Abstract:

To provide guidance for high−quality rice cultivation management under future climate change, a quantitative model between the average temperature during the grain−filling period (T_ave) and rice quality traits was constructed and validated using an optimized Gaussian mixture model (GMM) and Mean−Shift clustering algorithm. In this study, a staged sowing experiment was conducted in Changsha county of Hunan province in 2022 using 53 different mid−and late−season indica rice varieties. Eight sowing dates were set, including May 1, May 11, May 21, May 31, June 21, July 1, July 11 and July 21. The results showed that grain filling temperature significantly affected rice quality. Among different sowing dates, the coefficients of variation (CVs) for chalky grain rate and chalkiness degree were 39.0% and 48.6%, respectively, reflecting the most pronounced response to temperature. The CVs for alkali spreading value, gel consistency, transparency, and head rice rate were 23.9%, 20.2%, 17.5% and 8.9%, respectively, indicating a moderate response. In contrast, the amylose content, milled rice rate and brown rice rate showed minimal sensitivity, with CVs of only 4.0%, 2.5% and 0.8%, respectively. Based on the GMM and Mean−Shift clustering algorithms, the relational models between grain−filling temperature and rice quality traits were established across the eight sowing dates. R² values of the fitting functions exceeded 0.900 for chalkiness degree, gel consistency and alkali spreading value, while surpassed 0.800 for brown rice rate, milled rice rate, head rice rate and chalky grain rate, and were 0.778 for transparency and 0.573 for amylose content. Validation results demonstrated that, except for amylose content, the test R² values for the other eight traits were all above 0.600, confirming satisfactory model fit. Thus, the proposed functional models accurately captured how each quality trait responds to changes in grain−filling temperature. With sowing dates delayed, the average grain−filling temperature ranged from 30.8℃ to 18.4℃. Processing quality traits, brown rice rate, milled rice rate and head rice rate, exhibited a trend of initial increase followed by decrease, with optimal T_ave of 22.0℃, 22.0℃and 25.3℃, respectively. Appearance quality traits, chalky grain rate, chalkiness degree and transparency, showed an opposite pattern, first decreasing and then increasing, with optimal T_ave of 25.7℃, 25.5℃and 25.2℃, respectively. Eating quality traits responded differently: the amylose content remained relatively stable, gel consistency decreased monotonically, and alkali spreading value increased monotonically. Both excessive high and low temperatures adversely affected processing and appearance quality. Higher temperatures favored a softer gel consistency but resulted in a higher gelatinization temperature, whereas lower temperatures favored a lower gelatinization temperature but resulted in a harder gel consistency. Comprehensively, a grain−filling T_ave of 25℃ yielded relatively optimal processing and appearance quality. The corresponding trait values were 80.7% for brown rice rate, 71.2% for milled rice rate, 55.1% for head rice rate, 12.7% for chalky grain rate, 3.1% for chalkiness degree, 1.7 for transparency, 17.9% for amylose content, 62.7mm for gel consistency and 5.2 for alkali spreading. These met the Grade 3 high−quality rice standards specified in the agricultural industry standard "Quality of Edible Rice Varieties" (NY/T 593−2021).

Key words: Temperature at filling stage, Rice quality, Clustering algorithm, Curve fitting