January 2016
·
154 Reads
·
1 Citation
Despite the well-known advantages of using remote sensing data for burned area mapping, there are still drawbacks that should be circumvented. The presence of noise caused by cloud shadows, for instance, reduces the accuracy of burned area maps derived from remotely sensed images. The operational technique used to minimize this drawback is called multitemporal compositing, which consists of simply selecting the pixel that better represents the target of interest within a time interval. The aim of the present study was to evaluate multitemporal compositing techniques using Near Infrared (NIR) and Shortwave infrared (SWIR) channels from PROBA-V S1TOC 333m images. The images span September and October 2014 and cover the Brazilian territory (tile X13Y09). Four different multitemporal compositing approaches were analyzed, namely: the first, second and third lowest reflectance values from the NIR channel (NIRmin1, NIRmin2 and NIRmin3) and the largest value of SWIR within these three values (SWIR/NIR). The evaluation of the four composites approaches was conducted using the following criteria: (1) Presence of clouds and shadows; (2) Spectral separability to discriminate burned area; (3) Distribution of zenith angles. The results show that the compositing techniques SWIR/NIR, NIRmin3, and NIRmin2 considerably reduced noise caused by cloud shadows. Concerning the spectral separability to discriminate burned areas from different surface types, the NIRmin2 composite presented better results. Regarding the distribution of zenith angles, SWIR/NIR composite was closer to a Gaussian, while the other composites showed a high frequency between 50 and 60 degrees. Although compositing technique SWIR/NIR showed better results for the absence of clouds/shadows and distribution of zenith angles, the compositing technique NIRmin2 showed a greater capacity for discriminating burned areas. Since the compositing technique NIRmin2 presented lower computational cost, it may be a feasible option for burned area mapping using multitemporal composites of PROBA-V images. However, it is recommended that further studies covering a longer period of time should be conducted for comparison with the present work.