Period: 1982 – 2012

Sensor: AVHRR (1982 to 1999) & MODIS (2000 to 2012)

Frequency: 8 days

Resolutions: 5km

Extent: Global

Data can be accessed here.

Algorithm description: The algorithm theoretical approach is similar to Copernicus GEOV1 LAI product. It uses General Regression Neural Networks (GRNN) to retrieve LAI from time-series MODIS reflectance data. The network was trained (1 year period) on the fused MODIS and CYCLOPES LAI product over BELMANIP global sites. The CYCLOPES ‘effective LAI’ was first converted to ‘true LAI’ to match with MODIS LAI (equation 1 in Xiao et al (2014)). The GRNNs are trained with the fused time- series LAI from MODIS and CYCLOPES LAI products and reprocessed MODIS reflectance data.

Accuracy: Accuracy was estimated by comparing with ground data, MODIS LAI and CYCLOPES LAI. On average the GLASS LAI was estimated with a RMSE of 0.64 (R2=0.87).  MODIS and CYCLOPES LAI were estimated with an RMSE of 1.07 and 0.50 (R2 of 0.68 and 0.82), respectively.

Limitations:

  • For the biome type of evergreen broadleaf forest GLASS LAI values were systematically higher than the CYCLOPES LAI values: because the CYCLOPES algorithm does not include clumping at the plant and canopy scales
  • Peak position of the MODIS LAI frequency distribution was higher compared to GLASS: due to the overestimation of the MODIS LAI values for broadleaf forests.

Reference:

Xiao, Z. et al. Use of General Regression Neural Networks for Generating the GLASS Leaf Area Index Product From Time-Series MODIS Surface Reflectance. IEEE Trans. Geosci. Remote Sens.52, 209–223 (2014).