Search

• We determined frequency of environment types (ENVT) for US maize, soybean, and wheat. • Each ENVT corresponded to a combination of drought and heat stress during critical period. • Inter-annual variation in drought and heat stress was higher than spatial variation. • The ENVT scheme explained observed variation in average actual yield for the three crops. • ENVTs can be used to link field-year experiments with target production environments. The impact of agricultural technologies on crop yield is influenced by the environment type (ENVT) as determined by weather and soil. Understanding the correlation between the ENVT of the testing site in relation to the ENVT of the target production region is important for the evaluation and scaling out of agricultural technologies. Here we propose and apply the first explicit method to characterize ENVTs for major rainfed maize, soybean, and wheat producing regions in the United States. We combined a tested spatial framework, Technology Extrapolation Domain (TED), with crop modeling, long-term (30-y) daily weather records, and soil and management databases to calculate the frequency of ENVTs per crop for major harvested areas. Each ENVT was determined based on the intensity of drought and heat stress during key crop stages for yield determination. The ENVT repeatability was calculated based on the frequency of the most dominant ENVT in each TED. We found that inter-annual variation in drought and heat stress was larger than spatial variation. Our ENVTs explained 2x to 7x larger portion of the variance in actual yield compared to the existing TED framework that is based on long-term annual climate means and soil water storage. For maize and soybean, ca. 30% of their harvested area was located in TEDs with highly repeatable ENVTs (>66% of years). In contrast, only 15% of the wheat harvested area was located in TEDs with high ENVT repeatability. In comparison to the TED framework, the ENVTs defined here can help better capture G×E×M interactions and determine the environmental correlation between testing sites and target production environments. [ABSTRACT FROM AUTHOR]