| 摘要: | 稻米在臺灣為重要且廣泛食用的糧食作物，臺灣的糧食綜合自給率約在三成上下，而其中稻米自給率在九成以上，因此掌握當期稻米產量對於政府是重要的議題。遙測影像能提供多時序、大範圍的資料，可用以長期觀測稻作生長狀況，本研究使用2000年至2018年的MODIS (Moderate Resolution Imaging Spectroradiometer, MODIS)影像資料與臺灣稻米統計資料，建立稻米產量估算模型並比較估算成果與統計資料的差異。本研究分為三個步驟：第一步驟，使用MODIS影像計算並建立常態化差異植生指數(Normalized Difference Vegetation Index, NDVI)、地表溫度(Land Surface Temperature, LST)，設定稻米生長期之時期與稻米分佈位置，蒐集稻米歷年產量統計資料。第二步驟，使用2000年至2015年資料，包含NDVI、LST、臺灣各鄉鎮歷年產量資料，建立稻米產量估算模型。第三步驟，使用2016年至2018年之NDVI、LST資料輸入至稻米產量估算模型，計算出2016年至2018年稻米產量估算結果，比較與探討估算成果與統計資料間之差異。研究成果顯示，2016年至2018年第一期稻作均方根誤差(Root Mean Squared Error, RMSE)分別為792 (2016)、717 (2017)、1385 (2018)公斤/公頃，均方根百分誤差(Root Mean Square Percentage Errors, RMSPE)為11.9% (2016)、10.5% (2017)、17.3% (2018)，第二期稻作為1186 (2016)、930 (2017)、1308 (2018)公斤/公頃與36.0% (2016)、19.7% (2017)、24.5% (2018)。結果顯示在一期稻作中有較佳的估算成果，二期稻作穩定度較差。;Rice is an important and widely edible food crop in Taiwan.Grain supplies in Taiwan consists of approximately 30% self-efficiancy rate. Among it, 90% comes from rice.Therefore, understanding current rice yield is an important issue for the government. By using satellite images, multiple time series and large-scale data can be acquired to observe long-term rice growth. This study used Moderate Resolution Imaging Spectroradiometer (MODIS) data acquired from 2000 to 2018 and rice statistics in Taiwan to establish a rice yield model, comparing the differences between the estimation results and the existing statistics. This research is divided into three steps: (1) Using MODIS images to calculate Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST). Setting the rice growth period and rice distribution location, collecting statistics on historical rice yield. (2) Using data from 2000 to 2015, including NDVI, LST, and township-based yield data to establish a rice yield model. (3) Calculating NDVI and LST data from 2016 to 2018 in the rice yield model to obtain the rice yield estimation results to find out the difference between the estimation results and existing statistics. The results show that Root Mean Squared Error (RMSE) of first crops from 2016 to 2018 are 792 (2016), 717 (2017), and 1385 (2018) kg/ha. Root Mean Square Percentage Errors (RMSPE) are 11.9% (2016), 10.5% (2017), and 17.3% (2018). RMSE of second crops are 1186 (2016), 930 (2017), 1308 (2018) kg /ha, and RMSPE are 36.0% (2016), 19.7% (2017), 24.5% (2018). To conclude the above findings, first crops perform a relatively better estimation results than the second crops. |
