Prepared for Special Topics in Remote Sensing course given in the Geomatics Engineering Department of Istanbul Technical University. Thanks Prof. Dr. Elif Sertel and Res. Assis. Samet Aksoy for their contributions and efforts within the scope of the course.
Changing climate, agriculture, hydrology, and biodiversity, for example, all require precise air temperature data at varying temporal and geographical resolutions (Vogt et al., 1997; Monestiez et al., 2001; Yang et al., 2013; Ge et al., 2014). In furthermore, temperature of the air is utilized to investigate the influence of air temperature on plant photosynthesis or respiration. For illustration, the leaf temperature required to achieve the greatest net photosynthesis rate is in the 20–30°C range. As a result, environmental factors linked with this energy balance (light intensity, temperature, humidity, air flow rate, and so on) impact leaf temperature. As a result, this affects photosynthesis and growth. The most important element of the climate is the reason why humidity, precipitation, pressure and wind are formed from other elements. At this point, the importance of the work done, the air temperatures of the regions can be estimated with remote sensing and indirectly information about the rate of photosynthesis or respiration in plants can be obtained. By means of remote sensing of the air temperature apart from the local data, the air temperature data can be transformed into a continuous data instead of being a point data. Many articles have been written on the use of different algorithms and different data in order to predict the air temperature correctly, and the most accurate method has been tried to be found in different areas. Shen et al. extensively used deep learning to map air temperature using satellite images and base station inputs. China was chosen as the venue for the study. The study used MODIS LST and MOD13A2 for NDVI data, MODIS Annual Terra and Aqua combined MODIS land use / cover product (MCD12Q1) of 2015 for land cover, and SRTM for elevation. Data was fed into the computer, and the results were achieved by using a deep belief network as a deep learning approach (2020). Xu et al. investigated the feasibility of forecasting month air temperature using machine learning algorithms using remote sensing data. The study's focal point is the Tibetan Plateau. Ten deep learning methods were applied with the data to correlate the air temperature and environmental parameters supplied by MODIS. The Cubist methodology was chosen as the best method since it has the lowest prediction error and also the lowest variation among errors achieved with higher overall values and errors obtained with poorer quality samples (2018). Adab et al. employed four forms of quick linear data structures to predict soil moisture in their study (ANN, SVM, RF, and EN). As datasets, Landsat-8 satellite data, soil samples, CHIRPS, and SMAP data were employed. The region of Khorasan-Razavi in Iran was chosen as the location for the investigation. According to the findings obtained from the RF, SVR, ANN, and EN models, the RF model beat the other three approaches for estimating soil moisture (2020). Melinho et al. used MODIS images and air temperature measurements collected at meteorological stations in Morocco's Souses basin to train six machine learning algorithms on monthly and weekly air temperature predictions. It exhibits the capacity of machine learning algorithms and remote sensing data to forecast temperature using effective models. Aqua and Terra sensors are suitable for LST. In Morocco's Souses basin, the Cubist and RF ML algorithms were the most successful in forecasting air temperature (2022). Wang et al. (2010) assessed the efficacy of two regression studies and three spatial interpolation approaches in forecasting monthly air temperature in China. The findings of many GWR models were considered to be the most accurate, and hence adequate for predicting monthly air temperature on a broader scale than the MLR model. Kriging produced the best results of any interpolation method (2017). Satellite images and terrestrial station data from four state regions in the United States (Washington, Oregon, Montana, and Idaho) were used in this study; regression was performed using four different algorithms: Random Forest (RFR), Support Vector Regression-Radial Basis Function (SVR-RBF), Support Vector Regression-Linear (SVR-Linear), and smileCart. It is aimed to determine the most accurate algorithm as a result of the accuracy values of the regressions. In addition to the monthly average data of 1102 local stations in line with the study; One-month average temperature data from MODIS and Landsat-8 satellites were used.
With the regressions performed, estimation maps were obtained, and the obtained maps were displayed as a result of the classification. The Coefficient of Determination (R²) and Root Mean Square Error (RMSE) values resulting from the regressions enabled us to make sense of the algorithm's accuracy at this point, not only visually, but also numerically. Here, R² is expected to converge to 1 and RMSE to 0. In the regressions carried out in this direction, smileCart is the algorithm with the most accurate prediction result in all data combinations, while the best regression has emerged when three data are together. When a satellite image and terrestrial station data are used, the regression with Landsat-8 has more erroneous predictions than the MODIS satellite image. When Landsat-8 entered into regression with MODIS satellite images, the most accurate prediction results and values were obtained.
All data used for exercises are obtained from open source database/resources such as The National Centres for Environmental Information open source data access site and Google Earth Engine datasets.
https://code.earthengine.google.com/de6b52ce3b004063ffe5243f115d86c4 https://code.earthengine.google.com/d1284baf94c2c4da3652c37eaeca9266 https://code.earthengine.google.com/eed01603d9e28101ce359f1e76f57e99
This work is licensed under the GPL v3.0.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent