Taiwan (TARI) Site

Specific Project Objectives & Deliverables


1.Rice crop mapping

  • The research findings of rice crop mapping obtained for 2015 from Sentinel-1A data have been archived, using the following processing steps:
    • Data pre-processing including radiometric calibration, speckle noise filtering, terrain correction and re-projection, image co-registration;
    • Image classification using the normalized difference backscattering index (NDSI) and object-based image analysis (OBIA); and
    • Accuracy assessment using rice crop map obtained from TARI in 2015.
  • The mapping results indicated the overall accuracy and Kappa coefficient achieved for VH data were 75.7% and 0.69, respectively, while those for VV data were 49.4% and 0.5, respectively (Figures 2 and 3).

Figure 2. Results of rice crop mapping using Sentinel-1A VH data: (a) classification map, and (b) ground reference data.


Figure 3. Results of rice crop mapping using Sentinel-1A VV data: (a) classification map, and (b) ground reference data.


2. Rice yield estimation

  • We estimated rice crop yield in the study site following three main steps of data processing:
    • Data pre-processing to construct input parameters, including weather data, soil data, crop genotype coefficients, crop management data, and rice crop map;
    • Crop yield estimation by assimilating MODIS LAI data into CERES-rice model using the particle swarm optimization (PSO) algorithm; and
    • Error verification using the government’s rice yield statistics.
  • The robustness of the yield simulation approach was evaluated by using the government’s rice yield statistics collected from 46 townships. The crop yields achieved from CERES-Rice model were spatially averaged for each township and compared with those from the government. The results indicated that the values of the root mean square error (RMSE) and the mean absolute error (MAE) were 17.3% and 12.7%, respectively, which were lower than 20%, indicating good agreement between the two datasets.
  • The spatial distributions of simulated rice yields indicated the yield variability within the township, and higher yields were more concentrated in the north and south parts of the study region. The lower yields were especially observed for rice fields located along the coastal zones and in areas where the terrain was complex (Figure 4).

Figure 4. Spatial distribution of simulated rice yields for the first crop in 2014.

  • To certain extent, our objectives have been met. However, we are still developing algorithms for operational purposes in respect to “best practice”.
©2015 Joint Experiment for Crop Assessment and Monitoring