Statistical Models for Rice Yield Estimation from Remotely Sensed Data in Taiwan
Abstract
Rice is the most important food crop in Taiwan. Estimating rice production has been conducted yearly because policymakers need such information to devise crop management strategies to ensure domestic food consumption and rice grain exports. This study aims to develop methods for rice yield prediction in Taiwan using Moderate Resolution Imaging Spectroradiometer (MODIS) time-series data. We processed the data for the period from 2007 to 2015, following three main steps: (1) Data pre-processing to construct the smooth Normalized Difference Vegetation Index (NDVI) and land surface temperature (LST) data. (2) Establishment of yield models using the regression techniques with breakpoint. The rice yield was empirically modeled using the filtered NDVI, reconstructed LST, and the township yield statistics during the growing seasons. The crop yield was considered as the dependent variable determined by one or two independent variables (NDVI and LST). The data of the period 2007 to 2015 were used for establishment of yield models, leaving the 2016 data for error verification of the predicted results. (3) Error evaluation. To validate the model robustness, we used the established models to predict rice yields for crops in 2016. The statistical yields compared with the predicted yield results indicated that the median relative errors achieved for the first and second crops were -1.9% and 15.5%, respectively. The larger prediction error was observed for the second crop because the satellite data captured in this season were often contaminated by cloud cover and typhoons. The occurrence of heavy rains could decline potential crop yields and affect the vegetation greenness, influencing the spectral quality of satellite data, and thus increasing prediction errors. Eventually, although there are several model uncertainties, attributed to input parameters, our study demonstrates the potential application of time-series MODIS satellite for national prediction of rice crop yields in Taiwan.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMGC23H1454C
- Keywords:
-
- 0402 Agricultural systems;
- BIOGEOSCIENCES;
- 1632 Land cover change;
- GLOBAL CHANGE;
- 1640 Remote sensing;
- GLOBAL CHANGE;
- 6309 Decision making under uncertainty;
- POLICY SCIENCES & PUBLIC ISSUES