Satellite remote sensing of active fires provide an important tool for monitoring and locating wild fires.At present

more and more satellites and sensors have been used to monitor the fires.The Moderate Resolution Imaging Spec- troradiometer（MODIS）sensor

on board the Terra and Aqua satellites in Earth Observing System（EOS）of National Aeronautics and Space Administration（NASA）

have offered an improved combination of spectral

temporal and spatial resolution for global fire detection compared with previous sensors

and have become the major data source as the substi- tute for the Advanced Very High Resolution Radiometer（AVHRR）sensor

widely used for fire detection.MODIS global active fire detection algorithm

proposed by the MODIS fire team from NASA

and based on original MODIS detection algorithm and heritage algorithms which is developed for the AVHRR and the Visible and Infrared Scanner（VIRS）

was designed for global active fire products.But the MODIS fire algorithm is imperfect for fire detection in China.There are two reasons which could cause mistakes in fire detection:one is the threshold of mid-infrared channel used to exclude the false fire pixels is so big that many fires with lower brightness temperature are eliminated

the other is some fires with higher brightness temperature

which could be identified easily with human eyes in 4μm band

are omitted for the errors of potential fire pixels during the selection of background contextual pixels.This paper introduces an improved algorithm of self-adaptive fire detection for MODIS data based on MODIS global fire detection algorithm

which makes use of the bright- ness temperature derived from the MODIS 4μm and 6μm channels to carry out fire detection

and the improved algorithm is described in detail.The active fire detection strategy is based on absolute detection of the fire.The latter test identifies pixels with values elevated above a background thermal emission obtained from the surrounding contextual pixels.This method accounts for variability of the surface temperature and reflection by sunlight.Since the improved self-adaptive algo- rithm can acquire the threshold of mid-infrared for exclusion of the false fire pixels by histogram statistics

and recognize the potential fire pixels by relative temperature augment in background pixels

it enhances the ability of active fire detection.As for higher brightness temperature the thermal emission of surface active fires could penetrate through the smooth and thin cloud and reflect on combined color satellite image

which could be easily interpreted as fires by human eyes

this paper introduces a method to discern the active fires under the cloudy condition by the satellite image texture characteristics in visible spectrum and brightness temperature variance in mid-infrared spectrum.Finally

Guangdong region is selected as the experiment region to verify the improved fire detection algorithm

and the results of fires detection are compared between improved self-adaptive algorithm and MODIS global algorithm.The analysis show that this improved self-adaptive algorithm has a perfect discernment for active fires

which could increase sensitivity to smaller

cooler fires

and discernment for active fires in higher brightness temperature omitted in MODIS global algorithm.Also

the improved algorithm has a good discernment for the active fires under the cloudy condition.