Modeling Land Surface Temperature Using UAV-Derived RGB and NIR Data Through Machine Learning Techniques

Land Surface Temperature (LST) estimation is an important part of climate research, helping understand surface heat and environmental changes. This study introduces a simple and innovative way to estimate LST using machine learning and data collected by unmanned aerial vehicle (UAV). The UAV used RGB and near-infrared (NIR) sensors, which are commonly available and affordable.

The research took place in the Šumava Mountains of the Czech Republic, an area with unique landscapes and sensitive ecosystems. The UAV surveys used two types of cameras: one combined RGB and NIR sensors to capture visual and near-infrared data, and the other was a thermal camera to measure ground temperature. The thermal images provided the training data for machine learning models, which were designed to estimate LST using only RGB and NIR data. To test and validate the model, an integrated approach is used: sensors installed in different land cover types, direct measurements of air temperature from ground stations and medium-resolution satellites with a thermal band. This correlation with reference temperature sources ensures the model reflects real thermal conditions rather than relative differences alone. This method can be very useful when thermal cameras are not available, as they are often expensive and need careful calibration.

The models created in this study showed good accuracy, with strong agreement between the predicted and actual LST values but it is still necessary to check the LST directly. Incorporating reference temperature values enhances the model’s accuracy and applicability, allowing for consistent results. This means the models can reliably predict LST using just RGB and NIR data. This approach offers a practical alternative to traditional thermal measurements, which are more costly and harder to use for large-scale or frequent studies. One key advantage of this method is its affordability and ease of use. RGB and NIR sensors are much more accessible than thermal cameras, making it possible for researchers with limited budgets to monitor LST  effectively. 

This study offers a novel method for estimating LST by combining UAV technology, RGB and NIR sensors, and machine learning. The results show that the proposed approach is reliable and applicable for environmental and climate research. By integrating reference temperature sources, this study overcomes the challenges of relative-only measurements, providing reliable LST values for diverse applications. By overcoming the challenges of direct thermal measurements, this method provides an easier way to monitor land surface temperatures across different environments.