Hydrothermal germination models: comparison of two data-fitting approaches with probit optimization

Authors: Hardegree, Stuart; Walters, Christina; Boehm, Alex; OLSOY, PETER - Boise State University; Clark, Pat; Pierson Jr, Frederick

Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/10/2015
Publication Date: 6/26/2015
Citation: Hardegree, S.P., Walters, C.T., Boehm, A.R., Olsoy, P.J., Clark, P., Pierson Jr, F.B. 2015. Hydrothermal germination models: comparison of two data-fitting approaches with probit optimization. Crop Science. 55(5):2276-2290. https://doi.org/10.2135/cropsci2014.10.0703.
DOI: https://doi.org/10.2135/cropsci2014.10.0703

Interpretive Summary: Hydrothermal germination response models are widely used to compare seedlots for relative ability to germinate under temperature and water stress. The predominant model that has been used in previous scientific studies and agricultural modeling applications requires adoption of a large number of assumptions that degrade the predictive accuracy of these models. We tested two additional models that provided significant improvement in model predictions and that are easier to develop and use. These models can be used to improve germination and emergence prediction of crop models, and to interpret and identify important germination strategies of wildland plant species. This could lead to new strategies for selection of appropriate plant materials for planting in different field environments, or selection of more effective field planting dates.

Technical Abstract: Probit models for estimating hydrothermal germination rate yield model parameters that have been associated with specific physiological processes. The desirability of linking germination response to seed physiology must be weighed against expectations of model fit and the relative accuracy of predicted germination response. Computationally efficient empirical models have been proposed that do not require a priori assumptions about model shape parameters, but the accuracy of these models has not been compared to the more common probit-optimization procedure. Thirteen seedlots, representing 6 native perennial rangeland grasses and an invasive annual weed, were germinated over the constant temperature range of 3 to 36 oC, and water potential range of 0 to -2.5 MPa. Hydrothermal germination models were generated using probit optimization, optimized-regression/equation discovery, and statistical gridding. These models were evaluated for the pattern and magnitude of residual model error, and the relative magnitude of predictive errors under field-simulated temperature and moisture conditions. Residual model errors in predictions of germination rate were greatest for the probit optimization procedure. Statistical gridding and optimized regression produced lower predictive model error but the latter procedure could not resolve germination response of slower-germinating seed populations. Choice of hydrothermal germination model should be dependent on study or management objectives. Probit optimization produced model parameters that may reflect the physiological processes underlying rate response, but this procedure produced significant and consistent patterns of model error across all seedlots tested. The more computationally efficient and accurate regression and statistical gridding procedures may be desirable for identifying germination strategies and syndromes that are based on predicted response to simulated conditions of field temperature and moisture.