CLOUD DETECTION BASED ON DECISION TREE 
    
OVER TIBETAN PLATEAU WITH MODIS DATA 
Lina Xu*® * Shenghui Fang*, Ruiging Niu ^ Jiong Li? 
2 School of Remote Sensing and Information Engineering , Wuhan University, Wuhan 430079,China, silvaxu@sina.com 
? China University of Geosciences, Wuhan 430074, China 
Commission VIII, WG VIII/10 
KEY WORDS: cloud detection, MODIS, Tibetan Plateau, snow cover, decision tree 
ABSTRACT: 
Snow cover area is a very critical parameter for hydrologic cycle of the Earth. Furthermore, it will be a key factor for the effect of the 
climate change. An unbelievable situation in mapping snow cover is the existence of clouds. Clouds can easily be found in any 
image from satellite, because clouds are bright and white in the visible wavelengths. But it is not the case when there is snow or ice 
in the background. It is similar spectral appearance of snow and clouds. Many cloud decision methods are built on decision trees. 
The decision trees were designed based on empirical studies and simulations. In this paper a classification trees were used to build 
the decision tree. And then with a great deal repeating scenes coming from the same area the cloud pixel can be replaced by “its” real 
surface types, such as snow pixel or vegetation or water. The effect of the cloud can be distinguished in the short wave infrared. The 
results show that most cloud coverage being removed. A validation was carried out for all subsequent steps. It led to the removal of 
all remaining cloud cover. The results show that the decision tree method performed satisfied. 
1. INTRODUCTION 
In high altitude regions one of the important water sources are 
snow. So snow cover is very important in high mountainous 
areas where snow pack can often remain through the summer 
months and snow melt provides runoff and water supply for the 
downstream population. Many areas of the world have large 
gaps between observation locations or have no observation 
stations at all. Satellites have enabled researchers to obtain 
snow information on a global scale and monitor its effects on 
global climate (Rango, 1996). The Tibetan Plateau is a unique 
geomorphic unit and is called “the third pole” of the Earth by its 
highest altitude. So it is the most sensitive area in the world to 
hydrological cycle and climatic change. Mapping the snow 
cover area of the Tibetan Plateau is very important for the 
regional climatic change and Hydrological cycle. 
Considering that the Earth’s surface is normally covered by a 
great amount of cloud at any time (Partridge and Platt 1976), 
for reliable results from the retrieval of surface characteristics 
using remotely sensed data absolutely cloud-free pixels are 
required. But it is unrealistic. We have to consider about how to 
detect clouds and to remove them. 
Clouds have a very similar reflectance as snow. So it is very 
difficult to distinguish clouds from snow. There are many 
challenges in mapping snow cover because of the existence of 
cloud, but three items would be the most inconvenient truth (1) 
the snow cover are marked by the high reflectance of the cloud; 
(2) the snow cover are disturbed by the radiance of the cloud; (3) 
the clouds are misestimate to be snow. 
The Moderate Resolution Imaging Spectroradiometer (MODIS) 
is a 36-band spectroradiometer measuring visible and infrared 
radiation and obtaining data that are being used to derive 
  
* silvaxu@sina.com; phone +86-27-67883251; fax +86-27-67883251 
products ranging from vegetation, land surface cover, and ocean 
chlorophyll fluorescence to cloud and aerosol properties, fire 
occurrence, snow cover on the land, and sea ice cover on the 
oceans. The first MODIS instrument was launched on board the 
Terra satellite in December 1999, and the second was launched 
on Aqua in May 2002. MODIS provides snow cover 
information at 500m spatial and daily temporal resolutions. 
The MODIS snow product is part of the MODIS snow and sea 
ice global mapping project conducted by NASA’s Cryospheric 
Sciences Branch at the Goddard Space Flight Center. NASA 
creates daily snow maps indicating snow covered land, land 
without snow cover, cloud cover, seawater, lake water, and lake 
ice (Ault, 2006). There are often multiple views of snow cover 
in each day under clear skies. 
In winter MODIS snow cover products had high accuracy, but 
decreasing accuracy for the rest of the seasons, especially in 
winter-spring or fall-winter season. The main factor is the 
existence of the cloud in the snow mapping. So for the snow 
mapping the biggest challenge is to detect cloud and move it. 
In this paper, a decision tree was built to recognize the clouds 
from background for the MODIS data over Tibetan Plateau. 
And then cloud pixels were replaced by snow or vegetation or 
other pixels with a great deal repeating scenes coming from the 
same area. The goal of this study was to remove the cloud 
covered pixels from the snow cover data completely and to 
produce continuous maps of snow coverage over the Tibetan 
Plateau.