Spatial and temporal precipitation variability as a component of site-specific crop yield variability
Abstract
The purpose of this study was to determine spatial and temporal precipitation variability, and the effect of this variability on yield and profitability. On-farm precipitation data is currently being measured by site-specific farmers. One potential use of the data is to provide inputs for corn yield modeling, which has been performed with neural networks and simulation. Profitability of measuring on-farm data depends on spatial precipitation variability and its effect on yield. Precipitation and air temperature from corn silking to dent stages, scale of yield data, and a technology factor were used to model corn yield in east central Indiana at farm (250 ha), county, and state scales, using backpropagation neural networks with five data coding schemes. The best net gave a root-mean-squared error of 11.9% overall (10.9% farm, 10.5% county, 7.9% state yield), with maximum-value interval coding. Four rain gauges on the same farm, spaced apart 1.02 to 3.04 km, gave a median absolute deviation of precipitation among gauges, by corn and soybean phenological phase, of 0.25 to 1.73 mm·day−1 (spatial variability). Median absolute deviation from a reference year was 0.17 to 3.40 mm·day−1 (temporal/year-to-year variability). Spatial variability was less than temporal variability, and frequently less than 1 mm. Three precipitation data sources—a National Weather Service (NWS) station on the same farm, the nearest non-urban NWS station, and a weighted mean of three nearest non-urban NWS stations (27–35 km distance)—were used to simulate corn yield by 1-ha grid cells with CERES-Maize. The percent absolute difference of simulated yield among the three sources (effect of spatial precipitation variability) was 15.8%. The percent absolute difference from long-term mean (temporal variability) was 21.5%, of the same order as spatial variability. A choice among nitrogen application strategies—variable-rate versus whole-field application, starter versus no starter—was made for the same farm, using probability of profit from simulated corn yield based on the three precipitation data sources. The most profitable strategy was not dependent on the data source. Using data from the nearest NWS station was more profitable than measuring precipitation on-site.
Degree
Ph.D.
Advisors
Ess, Purdue University.
Subject Area
Agricultural engineering|Agriculture|Atmospheric sciences
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.