Transactions of the ASAE Vol. 46(5): 1365–1374 E 2003 American Society of Agricultural Engineers ISSN 0001–2351 1365 SIMULATION OF WITHIN–FIELD YIELD V ARIABILITY IN A FOUR–CROP ROTATION FIELD USING SSURGO SOIL–UNIT DEFINITIONS AND THE EPIC MODEL J. F. Perez–Quezada, J. Cavero, J. Williams, A. Roel, R. E. Plant ABSTRACT. Soil data were collected from a 30 ha commercial field using a 60 m sampling grid. Monitored yield data were also collected in this field between 1996 and 1999, when it had a wheat–processing tomato–bean–sunflower crop rotation. A comparison between SSURGO–NRCS soil–unit definition and field–measured soil data showed that in this field the former are a good approximation and starting point for precision agriculture studies and management. In a second test, the EPIC model, using the SSURGO database soil type definitions, was found to reproduce the yield variability within this field with reasonable accuracy. The model’s performance was not as good when tested with data from soil samples, apparently due to the way EPIC simulates water holding capacity from texture information and the lack of some key variables (not sampled), such as water content at field–capacity (FC), wilting–point (WP), and soil saturated conductivity. A set of runs was performed to simulate the yield at 13 point–locations in the field using FC, WP, and bulk density. Although the accuracy of the simulation did not improve greatly, the model reproduced the yield trend of two of the crops (wheat and sunflower). Keywords. EPIC model, Management zones, Precision agriculture, Spatial variability, SSURGO data, Temporal variability. omputer modeling can be used to address many questions that would need a great expenditure of resources to test experimentally. The Erosion– Productivity Impact Calculator (EPIC) cropping systems model (Williams et al., 1984) simulates conditions of weather, irrigation, fertilization, tillage, and management at the field level. EPIC has been tested for the study of complex crop rotations in southern France (Cabelguenne et al., 1990), simulating growth and yield of corn, grain sorghum, sunflower, soybean, and wheat. After calibrating and validating the model with two years of data, Cabelguenne et al. (1990) concluded that EPIC was able to simulate complex rotations with acceptable accuracy. Although this was not a replicated experiment, it attempted to measure the accuracy of the model’s predictions by using 28 pairs of plots that were coincident in terms of year of harvest, crop, preceding crop, and input level. Among these comparisons between field research plots, 85% had yields within 20% of each other. When computer simulations were compared with measurements of yield, 81% of the simulated yields were Article was submitted for review in December 2002; approved for publication by the Soil & Water Division of ASAE in June 2003. The authors are Jorge F. Perez–Quezada, Graduate Student, Graduate Group in Ecology, University of California, Davis, California; José Cavero, Tenured Scientist, Dep. Genética y Producción Vegetal, Estacion Experimental de Aula Dei (CSIC), Zaragoza, Spain; Jimmy Williams, Research Scientist, Texas Agricultural Experiment Station, Grassland Soil and Water Research Laboratory, Temple, Texas; Alvaro Roel, Graduate Student, Graduate Group in Ecology, University of California, Davis, California; and Richard E. Plant, ASAE Member Engineer, Professor, Departments of Agronomy and Range Science and Biological and Agricultural Engineering, University of California, Davis, California. Corresponding author: Richard E. Plant, Dept. of Agronomy and Range Science, University of California, 1 Shields Avenue, Davis, CA 95616; phone 530–752–1705; e–mail: replant@ucdavis.edu. within 20% of the observed yields (76 plot–years were analyzed). They concluded that EPIC was almost as good a predictor of plot yield as the yield of a similar (paired) plot in the same experiment. Bryant et al. (1992) used EPIC to measure yield response of corn to changes in irrigation timing, finding that it was not only able to simulate the effect of total amount of water but also the effects of the distribution of irrigation events. Their results after simulating three years of data showed coeffi- cients of determination between the observed and simulated yields that ranged from 0.72 to 0.86. The authors were able to improve the highest value to 0.91 by changing a parameter that simulated the effect of a hailstorm. This indicates that the EPIC model is very versatile but also that many years of data may be needed to get a calibration that adequately accounts for the variability caused by climatic conditions from year to year. EPIC has not been considered as necessarily providing an accurate simulation of a particular crop in a given field and a given year (Steduto et al., 1995). Working under research station conditions, Cavero et al. (2001) sampled intensively a 27 × 27 m field, measuring bulk density, infiltration rate, and soil texture at 91 points. Soil water retention at wilting point, field capacity, and soil depth were measured at 100 points. The advance and recession time of an irrigation front (which when combined define the irrigation opportunity time) and soil surface elevation were measured at 361 points. Corn yield measure- ments were made in 73 1.5 × 1.5 m areas. Cavero et al. (2001) found that when they used the estimated values of irrigation depth at each location as input to the EPICphase crop model, the best correlation with the measured yield variability occurred when they used only the spatial variability of infiltration rate (leaving opportunity time fixed), obtaining a coefficient of determination of 0.51. EPICphase is a modifi- cation of EPIC, especially improved for water and N stress C