USING MCSST METHOD FOR MEASURING SEA SURFACE TEMPERATURE WITH MODIS IMAGERY AND MODELING AND PREDICTION OF REGIONAL VARIATIONS WITH LEAST SQUARES METHOD (CASE STUDY: PERSIAN GULF, IRAN)
- 1Dept. of GIS and Remote Sensing Engineering, Islamic Azad University, Yazd Branch, Iran
- 2Dept. of Industrial Engineering, Iran University of Science and Technology, Iran
- 3Dept. of Information Technology, Shiraz University, Iran
Keywords: Thermal Remote Sensing, MCSST, MODIS, Modeling, Least Squares, Persian Gulf
Abstract. Nowadays, many researchers in the area of thermal remote sensing applications believe in the necessity of modeling in environmental studies. Modeling in the remotely sensed data and the ability to precisely predict variation of various phenomena, persuaded the experts to use this knowledge increasingly. Suitable model selection is the basis for modeling and is a defining parameter. So, firstly the model should be identified well. The least squares method is for data fitting. In the least squares method, the best fit model is the model that minimizes the sum of squared residuals. In this research, that has been done for modeling variations of the Persian Gulf surface temperature, after data preparation, data gathering has been done with multi-channel method using the MODIS Terra satellites imagery. All the temperature data has been recorded in the period of ten years in winter time from December 2003 to January 2013 with dimensions of 20*20 km and for an area of 400 km2. Subsequently, 12400 temperature samples and variation trend control based on their fluctuation time have been observed. Then 16 mathematical models have been created for model building. After model creation, the variance of all the models has been calculated with ground truth for model testing. But the lowest variance was in combined models from degree 1 to degree 4. The results have shown that outputs for combined models of degree 1 to degree 3 and degree 1 to degree 4 for variables does not show significant differences and implementation of degree 4 does not seem necessary. Employment of trigonometric functions on variables increased the variance in output data. Comparison of the most suitable model and the ground truth showed a variance of just 1⁰. The number of samples, after elimination of blunders reduced to 11600 samples. After this elimination, all the created models have been run on the variables. Also in this case, the highest variance has been obtained for the models that have trigonometric functions for variables. The lowest variance has been obtained for combined models of degree 1 to 3 and 4. After the elimination of the blunders and running the model once more, the variance of the most suitable model reached to 0.84⁰. Finally, among the 16 models, 13 models showed less than 1⁰ that demonstrates the accuracy of the selected models for scrutiny and predict variations of the surface temperature in this area.