The impact of potential errors in rainfall observation on the simulation of crop growth, development and yield

Precipitation is the source for almost all soil moisture available for plant extraction by crops that are grown in rainfed cropping systems. In many cases precipitation is measured using a tipping-bucket rain gauge (TBRG) or similar device. Most automated weather station (AWS) networks employ TBRG to observe rainfall, due to the need of automating the process of collecting rainfall data. However, it is known that these gages have some type of error due to aerodynamic effects, wetting losses and the actual design and operation of the tipping-bucket sensor. The data collected by AWS are frequently used as input for complex computer simulation models that predict crop yield as a function of weather and soil conditions and crop management scenarios. The objective of this study was to evaluate the impact of potential errors in rainfall observations on simulated growth, development and yield. The generic grain legume model cropgro and the generic cereal model ceres were used to simulate growth and development for soybean, peanut, wheat and maize under different precipitation scenarios. These crop models use daily weather data as inputs. In this study, 36 years of daily historical records from Tifton, Georgia were obtained and randomly modified with relative errors when the values for daily precipitation were greater than zero to emulate the inaccuracy of TBRG observations. Two approaches were considered: (a) rain gauges randomly underestimate (negative bias) rainfall and (b) rain gauges randomly overestimate (positive bias) rainfall. To account for the random variability of precipitation data and to avoid a trend that could affect crop growth and development, the random modifications were replicated 32 times for each individual weather year and for each inaccuracy case that was considered. In this study, the modifications in daily precipitation amounts did not impact crop phenology, but resulted in substantial changes in both the mean and the variability of simulated yield, biomass, evapotranspiration, and drainage. The underestimation, e.g. negative bias, of rainfall measurements had a larger impact on the simulated variables than an overestimation. This study showed that the accuracy of rainfall observations is critical for the simulation of yield and that the variability of the simulated outputs is directly correlated to the accuracy of model inputs. It also demonstrated that complex soil–plant–atmosphere models are sensitive to variation in precipitation and possibly other environmental inputs.