Publications

Li, L; Zha, Y (2019). Estimating monthly average temperature by remote sensing in China. ADVANCES IN SPACE RESEARCH, 63(8), 2345-2357.

Abstract
Air temperature is an important parameter affecting biosphere processes, air quality, regional climate, and human health, but is difficult to estimate, especially on a nationwide scale. Remote sensing provides high spatial resolution data that influence air temperature and can help overcome this problem in spatially heterogeneous environments. Generally, the simple linear relationships between air temperature, land surface temperature (LST), and auxiliary data, seldom consider the complex spatial variations in air temperature. The random forest (RF) regression model can characterize the complex temperature distribution by establishing the non-linear relationships among them. The main purpose of this paper is to present an RF regression method to estimate monthly average temperatures in China, using MODIS LST, normalized differential vegetation index (NDVI), night-time light, and a digital elevation model (DEM). Results based on cross-validation show that mean absolute error (MAE) and root mean square error (RMSE) at national level in different months ranged from 1.15 to 1.44 degrees C (mean +/- standard deviation: 1.30 +/- 0.10 degrees C) and from 1.57 to 1.99 degrees C (1.79 f 0.13 degrees C), respectively. Potential sources of error and uncertainty were analyzed and investigated. Results show that regression model performs better in eastern (MAE: 0.68 +/- 0.79 degrees C; RMSE: 0.75 +/- 0.81 degrees C) and central China (MAE: 0.68 +/- 0.41 degrees C; RMSE: 0.74 +/- 0.41 degrees C). The model performance was strongly dependent on elevation, slope, land cover, and satellite data quality. In addition, the percentage increase in mean squared error from RF regression model was used to quantify the importance of variables. Positive importance scores indicated an improvement in accuracy with the assistance of NDVI, night-time lights, and the DEM. LST, and more auxiliary data sources, when integrated into an RF regression model, may be ways in which to improve the modeling of air temperature variations in the future. (C) 2019 COSPAR. Published by Elsevier Ltd. All rights reserved.

DOI:
10.1016/j.asr.2018.12.039

ISSN:
0273-1177