Cropping system diversification for food production in Mindanao rubber plantations: a rice cultivar mixture and rice intercropped with mungbean

Experimental sites

The experimental sites were in the Arakan Valley Complex in the province of Cotabato in Central Mindanao, Philippines. Arakan has 28 villages, 10 of which are major rice producing villages. Its total land area is 69,432.79 ha with about 16,798 ha utilized for crop production. The landscape of Arakan is dominated by rolling hills, valleys, and mountain ranges.

Planting design

We evaluated the set of eight cropping treatments (Table 1), described in more detail below, in farmers’ fields for three seasons (2006–2008), with planting date details in Table S1. In 2006, two fields, one with 1-year-old rubber trees and one with 3-year-old rubber trees, were selected in each of three municipalities—Antipas, Arakan, and Pres. Roxas—to represent local systems. The experimental design in 2006 was a split-plot design with rubber tree age as the whole plot treatment (Fig. S1A). Each farm contained three complete blocks, with each block containing 10 m × 4 m subplots that were assigned to the eight intercropping treatments (Table 1; Fig. 1 and S1B).

Due to water-logging in some municipalities, the experiments in 2007–2008 were moved. During these years, two-year experiments were conducted in three fields with 1- to 2-year-old rubber trees in the municipality of Arakan, in the villages of Doroluman, Naje, and Sabang. The experiment was established in 2007 on 17 July (Doroluman, with 1-year-old rubber), 12 June (Sabang with 1.5-year-old rubber), and 16 May (Naje with 2-year-old rubber) (Table S1). The experimental design in 2007–2008 was a split-plot design with the two years as repeated measures (Fig. S2). The whole plots were again individual farms, with farm effects considered a random effect. Each farm again contained three complete blocks, with the eight intercropping treatments (Table 1) applied to the subplots.

Eight rice and mungbean treatment combinations (Table 1) were planted in plots between the rubber tree rows. Two rice cultivars were included in the design: Dinorado, a higher-value but lower-yielding cultivar, and UPL Ri-5, a higher-yielding but lower-value cultivar, along with a high-yielding mungbean variety, Pag-asa 7. Subplots were positioned between the rows of rubber trees. Each subplot was 10 m long and included 10 rows of rice and/or mungbean, with 0.4 m between rows and 0.2 m between hills for both rice and mungbean. The distance between the subplots and the base of the rubber trees was 1–1.5 m, and there was a 4 m border between the experimental plots to reduce interplot interference. In general pesticides were not used, with the exception of the Oreta one-year-old rubber farm in 2006. Hand weeding was implemented following standard farmer practices to support establishment of the crops.

Statistical analysis

The response variables evaluated were crop yield, land equivalent ratio, weed biomass, crop height, and the disease and pest ratings for each crop. The responses represent a range of different probability distributions for statistical analyses. Several response variables were approximately normally distributed, so these data were analyzed in a linear mixed model using the MIXED procedure in SAS (Version 9.2). The details of the ANOVAs are given in Tables S2 and S3, following the design structures in Figs. S1 and S2.

The same model structure was used for generalized linear mixed models for response variables that were not normally distributed, such as some disease ratings. The assumption of normality was tested using a Shapiro–Wilk test, and a Q-Q plot was evaluated where a heavy tail suggested use of a gamma distribution. Because of an upward skew in some data, the gamma distribution with log link function was used to analyze those responses using SAS Proc GLIMMIX. The ilink option was used for calculating the least square means. A Tukey–Kramer adjustment (at significance level 0.05) was used in multiple comparisons of performance within a crop across the different treatments in which that crop occurred. Weed dry weight was evaluated for each treatment (not evaluated separately for different crops). Weed weight was approximately normally distributed after square root transformation. We evaluated the effects of rubber tree age and cropping treatments (including the no-crop treatment) on weed weight in a linear mixed model. Ratings of disease severity were treated in analyses as approximately continuous.

Yield and economic value of intercrops: rice and mungbean

Both UPL Ri-5 and Dinorado have been preferred rice varieties of Arakan farmers for various reasons (RF Hondrade, pers. obs., 2006). Both mature in approximately 128 days. Their cooked grain quality is acceptable to consumers, while Dinorado has greater volume increase and better palatability. UPL Ri-5 is higher yielding (average yield approximately 2.5–3.5 t ha⁻¹ under favorable conditions) compared to Dinorado (yield less than 2 t ha⁻¹ up to 2.7 t ha⁻¹ under the same management practices). However, milled Dinorado had a local market price of more than a dollar (in Philippine pesos, P45–50 kg⁻¹, compared to UPL Ri-5 at P32–P35 kg⁻¹). Higher yielding UPL Ri-5 helps ensure farmers’ household food security while the higher market value of Dinorado provides additional income to the farmer. The average yield of Dinorado in Arakan ranged from <1 to 1.5 t ha⁻¹ while UPL Ri-5 produced 1 to 2 t ha⁻¹. The national average yield of upland rice was <1 to 1 t ha⁻¹.

Mungbean is a preferred legume of local farmers in Arakan, with a market price similar to Dinorado (RF Hondrade, pers. obs., 2006). Average yields under favorable growing conditions in the uplands ranged from 0.9 to 1.7 t ha⁻¹. Farmers grow mungbean both for market and household food consumption.

In addition to the analysis of grain yield described above, yield was also used to calculate the corresponding land equivalent ratio (LER) for evaluating the rice mixture and the rice-mungbean intercrops in each subplot. The LER for each subplot was calculated using the following formula, illustrated for a two crop mixture. ${\displaystyle P_1\cdot \frac{{\text{Yield}}_{\text{1}}\text{in mixture}}{{\text{Yield}}_{\text{1}}\text{in monoculture}}+P_2\cdot \frac{{\text{Yield}}_{\text{2}}\text{in mixture}}{{\text{Yield}}_{\text{2}}\text{in monoculture}}}$

where P_i is the proportion of rows planted to Crop i in the subplot being considered (Table 1), and Yield_i is the yield of Crop i calculated based on the dry (air- or sun-dried) weight per row. The LER for the three crop mixtures (with Dinorado, UPL Ri-5, and mungbean) was calculated similarly.

The LER for each mixture is a measure of how well the mixture or intercrop yield compared to the monoculture yield. Pair-wise comparisons between the means of the LER for the four cropping treatments (Table 1) that included more than one type of crop were performed. A one-tailed t-test was performed to see if there was evidence that the four LER means were larger than 1.

Crop weed, disease, and insect evaluation

In each of the subplots, including the control, the weed biomass was weighed at the end of the season. Common diseases were evaluated visually by an experienced disease observer, as described below. A range of beneficial and pest arthropod species were also sampled within the subplots, including the rice bug (Leptocorisa acuta). The following diseases and insect pests were evaluated at crop maturity, where evaluation generally followed standard rating methods (INGER Genetic Resources Center, 1996). For rice panicle blast (caused by Magnaporthe oryzae), three categories of infection were recorded: no visible lesion, lesions on several pedicels or secondary branches, or lesions on few primary branches or the middle part of the panicle axis. Rice leaf blast (caused by Magnaporthe oryzae) was evaluated in 2007 and 2008, using a scale of ten possible categories of infection (INGER Genetic Resources Center, 1996). Rice brown spot (caused by Cochliobolus miyabeanus) was evaluated on a scale of 10 potential levels of severity. Pod rot of mungbean (caused by Gluconobacter sp.) was evaluated in 2008 based on percentage incidence, the percentage pods infected. Rice bug damage was evaluated in 2006 by damage category.

Analysis of relative yield ranking of treatments, and stability of yield rank across environments

To evaluate the relative yield and yield stability of the different cropping systems, we compared cropping system performance across the different environments, in an analysis similar to the Mundt (2002) analysis of wheat cultivar mixture performance. The analysis was performed as follows. The mean yield per row was computed for each treatment in each site for the three years. For each of the twelve site-years, the weighted mean yield across the seven cropping systems in Table 1 (excluding the control) was calculated, as an index of the quality of that site-year as an environment for crop production (“environment index”). For each site-year, the rank of each treatment mean yield was also computed (1 = lowest, 7 = highest), indicating the relative performance of each treatment in each site-year.

The performance of the intercropping systems was evaluated in two analyses, one based solely on yield and one based on yield weighted by economic value. For the first, we used a regression analysis to evaluate the relationship between the environment index (mean yield in each site-year) and the yield ranks for each treatment. The best intercropping system will have the highest mean yield rank and will maintain its yield rank across environments. In the regression analysis, a positive slope indicates higher performance rank in high yield environments, while a negative slope indicates higher performance rank in low yield environments. Lower P-values associated with the slope indicate stronger evidence that the performance rank of a cropping treatment changes with environment. A relatively small mean square error from the F test in the regression analysis indicates that the quality of the environment explains most of the variability in performance rank of a cropping system. Second, we performed the same type of analysis but with the crop yields weighted by their relative economic value. The economic value of Dinorado and mungbean was approximately equivalent, and 40% higher than the value of UPL Ri-5, based on typical valuation of these crops. Thus relative economic weights were 1 for UPL Ri-5, 1.4 for Dinorado, and 1.4 for mungbean, and the treatment mean yield per row was multiplied by these weights in the economic analysis.

Rubber tree age effects

Rubber tree age did not have a significant effect on cropping system productivity (Table 2). Although the mean subplot yield in one-year-old rubber sites was 1.5 times that in three-year-old rubber sites, the variation among sites was very high (with yield 4.3 times higher in the highest yielding one-year-old rubber site compared to the lowest yield one-year-old rubber site).

Rice and mungbean yield responses to cropping system

Rice yields per row were greater in intercropping systems than in monoculture in some cases, although yields varied widely within and among years (Fig. 2 and Tables 2–5). There was a tendency for rice yields to be higher when rice made up a smaller proportion of the intercrop rows (Fig. 2 and Figs. S3). Dinorado yield responded significantly to intercropping treatments only in 2006 (p = 0.03), when Dinorado yield per area was significantly higher in 0.8 MB than in 0.2 MB (Tables 2 and 3). UPL Ri-5 yield also responded significantly to intercropping treatments only in 2006 (p = 0.001). UPL Ri-5 yield per area was significantly higher in 0.8 MB compared to the monoculture and RM in 2006 (Table 3). There were no significant treatment effects or significant mean differences in pairwise comparisons for the two rice varieties in 2007 and 2008 (Tables 3 and 2). As expected, UPL Ri-5 yield was generally higher than Dinorado yield.

Mungbean yield did not respond significantly to intercropping treatments in 2006 (Table 3). Mungbean also showed a tendency for higher yield when mungbean made up a smaller proportion of the intercrop rows (Fig. 2 and Fig. S3). There was a significant treatment effect for mungbean yield in 2007 and 2008 (p = 0.03; Table 2). There was a significant treatment × year interaction (p = 0.002). In 2007 and 2008, mungbean yield per area was significantly higher in 0.2 MB compared to 0.8 MB (Table 3). Mungbean data were missing from one farm in 2006 and there were also other cases of missing data that reduced the statistical power for mungbean comparisons. One component of variability in mungbean response was the observed 90% severity of pod rot in one site (P. Roxas) in 2006.

Land equivalence ratios

The estimated LER for the rice mixture varied with year (Fig. 3; Table 4). The intercrop LER estimate for 0.2 MB was significantly greater than 1 in all three years, and 0.5 MB and 0.8 MB were also significantly greater than 1 in two out of three years. The intercropping treatments differed significantly in their effect on LER in 2006 (p = 0.006; Table 4). The mean LER was significantly greater in RM and lower in 0.8 MB in 2006 (Table 4). Also, the LER for RM was significantly greater than one. In 2007 and 2008 there was a significant intercropping treatment effect on LER (p = 0.03). Among the pairwise comparisons, the 0.5 MB LER was significantly greater than the RM LER. Also the 0.5 MB LER mean was significantly greater than one, but the mean of RM was not significantly less than 1.

Weed responses to cropping systems

Weeds commonly observed in the experiments were Ageratum conyzoides (Asteraceae), Borreria laevis (Rubiaceae), Calopogonium mucunoides (Fabaceae), Chromolaena odorata (Asteraceae), Ludwigia octovalvis (Onagraceae), Murdannia nudiflora (Commelinaceae), and Rottboellia cochinchinensis (Poaceae), where A. conyzoides was particularly abundant. As expected, all crop treatments resulted in lower weed biomass than the unplanted control (Table 5). The mungbean monoculture had the lowest weed biomass in 2007–2008, and was significantly lower than the weed biomass in the Dinorado monoculture and RM; however, in 2006 the mungbean monoculture did not significantly reduce weed biomass (Table 5), probably because of problems with mungbean establishment in that year. The intercropping treatment effects were significant in both 2006 and 2007–2008. There was also a significant treatment by year interaction in 2007–2008 (Table 5).

Disease and insect responses to cropping treatment

Disease and insect pests were present at low levels overall (Tables S10 and S12). Brown spot, panicle blast, leaf blast, and rice bug were common enough in at least some years to be evaluated for differential responses to the intercropping treatments. Brown spot (caused by Cochliobolus miyabeanus) was generally more severe in Dinorado than in UPL Ri-5, and there was evidence for decreased brown spot in Dinorado in the rice mixture and increased brown spot in Dinorado in the 0.8 MB intercrop in 2007–2008 (Table S12). Panicle blast (caused by Magnaporthe grisea) was generally more severe in UPL Ri-5 compared to Dinorado in 2006 (Table S10), but more severe in Dinorado than in UPL Ri-5 in 2007–2008 (Table S12). There was evidence that panicle blast was higher in Dinorado in the 0.2 MB intercrop and lower in the 0.5 MB intercrop in 2006 compared to other treatments (Table S10). Leaf blast could be evaluated in 2007–2008. Leaf blast severity was generally higher in Dinorado than in ULP Ri-5 (Table S12), but there were no significant cropping treatment effects (Table S13). In 2007–2008 there was a year effect on brown spot and leaf blast in UPL Ri-5 (Table S13). In 2006 there was an effect of rubber age on panicle blast in UPL Ri-5, where older rubber trees were associated with higher disease (Table S11). Pod rot of mungbean was observed at one site in P. Roxas in 2007–2008 where severity reached 90%, and the treatment effect was significant (Table S13).

The rice bug, Leptocorisa acuta, could be evaluated in 2006 and was generally more abundant in UPL Ri-5 than in Dinorado (Table S10). Rice bugs were significantly less common in Dinorado in all intercropping systems compared to Dinorado monoculture, but were more common in Dinorado in the rice mixture compared to the monoculture (Table S10). Other insect pests observed in some sites were green leafhoppers (Nephotettix spp.), brown planthoppers (Nilaparvata lugens), and grasshoppers. The most commonly observed beneficial insects were spiders, wasps, red ants, and lady bugs, but these were observed too infrequently to allow statistical comparison of treatment effects.