Rice Population Simulation Model

Rice (Oryza sativa L.) growth and development is greatly affected by physical and agronomic factors. The complex interaction between rice plants and those factors makes it difficult to clearly determine which combination of physical conditions and inputs promote optimal cultivar performance using traditional methods. Physiologically-based crop simulation models have provided a rigorous structure which has proven to be useful for understanding and quantifying the impact of biotic and physical factors on crop growth and development.

A number of rice simulation models have been developed (van Keulen, 1978; Angus & Zandstra, 1980; McMennamy & O'Toole, 1983; Huang & Wang, 1986; Ritchie et al., 1987; Costello et al., 1988; Zhang & Chen, 1990; Graf et al., 1990b; Miller et al., 1993; Salam et al., 1994; Singh & Padilla, 1995).

These models represent a range of approaches to analyzing the response of rice to physical, agronomic, and biotic factors. However, those models have not been verified or validated thoroughly using a statistically rigorous procedure. As a result, those models have only limited applications.

The objectives of this research were to develop a statistically rigorous model parameterization, verification, and validation procedure, and to verify and validate RICEPSM following the procedure using successive sets of data. Eleven data sets, representing two cultivars in 1992, and three cultivars each at three plant densities in 1993, were collected from field experiments conducted at the Texas A&M University Systems Agricultural Research & Extension Center at Beaumont, TX. Each data set contains 13 data types (total dry mass, number of tillers, number of nodes per main stem, and mass of root, stem, leaf blade, panicle vegetative component, and grain, for the main plant and collectively for the tillers).

The proposed criterion for model parameterization, verification, and validation is the coefficient of multiple determination (R²), which measures the proportion of variability for single and multiple data types that was explained by the simulation model.

where: SSTO_j is the observed variation for the jth data type corrected for the mean, SSE_j is the amount of variability for the jth data type not explained by the model, and S_{_j}^² is the variance of the jth data type. Goodness of fit was evaluated across sampling dates for 13 data types. A critical value (R_{_c}^² = 0.85) was used to test overall goodness of fit. When R² was greater than the R_{_c}^², the model was considered verified with respect to its fit to a data set, if compared with a data set used to parameterize the model; or validated, if compared with an independent data set. After the model was verified or validated against a data set, the previously unused data set was used to evaluate the goodness of fit. If the model failed to adequately fit a data set, the parameterization and analysis cycle was repeated until the model was verified or validated against all data sets. Once a data set has been used for parameterization or verification, it could be used to re-parameterize the model but was no longer used for validation.

RICEPSM requires input parameters associated with a specific cultivar, as well as data on the management of the crop, the weather, and the soil. Management data consist of the latitude of the site, plant density at emergence, amount and time of fertilizer applications. Daily weather data include maximum and minimum temperature, and solar radiation, from historic climate records for a site or through a weather data generation model. Soil data include residual fertilizer content and organic matter content of the soil at planting. Rice plant population processes are simulated throughout a growing season on a per unit area basis. The time step for most physiological processes such as birth, growth, aging, and death of plant organs is 40 Degree-days (>10⁰C). Photosynthesis and respiration, which have nonlinear relationship with temperature, however, were simulated hourly. The rice population was categorized into 12 tiller age-cohorts. For each time step, plant carbohydrate and nitrogen balances were simulated for each age class of each structure type for each tiller cohort, and were used in simulating age and structure specific birth and death rates. A soil nitrogen balance was estimated in 10 cm depth increments taking into account mineralization, nitrification, denitrification, clay fixation, volatilization, and plant uptake. A distributed-maturation algorithm was used to describe variability in growth and aging of different structures. Each time step, the model updated numbers and mass for each age-class of root, culm, leaf sheath, leaf blade, panicle vegetative component and grain for each tiller age-cohort. The default outputs from RICEPSM include metabolic supply demand ratios for carbohydrate and nitrogen, total dry mass, number of tillers, number of nodes per main stem, nonstructural carbohydrate, nitrogen, and total mass of root, stem, leaf blade, panicle vegetative component, and grain for main stem and collectively for tillers for each time step interval.

Lemont Rice

The ability of RICEPSM to accurately simulate rice growth and development has now been verified and validated by evaluating the goodness of fit of the simulated results to the 13 data types of each data set for three cultivars, Gulfmont, Rosemont, and Teqing, and for a wide range of plant densities.

A major advantage of a well validated physiologically-based model is that it can help in analyzing the relative importance of the various factors, both genetic and environmental. Hence, RICEPSM can be used to select optimal timing of agronomic practices (such as planting date, fertilization, and pesticide application), test proposed management strategies, examine the effects of different genetic traits, and to quantify the effect of physical and biotic factors on rice growth.