W (N + P) and W-Lent (N + P) very closely (Fig. 5). This suggests that soil humus mineralization is very low in these two systems, or possibly that some of our assumptions for lentil and for fall rye using the spring wheat based model (Eq. 2) are not appropriate. We calculated the change in SOC between the start of the experiment and the last date sampled, for each rotation, comparing values derived by measurements and those estimated by Eq. 1 (Table 5). We also calculated a mean annual rate of increase in SOC over the life of each treatment. The measured and estimated SOC changes and rates of change were generally similar and confirmed our earlier expectations regarding the influence of management on SOC changes. Based on our estimates at Swift Current, SOC changes were between 0.06 Mg ha 1 yr 1 in F-W to 0.22 Mg ha 1 yr 1 in well-fertilized continuously cropped systems. The changes are all positive, suggesting that prairie farmers using stubble mulch cropping techniques and proper fertilization are currently maintaining SOC, and in some cases, where fallow frequency has been reduced, SOC may be increasing. It is likely that C gains will be even greater if direct seeding management is adopted (Janzen et al. 1998). This bodes well for society and producers with respect to increasing C storage in soils. The constants we developed for assessing SOC dynamics in the Swift Current rotations may be applicable to Chernozemic soils that had been degraded for decades. However, when we used Eq. 1, to simulate SOC dynamics for Cont W and F-W (unfertilized) over a 40-yr study period in a Dark Brown Chernozem from Lethbridge, Alberta, our estimates far exceeded the measured values (Campbell et al. 1999b). This was partly because, at initiation of this experiment in 1951, this soil was not as degraded as the soil at Swift Current. In addition, long use of mixed rotations that included legumes, plus regular additions of manure, from breaking in 1910 to 1951, likely maintained this soil in a very fertile state. Evidence of this was the much higher SOC (34.5 Mg C ha 1 ) level at Lethbridge at start of the study compared to Swift Current, even though both soils are in the same ecozone. As suggested by Janzen et al. (1998), our ability to increase C in soils depends not only on the amount of residue C inputs, but also on the C content of the soil at the time the residue was added. It is difficult to increase the C content of a soil that is already high in C, but much easier to increase the C content of a degraded soil. If we are to estimate soil C changes reliably over time and space on a regional or larger basis, we will need to use process-level models that are more complex than Eq. 1. Nonetheless, Eq. 1 performed well when we provided adequate data to characterize a site. However, it suffers from Table 5.Total residue C input ( ) z, measured and estimated y SOC increase from start of experiment to final measurement, and efficiency x of conversion of residue C to SOC Total Measured Measured Equation Mean annual SOC Mean annual SOC Mean annual Efficiency of residue C SOC + eroded C estimated C change based on change based on change based on conversion of Rotation input increase increase increase measured data measured + eroded C equation estimates input C to SOC w Mg ha 1 Mg ha 1 yr 1 (%) Swift Current ( ) F-W (N+P) F-W-W (avg N+P, +P, +N) F-Rye-W (N+P) Cont W (N+P) Cont W (+P) W-Lent (N+P) z Values are means over rotation phase. y Carbon change estimated by Eq. 1 and erosion estimated by EPIC model. x Residue C input divided by measured + eroded C. w For F-Flx-W, total residue C input was Mg ha 1, measured + eroded C increase was 4.44 mg ha 1 and efficiency of conversion, 13.99%.