The Improved U-STFM: A Deep Learning-Based Nonlinear Spatial-Temporal Fusion Model for Land Surface Temperature Downscaling

Shanxin Guo,Min Li,Yuanqing Li,Jinsong Chen,Hankui K. Zhang,Luyi Sun,Jingwen Wang,Ruxin Wang,Yan Yang

DOI: https://doi.org/10.3390/rs16020322

IF: 5

2024-01-13

Remote Sensing

Abstract:The thermal band of a satellite platform enables the measurement of land surface temperature (LST), which captures the spatial-temporal distribution of energy exchange between the Earth and the atmosphere. LST plays a critical role in simulation models, enhancing our understanding of physical and biochemical processes in nature. However, the limitations in swath width and orbit altitude prevent a single sensor from providing LST data with both high spatial and high temporal resolution. To tackle this challenge, the unmixing-based spatiotemporal fusion model (STFM) offers a promising solution by integrating data from multiple sensors. In these models, the surface reflectance is decomposed from coarse pixels to fine pixels using the linear unmixing function combined with fractional coverage. However, when downsizing LST through STFM, the linear mixing hypothesis fails to adequately represent the nonlinear energy mixing process of LST. Additionally, the original weighting function is sensitive to noise, leading to unreliable predictions of the final LST due to small errors in the unmixing function. To overcome these issues, we selected the U-STFM as the baseline model and introduced an updated version called the nonlinear U-STFM. This new model incorporates two deep learning components: the Dynamic Net (DyNet) and the Chang Ratio Net (RatioNet). The utilization of these components enables easy training with a small dataset while maintaining a high generalization capability over time. The MODIS Terra daytime LST products were employed to downscale from 1000 m to 30 m, in comparison with the Landsat 7 LST products. Our results demonstrate that the new model surpasses STARFM, ESTARFM, and the original U-STFM in terms of prediction accuracy and anti-noise capability. To further enhance other STFMs, these two deep-learning components can replace the linear unmixing and weighting functions with minor modifications. As a deep learning-based model, it can be pretrained and deployed for online prediction.

environmental sciences,imaging science & photographic technology,remote sensing,geosciences, multidisciplinary

What problem does this paper attempt to address?

### Problems the paper attempts to solve

This paper aims to solve the problem of spatio - temporal resolution limitations when measuring land surface temperature (LST) through the thermal infrared band of satellite platforms. Specifically, due to the limitations of satellite orbits and sensor designs, it is difficult for a single sensor to provide land surface temperature data with both high spatial resolution and high temporal resolution simultaneously. This has led to challenges in obtaining high - quality LST data in daily applications, such as evapotranspiration observation and urban heat island effect modeling.

To solve this problem, this paper proposes an improved non - linear spatio - temporal fusion model (U - STFM), namely **non - linear U - STFM**. The model replaces the original linear mixing and weighting functions by introducing two deep - learning components - the Dynamic Network (DyNet) and the Ratio Network (RatioNet). These improvements are aimed at:

1. **Solving the deficiencies of the linear mixing assumption**: The traditional linear mixing assumption cannot fully represent the non - linear energy mixing process of LST, especially in coarse pixels where there may be hot or cold spots that are independent of the coverage fraction.
2. **Improving noise resistance**: The original weighting function is very sensitive to small errors in the input data, reducing the reliability of the final LST prediction.
3. **Enhancing the generalization ability of the model**: Through the deep - learning components, the model can be trained on a small number of data sets and maintain a high generalization ability, being applicable to data at different temporal and spatial scales.

Through these improvements, the non - linear U - STFM model can generate LST products with a high spatial resolution (30 meters) while maintaining high accuracy, thus better meeting the needs of practical applications.