Evaluation of MODIS surface reflectance products for wheat leaf area index (LAI) retrieval

Document Type


Publication Title

ISPRS Journal of Photogrammetry & Remote Sensing

Publication Date






First Page


Last Page



The accuracy of leaf area index (LAI) retrieval depends critically on the quality of the input reflectance. MODIS Collection 4 (C4) and Collection 5 (C5) land surface reflectance data are used for wheat LAI retrieval. Results are compared with in situ measurements. The uncertainty in the reflectance data of the two collections (C4 and C5) from both Terra and Aqua sensors is analyzed and its influence on LAI retrieval is discussed. The discrepancies of blue and near infrared reflectances between Terra and Aqua in the C5 data are less than the discrepancies between the sensors in the C4 data. For both Terra and Aqua, the C5 data have much lower blue reflectance than do the C4 data. This can be attributed to improvements in the atmospheric correction algorithm for the C5 data including cloud mask definition and aerosol retrieval. Using both empirical vegetation indices and inversion methods, the LAI is derived from the C4 and C5 surface reflectances. For daily C4 data, only Aqua Normalized difference water indices (NDWI) have significant correlations with the LAI (at a 99% confidence level); in contrast, for the daily C5 data, all the vegetation indices have significant correlations with the LAI. A three-layer neural network is used to invert a one-dimensional (1-D) radiative transfer model for LAI estimation. For the daily C4 data, the correlation between the modeled and measured LAIs is poor and the root mean square error (RMSE) is larger than 1.1; in comparison, the RMSE for the daily C5 data is 0.7. For both C4 and C5 collections, the LAI tends to be overestimated when the sensor is operated with a large view zenith angle in the backscattering direction. The error is either due to the mismatch between the measured reflectance and the modeled reflectance from the simple 1-D radiative transfer model in this direction or due to the assumption of a Lambertian surface in the MODIS atmospheric correction. Additionally, for both methods the results from the 8-day composite C4 data are much better than the results from daily C4 data because there is less cloud and aerosol contamination after compositing. In summary, the daily C4 reflectance has greater uncertainty than does the daily C5 reflectance. Daily C5 data are more preferable for LAI retrieval; if only C4 data are available, more reliable results can be achieved using the 8-day composite data if its temporal resolution is not a concern for dynamic growth monitoring.


Leaf area index; MODIS; Model inversion; Vegetation index




© 2008 Elsevier

This document is currently not available here.