Rice yield prediction from yield components and ...

European Journal of Agronomy 17 (2002) 41 – 61 www.elsevier.com/locate/eja

Rice yield prediction from yield components and limiting factors D. Casanova a,*, J. Goudriaan b, M.M. Catala Forner c, J.C.M. Withagen d a

Land and Water De6elopment Di6ision, Food and Agriculture Organization of the UN, Apartado Postal 1524, Managua, Nicaragua b TPE Sub-Department, Wageningen Agricultural Uni6ersity, PO Box 430, 6700 AK Wageningen, The Netherlands c IRTA Experimental Station (EEE), ctra Balada s/n, 43870 Amposta, Spain d DLO-Research Institute for Agrobiology and Soil Fertility, PO Box 14, 6700 AA Wageningen, The Netherlands Received 29 September 2000; accepted 7 November 2001

Abstract This article aims to quantify growth at field level in relation to crop status and soil properties in irrigated direct-seeded rice. Forty fields were selected in the Ebro Delta (Spain). Rice growth was monitored and soil properties measured. Yield was related to soil properties by a deductive process identifying the yield determining factors. First (i) yield components, then (ii) variables related to crop status, and finally (iii) soil properties were identified. To interrelate the different groups of variables, correlation analysis and an operational methodology of von Liebig’s Law of the Minimum of the limiting factors were used. To quantify yield at field level, besides panicle and spikelet number, the fraction of necrotic grains, intensity of weed infestation and spatial heterogeneity within fields were necessary. Yield prediction accuracy from these five variables was very high, r 2adj = 0.94. Four main factors limited rice growth: (i) potassium and (ii) zinc shortage, both with strong antagonism with sodium, (iii) plant establishment and (iv) length of the growing season. Yield prediction accuracy from these variables was moderately high, r 2adj = 0.76. The longer the causal distance between yield and its determining factors, the lower the prediction accuracy. K and Zn deficiencies in the plant were mainly induced by soil salinity. The length of the growing period was primarily determined by temperature, but also by soil properties. Long growing cycle, homogeneity of plants per unit area, and adequate K and Zn levels in the plant were favored by high clay concentrations in the topsoil. © 2002 Elsevier Science B.V. All rights reserved. Keywords: Ebro delta; Quantitative analysis; Soil salinity; Plant density; Crop status; Irrigated rice

1. Introduction

* Corresponding author. Tel.: + 505-278-8477; fax: + 505278-0804. E-mail address: [email protected] (D. Casanova).

Worldwide, rice is cultivated in many different ways and degrees of intensification. The main differences being between upland rice and lowland rice production systems. Lowland rice may be

1161-0301/02/$ - see front matter © 2002 Elsevier Science B.V. All rights reserved. PII: S 1 1 6 1 - 0 3 0 1 ( 0 1 ) 0 0 1 3 7 - X

42

D. Casano6a et al. / Europ. J. Agronomy 17 (2002) 41–61

irrigated or rainfed. Under fully irrigated conditions, rice may be transplanted or direct seeded. Direct seeding with irrigation is the most common system in Europe, Australia and USA (Hill et al., 1991). Rice cultivation in the Ebro Delta (northeast Spain) occupies around 21 000 ha, with an average yield of 6500 kg ha − 1 (Catala´ Fomer and Fosch, 1997). This article focuses on understanding growth in relation to crop status and soil properties in irrigated, direct-seeded rice on a large area that was until 30 years ago an unhealthy, swamp, now transformed into a productive rice-growing area Rice growth was monitored and soil properties measured in 40 fields spread over the study area (see Fig. 1). Agricultural research is usually performed in single fields with sub-plots to assess the effect of single variables or combinations (Gomez and Gomez, 1984). This research tackled the problem differently, by observing directly what is happening in the farmers’

fields. This approach is typical of sociologists, biologists and epidemiologists (Webster and Oliver, 1990), and seeks to interpret the response to environmental variation. To elucidate the basic soil and plant processes determining rice yield in the area, the causal scheme is followed: soil properties and farm management influence the crop status, which, together with weather conditions, affects growth, finally determining yield. Ideally, a processed-based model is needed to relate soil properties to rice growth, but, as one was not available, a deductive process is followed that goes from yield to soil properties. The first step was to identify the main yield components and any influencing factors. In the second step, plant growth and crop status were related to yield components and influencing factors. The third step was to correlate the parameters of crop status with individual soil factors. Using this step-by-step approach, we were

Fig. 1. Sites of the fields ( +) within the Ebro Delta.


able to identify what and how soil properties affected rice growth.

2. Materials and methods Forty fields on 30 farms along the Ebro Delta, Spain, were selected for monitoring. Where soil conditions varied within farms, additional fields were selected to cover the soil variation (Fig. 1). As direct-seeded flooded rice is fully irrigated, water limitation is not considered. The fields were surveyed, and monitored during the 1995 rice growing season. Field sizes ranged from 0.5 to 3 ha (average 1.5 ha).

2.1. Soil properties For each parcel, based on augerings and measurements at three sites, the soil unit was classified at family level of the USDA classification by joining together several concepts: (i) The soil types were grouped based on profile development, texture and drainage. (a) Profile development was considered to occur if a histic horizon or calcium carbonate accumulation was present within 100 cm depth from the mineral soil surface. (b) Coarse texture indicates a sandy particle-size class (SSS, 1996). This parameter refers to a texture of loamy fine sand or coarser for all the materials from the base of the epipedon to 100 cm depth.Profile development and coarse texture were coded with 1 or 0 (as nominal-type of data) if present or not. (c) Drainage status was as an ordinal attribute with ordered rank. A system of quantifying soil drainage status was developed based on soil morphology and field measurements (Casanova et al., 1999). Values ranged from 1 to 9; the numbers increase with poorer soil drainage conditions. (ii) Particle-size class. The clay percentages at 40–60 cm arid 80– 100 cm depth were measured.

43

(iii) Topsoil texture, as explained below in the granulometry of the epipedon. (iv) Soil salinity, expressed as an average for 0 –120 cm depot, was measured before flooding (Richards, 1954). On each parcel, before flooding, a composite sample of topsoil from three sites within the parcel was taken for analysis. Analytical methods were those recommended by the Spanish Ministry of Agriculture, Fisheries and Food (MAPA, 1986), and included: texture (clay, silt and sand percentage); pH (soil/water extract 1:2.5); EC1:5 (topsoil salinity was also measured at the panicle initiation stage of the rice crop because of the importance of soil salinity in the study area); cation exchange capacity and exchangeable K and Mg were measured by extraction with ammonium acetate 1 N at pH 7; organic matter content (Walkey–Black method); available P (Olsen– Watanabe method); and total N (Kjeldahl method).

2.2. Farm management Farm management practices were recorded over time, including land preparation; fertilizer (N– P– K), both basal and cover; sowing (date and seed rate); cultivar; weed management (chemical application or mechanical); phytoprotection management; harvesting date; and post-harvesting practices (straw burning or incorporation). Shortgrain cultivars commonly grown in Spain, such as Bahia, Senia, Tebre and Niva, were grown in all the area covered by the research.

2.3. Weather conditions Daily totals of temperature, global short-wave radiation and wind speed were recorded at an agricultural experiment station situated an the study area, 15 km from the river mouth.

2.4. Crop status Fields were visited approximately weekly during the rice growing season. Crop status was determined on the basis of crop establishment, crop duration and nutrient status, as explained below.

44


(i) Crop establishment. In each field, sampling was done at three positions (1/4, 1/2 and 3/4 of the length) along an approximate diagonal transect. Three replicates per location were selected by throwing randomly an 80 cm diameter hoop (0.5 m2). Plant characteristics measured were grouped based on the development stage: (a) At tillering, the number of plants per square meter were measured. The fraction of intercepted photosynthetically active radiation (PAR), fPAR, was calculated (Casanova et al., 1998) from field reflectance measurements taken with a hand-held multispectral radiometer (CROPSCAN CROPSCAN Inc., 1993). An optimum fPAR for potential production was derived from ORYZA1, a physiological model (Kropff et al., 1994), based on development stage. The difference between the optimum and the measured fPAR was introduced in the matrix data This variable gives an estimate of the deviation from the most adequate fPAR that the field should have at that moment. It can be thought of as a measure of seedling vigour for the rice crop. (b) At flowering, as reflectance measurements saturate and are inadequate for dewing high values of fPAR (Casanova et al., 1998), direct measurements of PAR (400–700 nm) above and below the canopy were made using a Decagon Sunfleck Ceptometer (Model SF-80; Decagon Devices Inc., 1989). These values at flowering are, to a certain extent, a measure of the source capacity of the rice plant. (c) At ripening, the plant height was measured. (ii) Crop duration The dates of main phenological events were recorded, namely sowing, panicle initiation, flowering and maturity. (iii) Nutrient status was determined from crop samples taken at tillering and at maturity. Based on Yoshida, (1981), the nutrients in the different plant components were ana-

lyzed: at tillering in the leaves (N, P, K); at tillering in the stems (Ca, Na, Zn, K, Mg); and at maturity in the straw and grain (N, P, K, Ca, Na, Zn, Mg).

2.5. Yield components and influencing factors The sampling procedure for yield components and influencing factors was as for crop establishment, namely three 0.5 m2 samples at each of three equidistant points along an approximate diagonal transect. Plant characteristics measured were: (i) At flowering, the coefficient of variance of fPAR was used as an estimate of rice growth heterogeneity within each field. (ii) At ripening, rice panicles, Echinochloa spp., Cyperaceae and wild rice species per square metre were measured. In addition, fungal infestation was assessed in each parcel by selecting 15 plants at random and estimating: (a) the number of green leaves per stem at hard dough stage; (b) the fraction of leaf area affected by Helminthosporium, Blast (Pyricularia oryrae) or other disorder: (c) the fraction of stem affected by black (probably Fusarium spp.) or brown (probably Sclerotium spp.) disorders; and (d) the proportion of necrotic grain. (iii) At maturity, variables measured were: number of grains per panicle; percentage of filled spikelets; weight of 1000 dry grains (Yoshida, 1981); and grain moisture. The percentage of filled spikelets was determined from the number of unmilled grains that sank in a saline solution (SG= 1.06) in relation to the number of spikelets (Matsushima, 1957). The experimental fields were harvested independently by the farmer. To verify final yield data given by the farmers, a final harvest of 5 m2 was made at three locations each in 14 of the fields. Reported yields refer to unmilled rice containing 14% water. Maturity is taken as the time when the maximum grain weight is attained (Yoshida, 1981), independent of harvest date. To estimate the attainment of maturity, a model based on tempera-


ture summation with a base temperature of 10 °C, an optimum temperature of 28 °C and a maximum of 40 °C was used (Casanova et al., 2000). The input variables were moisture concentration at harvest and daily temperature. Maturity stage was taken to be when the final grain moisture concentration reached 25.5%.

2.6. Data analysis One method for examining grain yield performance is to break the yield into its components (Yoshida, 1981), such that: Y = PANO × SPP × FSP ×Wf ×10 − 5 −1

(1)

where, Y, grain yield (t ha ); PANO, panicle number m − 2; SPP, spikelets per panicle; FSP, fraction of filled spikelets; Wf, 1000-grain weight (g). In Eq. (1), the spikelet number includes filled, partially filled and sterile spikelets. Filled spikelets have mature grains. The filled spikelet fraction is the ratio of the number of grains to the total number of spikelets. The 1000-grain weight is the average weight, in grams, of 1000 grains. The variables in this study were neither controlled nor pre-selected, as they reflect the experiments’ being under field conditions. Thus, the levels of the independent variables can be considered random because they were not chosen at evenly spaced intervals. (i) Simple linear correlation analysis (Gomez and Gomez, 1984) was performed to show the degree of linear association between soil properties and crop status in Step 3 and the response to yield components and factors influencing crop status in Step (2a). In addition, simple linear regressions were fitted between the predicted and observed variables. (ii) The Law of the Minimum of the limiting factors (boundary line method), first proposed by Liebig (1855), was used for predicting yield in Step 1. Webb (1972) and Waggoner and Norvell (1979) were the first to develop mathematical techniques for this nineteenth-century law. Recently, boundary line or maximum line determination techniques have been developed by Schnug et al.

45

(1995). The line describing the highest yields observed over the range of values measured is known as the boundary line, since it lies on the upper edge of the body of data. Fig. 2 shows a scatter plot of an independent variable Xi with the dependent variable Y, where a linear regression line and a line over the maximum X-levels was drawn. The steps followed for defining the maximum line are: (a) Categorize the Xi variable in 10 groups. (b) Plot the histogram of the Y values for each Xi group variable, for groups with more than two members, and fit a normal distribution. (c) Select only the groups where a normal distribution test is not rejected. (d) For each of these groups, select the Xi average and the Y at 95% confidence (mean plus two times the standard deviation(S.D.)). (e) Fit a linear regression through these selected Xi and Y values. Non-linear regressions could also be fitted, especially for variables that are directly linked to crop performance, because at low supply, crop uptake is linearly related to supply, whereas, with increasing supply, yield increase tends to saturate (McBratney and Pringle, 1997). Variables, which act as yield-reducing factors, however, start at this plateau level and reduce yield as they increase. Thus, in this paper simple linear regressions were used to understand and predict yield performance. This procedure is followed for every independent variable (i= 1, 2, …, N). These maximum lines { f(x1) f(x2), …, f(xN )) describe the response to variation in the test parameter where all other factors are as close as possible to non-limiting in terms of crop yield. Data points below this fine relate to samples where some other factor is limiting. For every case, ‘N’ fictitious Ys are derived. To determine the final Y, the minimum value achieved with all these maximum lines is chosen (Waggoner and Norvell, 1979): Y= Min{ f(x1), f(x2), …, f(xN )}

(2)

46


Fig. 2. Scatter plot of an independent variable Xi (here grain number m − 2 as an example) and a dependent variable Y (yield in kg ha − 1). It shows the difference between the linear regression and the maximum line.

where, f(x1), maximum attainable Y from the independent variable 1; f(x2), maximum attainable Y from the independent variable 2; f(xN ), maximum attainable Y from the independent variable N. Finally, only the independent variables needed to cover the experimental cases are selected. It is a mathematical adaptation of von Liebig’s Law (the thirty-third statement of the ‘Principles of Agricultural Chemistry’). This method was used instead of multiple linear regression because of higher prediction accuracy (Casanova et al., 1999).

3. Results The framework for understanding rice growth was based on three steps: (i) Step 1 related yield to the variables of yield components and influencing factors, using the boundary line method. (ii) Step 2 was divided in two in order to accentuate the interrelationships between plant growth and crop status. Step (2a) related the variables of crop status with the significant variables of yield components and influencing factors of Step 1, using correlation anal-


ysis. Step (2b) related yield directly to the variables of crop status, using the boundary line method. (iii) Step 3 related the significant variables of crop status in Steps 2(a) or 2(b) with soil properties, using correlation analysis.

3.1. Deducing significant yield components and some influencing factors from yield 6alues (Step 1) The relationship between yield and two yield components and three influencing factors is shown in Fig. 3. The yield components have a positive effect on yield. For an influencing factor, it is the other way around. Additionally, if such a factor is zero, the upper limit reaches potential yield. In these scatter plots, straight lines over the maximum levels were drawn based on the boundary line method. These lines (Y1, Y2, …, YN ) can be considered imaginary ‘yield plateaux’, as if this variable alone was limiting yield. On this basis, an operational model was developed for determining yield, with five variables sufficient to cover the experimental cases: PANO, panicles per square metre; SPNO, spikelets per square metre; GN, percentage of necrotic grain (just before harvesting); WI, intensity of weed infestation; and FFD, standard deviation in fPAR at flowering. Other yield components-such as the fraction of filled spikelets, the weight of 1000 dry grains and the number of spikelets per panicle (per se)-were not significant in this analysis. Based on these five variables and the model Eq. (2), yield was predicted (Fig. 4a). The intercept was not significantly different from 0 (P\ 0.05) and the slope was not significantly different from 1 (P \ 0.05). Fig. 4b illustrates the observed and predicted yields with only the two yield components. The predictions were accurate for high values, but not for low. Rice necrosis, spatial heterogeneity and weed infestation, where present, each had a strong effect, but were not reflected in the yield components. However, this information should be used with caution. The occurrence of one of these five variables as a limiting factor does not exclude the possibility that other plant parameters are also very close to limiting

47

3.2. Deducting significant 6ariables of crop status from yield components and some influencing factors (Step 2a) Partial correlation coefficients were calculated and tested for the significant yield components arid influencing factors of Step 1, and variables related to crop status. Spikelets per unit area (SPNO) was split into the number of panicles per unit area (PANO) and the number of spikelets per panicle (SPP). SPP, in contrast to PANO and SPNO, had a high scatter and no positive relation with yield. Only those correlations that were significant at the h= 0.05 level between variables are marked in Table 1. with a positive or a negative sign depending on the type of correlation The results shown in Table 1 highlighted four factors: (i) Potassium (K) was positively related to spikelet number per unit area (Table 1 Fig. 5aa). K concentration in the stems was a better indicator for diagnosis, probably because less than 20% of the absorbed K is translocated to the panicles (Yoshida, 1981). The causal physiological relationship is that K in the vegetative plant parts promotes the translocation of assimilates into grains and thus has a favorable impact on grain protein synthesis (Mengel et al., 1981). K had the opposite effect to sodium (Na). Na concentration in the straw at maturity was negatively related to panicle number, as shown in Table 1. (ii) Zinc (Zn) concentration in the grains at maturity had an effect on panicle number, spikelet number and fraction of necrotic grain, as shown in Table 1. Grain Zn concentration was a better diagnostic indicator, probably because up to 50% of the Zn absorbed by a rice crop ends in the panicle (Yoshida and Tanaka, 1969). Fig. 6b indicates that Zn is antagonistic to Na. (iii) The number of plants available is related to various factors that determine yield. Low and heterogeneous germination values were recorded within farmers’ fields. Data on plants and spikelets per unit area in the study are shown in Fig. 7.. Additional data from trials rising a long-grain cultivar at an

48


Fig. 3. Measured values in 1995 of yield for various cultivars (Bahia, Senia, Tebre and Niva) and yield components together with some influencing factors: (a) PANO (panicles m − 2); (b) SPNO (spikelet m − 2); (c) GN (percentage of necrotic grain); (d) FFD (heterogeneity; standard deviation in fPAR at flowering) and (e) WI (weed infestation; weeds m − 2 at ripening). The straight lines represent the maximum fitted lines.


experimental farm were incorporated Fig. 7 illustrates that a threshold of about 160–180 plants m − 2 was necessary to maintain constant grain number. (iv) The length of the growing period, and especially the pre-heading period, had significant effects on the final yield, in particular on the grain number per unit area (Fig. 8a and b).

49

3.3. Deducing significant 6ariables of crop status directly from yield 6alues (Step 2b) The law of the minimum of the limiting factors (boundary line method) was used to predict yield values directly from crop status. The relation of various variables of crop status to yield is shown in Fig. 8. Based on the model of Eq. (2), in order to cover all the experimental cases,

Fig. 4. Comparison of observed and predicted yield in 1995 from the model of Eq. 2 using (a) five variables (two yield componets and three influencing factors) and (b) two variables (just the two yield components). Labels indicate the main limiting variable: PANO (panicles m − 2); SPNO (spikelets m − 2); GN (percentage of necrotic grain); FFD (standard deviation in fPAR at flowering) and WI (weeds m − 2 at ripening).

50


Table 1 Significant correlation coefficients (at the h =0.05 level, n =40) of yield components and influencing factors with variables of the crop status Crop status Plant establishment Germination (%) Plant number CV of plant number ( fPAR)til ( fPAR)flo Plant height

PANO

SPP

SPNO

+0.42 +0.42 −0.43

GN

FFD

−0.40 −0.37 +0.46

Crop duration Length of vegetative st. Length of pre-heading st. Length of post-heading st. Length total

+0.43

−0.43

+0.37

Nutrient status Tillering N – leaves K – leaves P – leaves Ca – stems Na – stems Zn – stems K – stems Mg – stems Maturity Na – straw Zn – straw N – straw P – straw K – straw Ca – straw Mg – straw Na – grain Zn – grain N – grain P – grain K – grain Ca – grain Mg – grain

WI

−0.39 +0.41

−0.37

−0.41 +0.51

−0.38

+0.44

−0.42 +0.41

+0.41 +0.45

−0.39

+0.41

Yield components: PANO (panicles m−2), SPP (spikelets panicle−1), SPNO (spikelets m−2). Influencing factors: GN (fraction of necrotic grain), WI (weeds m−2 at ripening) and FFD (heterogeneity, the standard deviation of the fraction of fPAR absorbed at flowering). Crop status: ( fPAR)til is the difference between the optimum and the measured fPAR at tillering. ( fPAR)flo is the difference between the optimum and the measured fPAR at flowering. The other variables are easily identifiable.

six limiting factors or crop status variables were necessary: K/Nams, ratio of K (mg K per g dry matter) to Na (mg K per g dry matter) in the straw at maturity. Znmg, Zn concentration (mg Zn per g dry matter)

in the grain at maturity. Nmg, N (mg N per g dry matter) in the grain at maturity. PLNO, plants per unit area (plants m − 2). LT, length of the total growing period (days). L – PreH, length of the pre-heading period (days).


Yield was predicted as shown in Fig. 9a. The slope is significantly different from 1 at the h = 0.05 probability level, but not at the h =0.01. The intercept is significantly different from 0 at both probability levels. Using three independent variables (Fig. 9b), the adjusted regression coefficient is similar to that of Fig. 9a, but the intercept and slope are both significantly different from 0 and 1 respectively, even at the h =0.01 probability level.

51

3.4. Deducing significant soil properties from crop status (Step 3) Partial correlation coefficients were calculated and tested between all measured soil properties and the variables of crop status that were significant in Steps 2(a or b). Only those correlations that were significant at the h= 0.05 probability level are shown in Table 2, with a positive or a negative sign depending on the type of correlation. Note that

Fig. 5. Spikelet number in relation to K concentration in the straw at maturity (a) Na and K concentration in the straw at maturity (b).

+0.36

+0.40

+0.35 +0.40

+0.36

Plant number

−0.47

−0.51 −0.44

−0.42 +0.37

+0.36 −0.37

−0.45 +0.38

Length total

+0.37

+0.51

Length pre-heading

−0.45

−0.54

CV of plants

+0.43

+0.41 −0.39

+0.37

Tillering Nleaves

Tillering Pleaves

−0.35

+0.38 −0.39

+0.55 +0.59 −0.56 +0.49

+0.48 +0.37 +0.45 −0.37

−0.51 −0.47 −0.40 +0.37

+0.61 +0.66 +0.62

Maturity Kstraw

−0.40 +0.36

−0.39 −0.51

Maturity Znstraw

+0.52 −0.42

Maturity Nastraw

−0.50

+0.37

Maturity Nstraw

−0.35 +0.52 −0.56

−0.39

+0.50 −0.40

Maturity Nagrains

+0.40 +0.50 −0.53 +0.49

+0.40

−0.53 −0.55

−0.43 +0.43

Maturity Zngrains

−0.35

Maturity Ngrains

Soil properties: development of structure is assigned 1 while non-development is 0. Drainage class is ranked from well drained (1) to very poorly drained (9). Coarse texture (1) was assigned for a sandy particle-size class (0 if non-coarse). EC1 is the soil salinity averaged from 0 to 120 cm depth before sowing. EC2 is the topsoil salinity before sowing. EC3 is the topsoil salinity at the panicle initiation stage. Other soil parameters are easily identifiable.

Development of structure Drainage class Coarse texture Clay (%) between 40 and 60 cm Clay(%) between 80 and 100 cm EC1 (dS m−1) EC2 (dS m−1) EC3 (dS m−1) Topsoil CEC (meq 100 g−1) Topsoil pH Topsoil clay (%) Topsoil sand (%) Topsoil silt (%) Topsoil SOM (%) Topsoil P (ppm) Topsoil K (ppm) Topsoil Mg (meq 100 g−1) Topsoil N (%)

Germination (%)

Table 2 Significant correlation coefficients (at the h=0.05 level, n =40) of significant variables of crop status with soil properties

52 D. Casano6a et al. / Europ. J. Agronomy 17 (2002) 41–61


53

Fig. 6. Spikelet number (SPNO) in relation to Zn concentration at maturity in the grain (a). Zn and Na concentration at maturity in the plant (b).

drainage status of the soil profile had no significant effect. The main soil properties with significant effect on rice growth can be grouped as: (i) Soil salinity had negative effects on the plant K and Zn. Fig. 10aa shows K concentration in the straw at maturity against soil salinity (of the saturated paste, averaged for the top 120 cm of the soil profile). The exchangeable K in the soil was grouped in four classes, and the exact values are indicated as labels. Fig. 10a illustrates that, with high soil salinity, straw K concentration was low even when high values of exchangeable K were mea-

sured m the soil. This indicates that in soils with high soil salinity, K absorption is hampered, leading to low levels of K in the plant and consequent growth reduction. Fig. 10b shows how Zn was negatively related to soil salinity, although not as strongly as K Soil Zn concentration was not measured. (ii) Soil organic matter, N, K and exchangeable Mg, together with the clay percentage and CEC of the topsoil, had positive effects on the homogeneity of plants per unit area. (iii) The clay and silt percentage, as well as the CEC of the topsoil, had positive effects on

54


plant Zn and K concentrations, and on crop growth duration. It is unclear if this effect is directly through soil fertility or indirectly through a higher retention capacity of the fertilizers. Fig. 11 illustrates how low clay fractions in the topsoil induced Zn shortages. Furthermore, in topsoils with a high clay fraction, Zn absorption was reduced by high soil salinity. (iv) pH had positive effects on the plant Zn and K concentrations. (v) Favorable N levels in the plant were promoted by low salinity, high pH, and fine soil texture.

4. Discussion The use of mathematical techniques provide a basis for understanding the processes of rice yield determination. We followed a deduction process in which there is a risk of deriving relations that are not really causal. Thus, knowledge of crop response to yield determining factors is necessary for interpreting the results (McBratney and Pringle, 1997). Yield values were predicted with high accuracy from yield components and some influencing factors (r 2adj =0.94), and to a lesser extent from the

variables related to crop status r 2adj = 0.76). This is partly due to the increase in causal distance between yield and the variables considered. Based on this reasoning, predicting yield from soil properties would give a lower regression coefficient. Such results were found in Casanova et al. (1999) and have been reported by other authors (Bouman, 1994; Greenwood et al., 1986). Sinclair and Park (1993) noted ‘the inadequacy of the limiting-factor paradigm because of the great flexibility of plants to acclimate morphologically and physiologically to changing environmental conditions’. This paper shows that this may be true within a given long causal distance between the limiting factor and yield, but not so for short causal distances. Therefore, the validity of the limiting factor paradigm depends on the causal distance between variables, as well as on other factors, such as the range or variability of the variables. The boundary line method allows one to quantify the overall deficiency in yield with respect to the best performance, and it provides information for deciding if investigations to increase yield are justified. The deduction process used in this paper proved to be adequate for sorting out the basic soil and rice plant processes, so that the agricultural potential of the area can be reached. Such a methodology can also be extrapolated to other

Fig. 7. Spikelet number against plants per unit area for four cultivars in 1995.


55

Fig. 8. Scatter plots of yield and variables related to the crop status: (a) L – T (length of the growing cycle): (b) L – PreH (length of the pre-heading period); (c) PLNO (plant number per square metre); (d) Kmg/Namg,. (potassium to sodium ratio at maturity in the straw); (e) Znmg (zinc concentration at maturity in the grain) and (f) Nmg (nitrogen concentration at maturity in the grain).

56


Fig. 9. Comparison of observed and predicted yield from (a) six and (b) three variables related to the crop status using the model of Eq. 2. Labels indicate the main limiting source: L – T (length of the growing period); L – PreH (length of the pre-heading period); PLNO (plants per unit area); K – Na (potassium to sodium ratio at maturity in the straw); Zn (zinc at maturity in the grain) and N (nitrogen at maturity in the grain).

crops and agricultural areas. Data and results obtained in this type of research are necessary for developing operational decision-support systems to improve decision making at farm level.

4.1. Deducing significant yield components and some influencing factors Panicle number and spikelet number, as yield components, were expected to be significantly,


related to yield because they form the sink capacity of the rice plant. Necrotic grain showed a high scatter with yield. Fig. 3c showed that a quick ‘rule-of-thumb’ yield estimate can be given prior to harvest. Intensity of weed infestation, if high, has a strong effect on yield.

4.2. Deducing significant 6ariables of crop status Step (2b), which related yield directly to crop status, showed that the significant variables were similar to those in Steps 1 and 2a. The various nutrient concentrations measured at maturity were

57

more highly correlated with the yield components and influencing factors than those measured at tillering. This effect, presumably, was caused by the progressive dilution (deficiency,) or concentration (toxicity) of the nutrients during the growing cycle. Five main groups of factors seem to be related to poor rice growth: (i) K deficiency, reduced significantly the panicle and spikelet numbers. High Na levels were also associated with heterogeneity within the fields. Fig. 5b illustrates the antagonistic effect of K and Na. A K/Na ratio represents this relation in the dataset (Fig. 8d). A low

Fig. 10. K concentration in the plant against soil salinity (a). Zn concentration in the grains against soil salinity (b).

58


Fig. 11. Zn concentration in the grains against clay (%) in the topsoil. Labels indicate soil salinity (electrical conductivity of the saturated paste averaged in the profile down to 120 cm).

K/Na ratio at the plant straw at maturity may be used indirectly as a first estimate of low production. Other authors (Qadar, 1995; Girdhar, 1988) reported that if high Na levels exist in the plant, the K levels are low. According to Yoshida (1981), the critical value for K is 10 mg K/g dry matter. In our dataset, the average K at tillering in the leaf blades was 22 mg K/g dry matter, with a standard deviation (S.D.) of 9 3. The average K at maturity in the straw was 15 mg K/g dry matter, with a S.D. of 4. These values were apparently above the critical limits of Yoshida (1981), yet, the scatter plots in Fig. 5 show that, for our farm management conditions and soil units, the saturation level of K was not reached. Presumably potassium deficiency may be induced by Na toxicity. This is in accordance with Khan et al. (1992b), Bohra et al. (1995). (ii) Zn deficiency, shown by low Zn concentration in the grains at maturity, was associated with panicle and spikelet number, and also with grain necrosis. Zn deficiency in lowland rice has been widely reported on submerged soils (Patrick et al., 1985; Khan et al.,

1992a,b; Beyrouty et al., 1994; Sharif Zia et al., 1994). Concentration of Zn in the soil solution generally decreases following flooding, because of precipitation as ZnCO3, ZnSO4. or Zn(OH)2 (De Datta, 1981). It is well known that this was the case in Pant Nagar in Uttar Pradesh, India (Yoshida and Tanaka, 1969). According to these authors, critical Zn concentrations in the shoot are 15–20 mg Zn /kg dry matter. In the dataset, the average Zn concentration in the shoot at tillering was 31.5 mg Zn /kg dry matter, but at maturity only 22.3 mg Zn /kg dry matter, with 7 values below 20 mg Zn /kg dry matter. These low values of Zn, together with its relation to yield components and influencing factors (shown in Fig. 6) induced a major inhibition of rice growth due to Zn shortage. (iii) Sufficient plant numbers are important to ensure enough panicles, and also for reducing weed infestation. Not only plant number but also homogeneity within a field had significant effects on weed infestation and on the fraction of necrotic grain. Data shown in Fig. 7. as well as Fig. 8c, imply that a threshold value of 160–180 plants m − 2 should be sought. Farmers sow between 500 and 600


seeds per square metre, and low germination percentages may occur if low temperatures or strong winds occur. Stands are difficult to restore with re-seeding because the oxygen concentrations in the soil and water are very low by the time a re-seeding decision is made (Hill et al., 1991). (iv) Temperature influenced the length of the growing period, and especially the pre-heading period and had significant effects on yield and the number of available spikelets per unit area Rice in temperate conditions is grown in summer under high solar radiation According to Angus et al. (1993), with full radiation commonly available, temperatures are indirectly limiting rice production either by effects on spikelet sterility or on development rate. Our results are in accordance with Dingkuhn et al. (1991), who showed that biomass at maturity was a function of crop duration. (v) High N concentration in the grains at maturity indicated that something else limited growth and as a result there was a surplus of N. (vi) fPAR values at flowering did not have significant effects on yield This result indicates that, in the study area, there is sufficient maximum absorbed fPAR or, in other words, there is an adequate maximum leaf area index (LAIm). A field may have areas with low end high fPAR simultaneously, as the results indicated with the deviation of fPAR at flowering. It is a matter of homogeneity or uneven growth within a field. Also note that variables such as the height of the rice plant at ripening and the deviation of fPAR from the optimal at tillering, did not have a significant influence. The fact that the effect of fPAR values at tillering and at flowering was not significant implies that most of the fields started well, or without showing any strong difference that would determine future yield. These results are in accordance with Ramasamy et al. (1997), who showed that most differences in rice growth due to soil properties are not clear before flowering, and mainly show up in the period between flowering and maturity.

59

4.3. Deducing significant soil properties Table 2, Figs. 10 and 11 illustrate the relation of soil properties to crop status. Soil submergence causes a substantial decrease in Zn concentration in the soil solution. Obermueller and Mikkelsen (1974) and Jugsujinda and Patrick (1977) found that increased tissue concentrations and uptake of Zn occurred with rice grown under aerobic conditions. After prolonged submergence, the Zn concentration declines towards a lower limit. According to Forno et al. (1975), bicarbonate and carbonate have inhibitory effects on Zn absorption in alkaline soils. In this study, Zn deficiency was not associated with high pH (extract 1:2.5 had a mean of 7.9 and a S.D. of 0.2). In this area, it is mostly associated with poor soil fertility (low clay and silt concentration) and soil salinity (Casanova et al., 1999). Both Zn and K were adversely affected by soil salinity, as shown in Fig. 10a and b. These results indicate that in soils with high soil salinity, K and Zn absorption was hampered, and only low levels of these canons were reached in the plant. Khan et al. (1992a,b) found that application of Zn improves the growth, yield and nutrition of rice even at 16 dS m − 1 salinity level. Further research in Zn assimilation and interaction with other cations under submerged conditions is necessary. Greater topsoil clay and silt concentrations also improved the levels of Zn and K, crop duration and plant establishment homogeneity.

5. Conclusions The results, based on monitoring of rice growth and measurement of soil properties in a wide range of farmers fields in the Ebro Delta (Spain), can be summarized as: (i) The source capacity, measured as the fraction of intercepted PAR at tillering and at flowering, rarely limited rice growth. Rather, the sink capacity, in terms of available panicles and spikelets, was often a limiting factor. (ii) Influencing factors, such as the fraction of necrotic grain, intensity of weed infestation and spatial heterogeneity within fields, were needed to quantify yield at field level.

60


(iii) A minimum of 160– 180 plants m − 2 was required to stabilize spikelet number and, additionally, to reduce spatial heterogeneity and weed infestation. (iv) Final yield was a function of the length of the growing season. The length of the pre-heading period, when the potential size of the crop is primarily determined, was of key importance. (v) Rice growth was limited strongly by K and Zn shortage. In both cases, there was strong antagonism with Na. (vi) K and Zn deficiencies in the plant were mainly induced by excess Na. High Na levels in the plant were caused by soil salinity. (vii) Long growing cycle, homogeneity of plants per unit area, and adequate K and Zn levels in the plant were favored by high clay concentrations of the topsoil. The actual minimum and maximum recorded yield varied from 4 to 11 t ha − 1. Hence, a high variability in rice growth and yield was found. The reason for such discrepancies is the unbalance between land characteristics (soil properties variability and farming practices based on ‘What the neighbor does’) and rice requirements (crop status, weather and soil conditions). Yield differences between soils and farming practices were compared and analyzed. This research offers the potential to move beyond ‘what the neighbor does’. For instance, it shows the importance of measuring plant establishment in its own parcel rather than checking the neighbour’s sowing density. The deductive process shown in this paper proved to be adequate for elucidating the basic soil end ricecrop processes, so as to derive ways and means to further improve the agricultural production system at field level. Such an approach could be used in other agricultural areas where there is a lack of data for identifying the main yield-limiting factors. Future research should be oriented towards developing decision support systems for major land units, combining soil, climatic and agronomic settings. Future farmers and decision-makers should be able to decide themselves what to do based on a thorough knowledge of the natural processes at field level, rather than repeat their neighbour’s practices or take arbitrary decisions!

Acknowledgements The authors are grateful to the staff of the IRTA experimental station in the Ebro Delta for providing data for this paper. Thanks are also due to J. Llop, J. Raco´ and A Forcadell for their field measurements. The authors acknowledge the assistance of Dr J. Bourna and Dr G.F. Eperna in revision of the manuscript, and Dr A Stein for support in developing the operational methodology based on von Liebig’s Law, all from the Laboratory of Soil Science and Geology at Wageningen Agricultural University.

References Angus, J.F., Bacon, P.E., Reinke, RF., Williams, RL., 1993. Rice yield variability in Australia in relation to cool-temperature damage, N fertilizer and water depth. In Proceedings Third Symposium on the Impact of Climatic Change on Agricultural Production in the Pacific Rim Central Weather Bureau, Taipei, pp. 11 –28. Beyrouty, C.A., Grigg, B.C., Norman, R.J., Wells, B.R., 1994. Nutrient uptake by rice in response to water management. J. Plant Nutr. 17(I), 39 – 55. Bohra, J.S., Do¨ rffling, H., Do¨ rffling, K., 1995. Salinity tolerance of rice (Oryza sati6a L.) with reference to endogenous and exogenous abscisic acid. J. Agron. Crop Sci. 174, 79 – 86. Bouman, B.A.M., 1994. A framework to deal with uncertainty in soil and management parameters in crop yield simulation: a case study for rice. Agric. Syst. 46(I), 1 – 17. Casanova, D., Eperna, G.F., Goudriaan, J., 1998. Monitoring rice reflectance at field level for estimating biomass and LAI. Field Crops Res. 55, 83 – 92. Casanova, D., Eperna, G.F., Bourna, J., Goudriaan, J., 1999. Yield-gap analysis in relation to soil properties in directseeded flooded rice. Geod. 91, 191 – 216. Casanova, D., Goudriaan, J., Bosch, A.D., 2000. Testing the performance of ORYZAl, an explanatory model for rice growth simulation, for Mediterranean conditions. Eur. J. Agron. 12, 175 – 189. Catala´ Fomer, M.M., Fosch, T., 1997. Estudio de los factores que inciden en el rendimiento del arroz en el Delta del Ebro (Study of the factors that influence rice production in the Ebro Delta). Agric. Vergel. 3, 120 – 128. CROPSCAN Inc., 1993. Multispectral Radiometer (MSR): User’s manual and technical reference. CROPSCAN Inc., Rochester. Decagon Devices Inc., 1989. Sunfleck Ceptometer User’s Manual. Decagon Devices Inc., Pullman (WA). De Datta, S.K., 1981. Principles and Practices of Rice Production. Wiley, Singapore.

D. Casano6a et al. / Europ. J. Agronomy 17 (2002) 41–61 Dingkuhn, M., Schnier, H.F., De Datta, S.K., Dorffling, K., Javellana, C., 1991. Relationships between ripening-phase productivity and crop duration, canopy photosynthesis and senescence in transplanted and direct-seeded lowland rice. Field Crops Res. 26, 327 – 345. Forno, D.A., Yoshida, S., Asher, C.J., 1975. Zinc deficiency in rice. I. Soil factors associated with the deficiency. Plant Soil 42, 537 – 550. Girdhar, I.K., 1988. Effect of saline irrigation water on the growth, yield and chemical composition of rice crop grown in a saline soil. J. Ind. Soc. Soil Sci. 36, 324 –329. Gomez, K.A., Gomez, A.A., 1984. Statistical Procedures for Agricultural Research, second ed. Wiley, New York. Greenwood, D.J., Neeteson, J.J., Draycott, A., 1986. An analysis of the dependence of yields of potatoes and vegetable crops on soil properties and weather conditions. J. Sci. Food Agric. 37 (7), 674 – 675. Hill, J.E., Bayer, D.E., Bocchi, S., Clampett, W.S., 1991. Direct seeded rice in the temperate climates of Australia, Italy, and the United States. In: Direct Seeded Flooded Rice in the Tropics. Int. Rice Res. Inst, Los Ban˜ os, pp. 91 – 102. Jugsujinda, A., Patrick, W.H., 1977. Growth and nutrient uptake by rice in a flooded soil under controlled aerobic – anaerobic and pH conditions. Agron. J. 69, 705 – 710. Khan, H.R., Yasmin, K.F., Adachi, T., Ahmed, I.U., 1992a. Effects of gypsum, Zn and intermittent saline irrigation on the growth, yield and nutrition of rice plants grown in a saline soil. Int. J. Trop. Agric. 38 (3), 421 –429. Khan, H.R., Faiz, S.M.A., Islam, M.N., Adachi, T., Ahmed, I.U., 1992b. Effects of salinity, gypsum and Zn on mineral nutrition of rice. Int. J. Trop. Agric. 10 (2), 147 –156. Kropff, M.J., Van Laar, H.H., Matthews, R.B., 1994. ORYZA1: an ecophysiological model for irrigated lowland rice production. SARP Research Proceedings. DLO-CAB and WAU-TPE, Wageningen, and Int. Rice Res. Inst., Los Ban˜ os. Liebig, J. von, 1855. Principles of agricultural chemistry with special reference to the late researches made in England. 17– 34. Reprinted in: Cycles of essential elements (Benchmark papers in Ecology Vol. I, L.R Pomeroy (1974)). Dowden, Hutchinson and Ross Inc., Stroudsburg, Pennsylvania, pp. 11 – 28. MAPA, 1986. Me´ todos oficiales de ana´ lisis de suelos. Vol 3. Madrid. Matsushima, S., 1957. Analysis of developmental factors determining yield and yield prediction in lowland rice. Bull. Natl. Inst. Agric. Sci. A5, 1 – 271.

61

McBratney, A.B., Pringle, M.J., 1997. Spatial variability in soil —implications for precision agriculture Precision Agriculture’97. In: Stafford, J.V. (Ed.), Spatial Variability in Soil and Crop, vol. 1. BIOS, Oxford, pp. 1 – 31. Mengel, K., Secer, M., Koch, K., 1981. Pottassium effect on protein formation and amino acid turnover in developing wheat grain. Agron. J. 73, 74 – 78. Obermueller, A.J., Mikkelsen, D.S., 1974. Effects of water management and soil aggregation of the growth and nutrient uptake of rice. Agron. J. 66, 627 – 632. Patrick, W.H., Mikkelsen, D.S., Wells, B.R., 1985. Plant nutrient behavior in flooded soil. In: Fertilizer Technology and Use, third ed. Soil Science Society of America, Madison, WI, pp. 197 – 228. Qadar, A., 1995. Potassium and sodium concentrations of shoot and laminae of rice cultivars and their sodicity tolerance. J. Plant Nutr. 18 (11), 2281 – 2290. Ramasamy, S., ten Berge, H.F.M., Purushothaman, S., 1997. Yield formation in rice in response to drainage and nitrogen application. Field Crops Res. 51, 65 – 82. Richards, L.A., 1954. Diagnosis and Improvement of Saline and Alkali Soils. USDA Agriculture Handbook No. 60. Schnug, E., Heym, J., Murphy, D.P., 1995. Boundary line determination technique (BOLIDES). In: Robert, P.C., Rust, R.H., Larson, W.E. (Eds.), Site-Specific Management for Agricultural Systems. ASA/CSSA/SSSA, Madison, WI, pp. 899 –908. Sharif Zia, M., Salim, M., Aslam, M., Gill, M.A., Rahmatullah, S., 1994. Effect of low temperature of irrigation water on rice growth and nutrient uptake. J. Agron. Crop Sci. 173, 22 – 31. Sinclair, T.R., Park, W.I., 1993. Inadequacy of the Liebig limiting-factor paradigm for explaining varying crop yields. Agron. J. 85, 742 – 746. SSS [Soil Survey Staff], 1996. Keys to Soil Taxonomy (Seventh Edition). USDA-NRCS. Pocahontas Press, Blacksburg (VA). Waggoner, P.E., Norvell, W.A., 1979. Fitting the Law of the Minimum to fertilizer applications and crop yield. Agron. J. 71, 353 – 354. Webb, R.A., 1972. Use of the boundary line in the analysis of biological data. J. Hortic. Sci. 47, 309 – 319. Webster, R., Oliver, M.A., 1990. Statistical Methods in Soil and Land Resource Survey. Oxford University Press, New York. Yoshida, S., Tanaka, A., 1969. Zinc deficiency of the rice plant in calcareous soils. Soil Sci. Plant Nutr. 15, 75 – 80. Yoshida, S., 1981. Fundamentals of Rice Crop Science. Int. Rice Res. Inst, Los Ban˜ os.