Friday, June 5, 2015

Improving predictive model for rice in arid environments

Projections of the likely impacts of climate change in the future are based on complex computer simulation models. A major emphasis in climate simulation modeling is determining the likely impact of climate change on agriculture.

The Africa Rice Center (AfricaRice) has long been involved with models that simulate crop development and predict planting dates to avoid the worst of seasonal stresses. In the mid- 1990s, AfricaRice researchers helped develop the model ORYZA_S by combining two existing models: ORYZA1, a rice crop model developed by Wageningen University and the International Rice Research Institute (IRRI) for irrigated conditions in Asia, and RIDEV, a much simpler model developed by AfricaRice simulating phenology of rice and sterility due to heat or cold stress under irrigated conditions in the Sahel.

An improved version, ORZYA2000, was released in 2001 by IRRI, followed by progressive revisions until 2009, but it was based on ORYZA1 and therefore fails to accurately simulate yield under Sahelian conditions.

“We chose to proceed with the 2009 version of ORYZA2000,” says Pepijn van Oort, crop modeler at AfricaRice, “because we hope that any improvement in the main model will also be useful under different conditions from those we tested, such as with water or nitrogen limitation, or in crop rotations. With ORYZA_S such applications were not possible.”

This meant that 20 years on, there was a need to take a fresh look at phenology and cold and heat stress in the Sahel. Developing new subroutines and other refinements to obtain a better predictive model for rice in the Sahel — in a changed climate with respect to the 1990s — became the new challenge for AfricaRice.

Computer-based models create simplified versions of reality and so should never be considered perfect. “Perfect prediction is suspect, may be caused by over-parameterization on a limited dataset, and runs a risk of adjusting parameter values without sound eco-physiological justification,” says van Oort. “We have tried to avoid this by using a large dataset, by making only modifications substantiated by solid experimental research, and by keeping calibration to a minimum.”

The large data set was obtained by Michiel de Vries, then AfricaRice irrigated-rice agronomist, from monthly sowings of variety IR64 at two sites in the Senegal River valley over 15 months in 2006–2007, a total of 29 treatments.

The modifications made, chosen on the basis of previous research, comprised: (i) so-called ‘cardinal’ temperatures for development; (ii) cardinal temperatures for early leaf growth; (iii) spikelet-formation process; and (iv) heat- and cold-induced sterility.

The model was specifically calibrated only for developmental characters. Moreover, to test the new heat- and cold-induced sterility subroutines, validation simulations were run to predict yield, first using observed development and number of spikelets, and second with simulated development and number of spikelets.

“The first thing we needed to adjust for IR64 grown in the Sahel was the cardinal temperatures,” says van Oort. “In particular, IR64 has a much higher base temperature than the default setting in ORYZA2000 (14°C cf. 8°C), a slightly higher optimum temperature (31°C cf. 30°C), and apparently experiences no delay in development at temperatures above the optimum (i.e. there is no maximum temperature, at least not under the conditions tested).” With these parameters corrected, the model gave improved simulation of rice development and therefore yield.

“We started with the situation in which ORYZA2000 over-estimated heat-induced sterility and underestimated cold-induced sterility,” says van Oort. “The new heat and cold subroutines give much better simulation of the two sterilities and, consequently, final yield.” The keys to improving heat-induced sterility simulation were transpirational cooling and flowering time, while the key to improved cold-induced sterility was using minimum daily temperature rather than average daily temperature.  

“These modifications are all logical if we think about where we’re working,” says van Oort. “ORYZA2000 was developed in and for Southeast Asia which, for the most part, is a humid tropical environment. In comparison, arid regions like the Sahel experience much lower humidity and much greater ranges in daily temperature.”

In a dry environment, the relative humidity (RH) is much lower than in a humid one. Thus, the ability of a plant to cool itself through transpiration is much greater in the arid zone (just like humans can sweat to cool themselves in a dry environment, while sweating in a humid environment just makes one wet!).

According to the subroutines developed by van Oort, at 35°C and 30% RH (typical of the Sahel), a plant can cool by 6°C relative to the air temperature via transpiration, while at 30°C and 90% RH (typical of humid tropical Southeast Asia), there is zero ability to cool via transpiration.

Flowering earlier in the morning means the rice plants are exposed to a lower temperature, which reduces the risk of heat sterility. In general, rice plants flower earlier during the day in hotter environments, but this characteristic is also genetically controlled and so varies with genotype. “Putting flowering time into the model now allows us to simulate how much yield gain can be obtained from breeding for earlier-flowering varieties,” says van Oort.

Arid environments also have much larger temperature differences during the day. “On one day in January, temperature increased from 8°C to 33°C. According to ORYZA2000, the cold-sterility risk was small, because average temperature was ‘safe’, but it was clear that the minimum of 8°C caused severe cold sterility. We therefore changed the subroutine to use minimum rather than average temperature.”

“Model calibration can be a tricky enterprise,” says van Oort. “At a certain point we found that the model was overestimating biomass production and therefore also yield. An effective trick to increase accuracy for yield was to modify the parameter that determines that number of spikelets formed per unit of biomass. But this led to unrealistic parameter values, because the real problem was that the model was overestimating total biomass. So we kept focused on the real causes of errors and played no artificial tricks with parameters.”

At the end of the day, van Oort and the team were able to modify the ORYZA2000 model to better predict IR64 rice development and yield in the arid Sahel of the Senegal River valley. Moreover, it did a better job of these predictions than the benchmark ORYZA_S that was developed for the same environment and optimized for IR64 in 1999.

“It is important to remember that this work was not done in isolation,” says van Oort. “It would not have been possible without the work done in the 1990s by Michael Dingkuhn (formerly with AfricaRice) and his co-workers in developing ORYZA_S and RIDEV.” In fact, the Sahel-adapted ORYZA2000 of van Oort and partners uses several equations and parameters derived from ORYZA_S and RIDEV.

“Our results indicate a need for further research into the components we identified, and to re-assess the climate risk to rice in arid regions,” concludes van Oort. “Our discoveries about the importance of cardinal temperatures, heat tolerance and heat avoidance also provide a basis for variety selection, as these three critical characteristics are genetically controlled and vary across cultivars.”

Related :