CERES-Rice作物模擬軟體之作物生長參數的敏感度分析與產量區間預測

Abstract

全球氣候變遷造成極端氣候發生頻度增加也影響全球糧食供應，世界第二大糧食作物、臺灣主要的糧食作物-水稻，也常作為氣候變遷研究之目標作物。最初用於預測作物生長發育的作物模擬軟體也因此越來越受到重視，在臺灣作物模擬軟體的研究及應用並不普遍，本研究欲利用作物模擬軟體預測水稻之產量，水稻研究中DSSAT CERES-Rice為最廣泛應用的作物模擬軟體之一。不過作物模擬軟體若輸入相同的設定及生長參數將會得到相同的模擬結果，如此無法反映出自然界中的個體變異。因此本研究目標利用區間估計代替作物模擬軟體的點估計以產生產量預測區間，首先利用Log Multivariate Normal與Uniform分布產生模擬資料，以3k 複因子混雜設計、複回歸分析與逐步變數篩選等方法篩選P1、P2R、P5、P2O、G1、G2、G3及G4等8個生長參數皆為重要的生長參數後，再以Log Multivariate Normal分布產生臺灣水稻品種-臺農67號與臺中秈10號之品種模擬資料得其模擬產量之10% 與90% 百分位數作為兩品種之產量區間分別為 (5360.4, 7049.6) Kg･ha-1及 (4995.4, 6881.6) Kg･ha-1，對照行政院農業委員會農業試驗所之TRIS台灣稻作資訊系統各育成品種之產量試驗資料，產量預測區間高估臺農67號之產量，低估臺中秈10號之產量。經本研究驗證CERES-Rice再經改善可做為臺灣水稻產量預測的工具之一，為改善作物模擬軟體於臺灣稉稻品種之模擬結果，有待後續參數化、模擬前的校正等以提升可信度與準確性，經驗證之模擬軟體也能推廣應用於臺灣農業，如政策評估、栽培管理、提供育種方向等多元的用途，促進臺灣農業更有效率的發展。

Climate change caused that increasing extreme climate occurred and resulted in food and water shortages all over the world. Rice, with the second-highest worldwide production and as one of the most important crops in Taiwan, was usually concerned as the target crop in crop models. In addition, the indica was the main cultivating and commercial variety instead of the japonica which was widely cultivated in Taiwan. Crop models were initially used to predict crop growth and development, but researches and applications in Taiwan were still scarce. DSSAT CERES-Rice was one of the widely used crop model in rice research. But there were some difficulties to create individual variations in the simulation because the same inputs would return the same outputs by crop model simulations. In this study, we aimed to use interval estimation to substitute point estimation to predict the yield intervals of two rice varieties in Taiwan, TNG67 and TCS10. First, we created simulation data from Log Multivariate Normal distribution and Uniform distribution then selected P1, P2R, P5, P2O, G1, G2, G3 and G4 as the effective genetic coefficients by 3k factorial confounding design, regression analysis and stepwise regression analysis. Second, we simulated simulation data of TNG67 and TCS10 from Log Multivariate Normal distribution then estimated 10% to 90% percentiles as the yield intervals, (5360.4, 7049.6) and (4995.4, 6881.6) respectively. Comparing to the yields of trials, from Rice Registered Varieties Database, Taiwan Agricultural Research Institute Council of Agriculture, Executive Yuan, the yields of TNG67 were overestimated and the yields of TCS10 were underestimated. CERES-Rice is able to be implemented to predict the yields of Taiwanese rice varieties To improve the simulation outputs, parameterizing varieties in Taiwan and calibration before simulation would be essential and efficient to promote the applications of crop models to agriculture in Taiwan, such as policy assessment, management decision, plant breeding and so on.