Comparing and combining process-based crop models and statistical models with some implications for climate change

Abstract

LETTER • OPEN ACCESSComparing and combining process-based crop models and statistical models with some implications for climate change AbstractWe compare predictions of a simple process-based crop model (Soltani and Sinclair 2012), a simple statistical model (Schlenker and Roberts 2009), and a combination of both models to actual maize yields on a large, representative sample of farmer-managed fields in the Corn Belt region of the United States. After statistical post-model calibration, the process model (Simple Simulation Model, or SSM) predicts actual outcomes slightly better than the statistical model, but the combined model performs significantly better than either model. The SSM, statistical model and combined model all show similar relationships with precipitation, while the SSM better accounts for temporal patterns of precipitation, vapor pressure deficit and solar radiation. The statistical and combined models show a more negative impact associated with extreme heat for which the process model does not account. Due to the extreme heat effect, predicted impacts under uniform climate change scenarios are considerably more severe for the statistical and combined models than for the process-based model.

A deep learning approach to conflating heterogeneous geospatial data for corn yield estimation: A case study of the US Corn Belt at the county level

“…This inconsistent WDRVI changes implied unrecoverable damage caused by extreme events occurring in early stages, such as pollination failure (Cicchino et al, 2010). In this context, linear statistical modeling may provide biased yield estimation because of the nonlinearity among input features, whereas biophysical simulation models can simulate the cumulative growth process deterministically but suffer its generality for large-scale estimations (Roberts, Braun, Sinclair, Lobell, & Schlenker, 2017 recognized by large-scale analysis and field traits (Edreira & Otegui, 2012;Liu et al, 2016;Lobell et al, 2014). For example, southern and central Corn Belt are more sensitive to high temperature and suffer more yield losses due to increased temperature (Butler & Huybers, 2013).…”

Integrating genotype and weather variables for soybean yield prediction using deep learning

“…Traditionally, crop growth models have been proposed to simulate and predict crop production in different scenarios including climate, genotype, soil and management factors [10]. These provide reasonable explanation on biophysical mechanisms and responses, however, these models have deficiencies related to input parameter estimation and prediction in complex and unforeseen circumstances [11]. Previous attempts at yield prediction across environments JOHNATHON SHOOK ET AL.…”

Is another genetic revolution needed to offset climate change impacts for US maize yields?

“…Without substantial gains in productivity, the rising global demand for food could lead to higher food prices thereby incentivizing conversion of rainforests, wetlands, and grasslands to farmland (Alston et al 2009, Duvick andCassman 1999). There has been much work estimating the potential impact of climate change on maize yields using historical data coupled with statistical models (Lobell and Asner 2003, Schlenker and Roberts 2009, Lobell et al 2011, Butler and Huybers 2013, Burke and Emerick 2016, Gammans et al 2017, and recent research suggests that these statistical-based approaches provide similar estimates to process-based models (Roberts et al 2017). A key empirical challenge for statistical models is unpacking the effect of weather on crop yields from that of technological differences across both locations and time.…”