National Aeronautics and Space Administration

Fig. 1. Global distribution of approximately 2,500 Terra/ASTER
reference scenes used for the validation of the MOD14 product
(Adapted from Giglio et al., 2016).

Stage 3 validation of the Terra MODIS Collection 6 active fire detection product has been achieved.

Approximately 2,500 ASTER scenes (60×60km each) were used to assess the performance of the MOD14 product globally. An equidistant grid (900km diameter cells) was used to provide an unbiased spatial sampling of fire pixels (Figure 1). Three temporal subsets of MODIS Terra data were represented in the validation, namely: 2001-2002, 2003-2004, and 2005-2006. A similar number of ASTER scenes were used for each temporal subset providing unbiased multi-year sampling of fire pixels. Approximately 16,000 daytime and 700 nighttime MOD14 fire pixels were sampled globally.

Omission errors for the Collection 6 MOD14 product is approximately 10% lower for large fires in areas with high percentage tree cover, compared to Collection 5 data (Figure 2, below left). Overall, omission errors tend to stabilize at 5% as fires grow larger in size. The Collection 6 MOD14 daytime commission errors (false alarms) were also reduced, compared to Collection 5. The improvement was particularly noticeable in tropical forests characterized by high percentage tree cover (Figure 3, below right). False alarms occurring in densely vegetated areas tend to coincide with recently burned plots identified by the presence of a discernable scar. Nighttime commission error was estimated to be negligible.

Fig. 2: Omission error rates calculated for MOD14
Collection 5 & 6 data as a function of fire size mapped
using coincident ASTER 30 m resolution reference
fire data. Results are stratified based on percentage
tree cover (TC) data derived from MODIS Vegetation
Continuous Fields (MOD44) product (Adapted from
Giglio et al., 2016).

Fig. 3 False alarm rates calculated for MOD14
Collection 5 & 6 data as a function of percentage
tree cover data derived from MODIS Vegetation
Continuous Fields (MOD44) product. Results are
stratified based on the presence/absence of
discernable burn scars within affected pixels.
(Adapted from Giglio et al., 2016).

Stage 1 validation has been achieved for the MODIS Fire Radiative Power (FRP) product

Validation of FRP data are challenging because FRP retrieval requires unsaturated radiances, which are generally not provided by medium resolution sensors (e.g., Landsat-class data) typically used as reference data. Opportunistic validation exercises to-date have used reference data from the experimental small satellite BIRD, or airborne thermal imaging systems (Dickinson et al., 2016; Schroeder et al. 2014). As a result of the limited data available, the MOD14 FRP data remains validated at Stage 1. The top-of-atmosphere FRP data uncertainty can be relatively high (˜50%) and will vary depending on the observation conditions (e.g., fire fractional area and pixel spatial response function effects, as well as atmospheric attenuation).