May 23 - July 29
Choosing the Most Accurate Thresholds in a Cloud Detection Algorithm for MODIS Imagery
Tracey Dorian and Michael Douglas
What is already known:
- High resolution cloud climatologies are needed for mesoscale model validation, solar energy mapping, and other applications.
- MODIS satellite imagery has not been fully exploited to produce high resolution cloud climatologies.
What this study adds:
- We have improved cloud detection by adjusting threshold values for our cloud detection algorithm.
- The MODIS-based climatologies are now more realistic -- especially in stratus regions over oceans.
- Both advantages and limitations exist in simple techniques to identify clouds from MODIS imagery.
Abstract:
We use a cloud detection algorithm that detects cloudy pixels from MODIS images by characterizing individual pixels as cloudy or non-cloudy based on the brightness values of the pixels and a predetermined threshold. The algorithm then produces mean fields of daytime cloudiness over different geographical regions. Although the cloud climatologies produced initially appeared realistic, it was found that the algorithm largely underestimated the cloud frequencies over some regions when using a threshold of 215. Analyzing various MODIS images and recording cloudiness over different sectors served as the “ground truth” data which we compared to the algorithm output. After comparing the subjective estimates and the algorithm output for four regions of the world, we found that the algorithm underestimates cloudiness over these additional regions and that lowering the thresholds to 170-190 over oceans and 190-215 over land generally identified the thick clouds most accurately. Studying more regions or extending research on certain regions will allow us to better understand how the algorithm behaves with certain types of cloudiness and geography. Even though the thresholding technique is somewhat arbitrary, by better understanding how the algorithm behaves we can modify the algorithm to ensure that the output more accurately describes cloud climatologies around the world. If we are able to do this, then our algorithm could be used for many applications such as validating the numerical model simulations of cloud climatologies or assessing climate and potential climate change.