محاسبه و ارزیابی دمای سطح زمین با استفاده از الگوریتم پنجره مجزای غیرخطی و تصاویر ماهواره سنتینل 3 مطالعه موردی: استان تهران

در سال‌های اخیر دمای سطح زمین (lst) اهمیت زیادی در مطالعات علوم‌ زمین و محیط‌زیست پیدا کرده است. فناوری سنجش‌ازدور، امکان پایش مکانی و زمانی این کمیت را در سطوح وسیع فراهم می‌آورد. این پارامتر از طریق تصاویر ماهواره‌ای با حداقل یک باند حرارتی فراهم می‌شود. در این مطالعه از روش پنجره مجزای غیرخطی توسط ماهواره  سنتینل3 در طول فصول مختلف سال 1397 برای محاسبه دمای سطح زمین استفاده شد و همچنین یک روش اعتبارسنجی مستقیم و غیرمستقیم برای آن ارائه شده است. روش اعتبارسنجی برمبنای ارزیابی قطعی این محصول با داده میدانی، و ارزیابی نسبی آن با محصولات دمای مادیس و slstr می‌باشد. همچنین از روش برآورد گسیلمندی برمبنای شاخص پوشش گیاهی برای تخمین دما از روش پنجره مجزای غیرخطی باتوجه به دو باند حرارتی تصاویر سنتینل3 استفاده شد. برای اطمینان بیشتر، محصولات دمای مادیس و slstr نیز به‌صورت مستقیم با داده میدانی ارزیابی قطعی شد. به‌طور کلی نتایج حاصل از محصول دمای مادیس، slstr و دمای برآورد شده از روش پنجره مجزای غیرخطی روندی مشابه را برای تغییرات دما در طول فصول سال نشان دادند. به‌طور خلاصه، با توجه به دو روش اعتبارسنجی مستقیم و غیرمستقیم برای دمای برآورد شده از روش پنجره مجزای غیرخطی، فصل تابستان با مقادیر بزرگ میانگین مربع خطاها (2/46)، و فصل زمستان با مقادیر کوچک میانگین مربع خطاها (0/86) به‌ترتیب کمترین و بیشترین نتایج را برای فصول در سال 1397 ارائه دادند. در نهایت، با توجه به نتایج به‌دست آمده دمای برآورد شده هم به‌صورت قطعی و هم به‌صورت نسبی نتایج مطلوبی را برای تمام فصول در مقیاس زمانی و مکانی گسترده فراهم می‌کند که می‌تواند در مقیاس‌های بزرگ برای برآورد دما در حل بحران‌های زیست‌محیطی و همچنین تغییر اقلیم از آن استفاده نمود.

Calculating land surface temperature using nonlinear split window algorithm and sentinel3 satellite imagery Case study: Tehran Province

Extended AbstractIntroduction   As a key parameter describing physics of land surface processes on local and global scales, land Surface Temperature (LST) is the result of all interactions and energy flows between land surface and the atmosphere. Temperature changes rapidly on temporal and spatial scales, and thus a complete description of LST require measurements involving spatial and temporal frequencies. Hence, climatological, meteorological, and hydrogeological studies require having access to wide scale information about spatial changes of air temperature. Since the LST product of SLSTR uses linear splitwindow algorithm, the present study has used nonlinear splitwindow algorithm to estimate LST in Sentinel3 images. Linearity of the radiation transfer equation in linear algorithm and some approximations used in splitwindow algorithms (such as transfer approximation as a linear function of vapor value) result in considerable errors because of which nonlinear algorithm is used in the present study. Using linear splitwindow algorithm to estimate LST in tropical climates also leads to a high level of error. The present study seeks to estimate LST using a nonlinear splitwindow algorithm and data retrieved from Sentinel3 in different seasons of 2018 and 2019. The results are also evaluated using temperature product of MODIS and SLSTR. Materials & Method   A time series of sentinel3 images retrieved from 2018 to 2019 was used as research data. Data were collected by Sentinel3 SLSTR sensors operated by the European Space Agency (ESA). Obviously, images shall be radiometrically corrected before calculating physical land surface parameters such as temperature, emissivity, reflectance and radiance, albedo, and etc. To reach this goal, it is necessary to omit or minimize the effect of atmosphere, epipolar geometry of sensor, sunlight, topography, and surface characteristics while estimating surface parameters in these images. The current study seeks to estimate LST applying a nonlinear splitwindow algorithm on Sentinel3 data collected during different seasons of 2018 and 2019 and to evaluate the results using temperature product of MODIS, SLSTR, and insitu data. Pearson Correlation Coefficient and Root Mean Square Error (RMSE) were also used as relative and quantitative criteria to evaluate the accuracy of the proposed method and determine the deference between temperature calculated by the proposed method and temperature product of MODIS and SLSTR sensor. Hence, four frames of LST product collected by MODIS, and SLSTR in April, June, and October, 2018 and January, 2019 were used to evaluate the proposed method. Results & Discussion   The proposed method was also indirectly evaluated using temperature products of MODIS and SLSTR sensor. Applying parameters of mean and root mean square error, the evaluation has shown that the results obtained from the proposed method in the oneyear reference period were more similar to the results obtained from MODIS sensor. Comparing nonlinear SplitWindow algorithm and MODIS products, RMSE ranged from 1.21 to 2.46 and the highest and lowest accuracy belonged to winter and summer, respectively. Comparing this algorithm with the SLSTR product, RMSE ranged from 0.76 to 2.24 and the highest and lowest accuracy belonged to winter and summer, respectively. Proper performance of the algorithm in winter is due to the relative balance of atmospheric water vapour in this season. Comparing nonlinear modelling of atmospheric water vapour in the nonlinear algorithm of a Splitwindow and the linear algorithm in SLSTR and MODIS products, the small difference between temperature calculated by the algorithm and the products can be justified. However, due to temperature fluctuations in summer, results obtained by the proposed method were not reliable enough compared to both temperature products. Generally, results obtained from the proposed method showed a higher correlation with the temperature product of SLSTR sensor, which is due to the similar spectral bands used in calculating the surface temperature. Relative comparison of the SplitWindow and the MODIS product’s nonlinear algorithm showed a coefficient of determination ranging from 0.76 to 0.96, while comparing this algorithm with the SLSTR product showed a determination coefficient of 0.80 to 0.98. Comparing temperature obtained from the nonlinear SplitWindow algorithm with SLSTR and MODIS temperature products, the proposed algorithm was relatively stable no matter which season was taken into account. Conclusion   The present study seeks to estimate Land Surface Temperature using a nonlinear SplitWindow algorithm and Sentinel3 data collected in different seasons. Values obtained from the algorithm were validated using insitu dataset retrieved from the meteorological station. They were also evaluated using temperature product of MODIS and SLSTR. To increase the accuracy level, temperature product of MODIS and SLSTR were also evaluated and compared with the insitu dataset and provided good results. Generally, there is a significant difference between temperature values estimated by the NSW algorithm for different seasons especially summer. However, a similar trend was observed in temperature changes reported by SLSTR and MODIS, and the proposed algorithm in different seasons of the study area. Although, the nonlinear SplitWindow algorithm showed a higher accuracy in spring and winter, overall results indicated that the proposed method was relatively stable no matter which season was taken into account. It can be concluded that LST estimation with nonlinear Splitwindow method and Sentinel3 satellite data has an acceptable level of accuracy and thus, can be used in large scale environmental crises such as climate changes.