Correlation between temperature and phenology prediction error in rice (Oryza sativa L.)
van Oort, PAJ and Zhang, T and de Vries, ME and Heinemann, AB and Meinke, H, Correlation between temperature and phenology prediction error in rice (Oryza sativa L.), Agricultural and Forest Meteorology: An International Journal, 151, (12) pp. 1545-1555. ISSN 0168-1923 (2011) [Refereed Article]
For rice (Oryza sativa L.), simulation models like ORYZA2000 and CERES-Rice have been used to explore adaptation options to climate change and weather-related stresses (drought, heat). Output of these models is very sensitive to accurate modelling of crop development, i.e. phenology. What has to date received little attention in phenology calibration is the temperature range within which phenological models are accurate. Particularly the possible correlation between temperature and phenology prediction error has received little attention, although there are indications that such correlation exists, in particular in the study by Zhang et al. (2008). The implication of such correlation is that a phenology model that is accurate within the calibration temperature range can be less accurate at higher temperatures where it can
systematically overestimate or underestimate the duration of the phase from emergence to flowering. We have developed a new rice phenology calibration program that is consistent with ORYZA2000 concepts and coding. The existing calibration program DRATES of ORYZA2000 requires an assumption of default cardinal temperatures (8, 30 and 42 °C) and then calculates cultivar specific temperature sums and development rates. Our new program estimates all phenological parameters simultaneously, including the cardinal temperatures. Applied to nine large datasets from around the world we show that the use of default cardinal temperatures can lead to correlation between temperature and phenology prediction error and temperature and RMSE values in the order of 4–18 days for the period from emergence to flowering. Our new program avoids such correlation and reduces phenology prediction errors to 3–7 days (RMSE). Our results show that the often made assumption of a rapid decrease in development rate above the optimal temperature can lead to poorer predictions and systematic errors. We therefore caution against using default phenological parameters for studies where temperatures may fall outside the range for which the phenological models have been calibrated. In particular, this applies to climate change studies, were this could lead to highly erroneous conclusions. More phenological research with average growing season temperatures above the optimum, in the range of 32–40 °C, is needed to establish which phenological model best describes phenology in this temperature range.