Publications

Meliho, M; Khattabi, A; Zejli, D; Orlando, CA (2024). Spatiotemporal prediction of daily air temperature using remote sensing and machine learning in Morocco. THEORETICAL AND APPLIED CLIMATOLOGY, 155(3), 2105-2117.

Abstract
Air temperature (AT) is an extremely useful metric for a wide range of climate change applications and studies. While its data are usually obtained from measurements at meteorological stations, this may be limited over large areas due to the unavailability of weather stations or lack of reliable data due to station-specific errors. With the availability of remote sensing data, including moderate resolution imaging spectroradiometer (MODIS), many of these constraints are overcome. In this study, two machine learning (ML) algorithms, including random forest (RF) and Cubist, and linear regression were employed to estimate AT using MODIS data over Morocco. The ML models were found to be quite well suited to predicting AT, recording RMSE values ranging from 2.40 to 2.76 degrees C. The Cubist and random forest models, with RMSE of 2.40 and 2.42 degrees C, respectively, and R2 of 0.82, were more accurate than the linear model. The linear model had the least predictive power, recording RMSE and R2 of 2.76 degrees C and 0.75, respectively. This study highlights the importance of station data availability, with model performance reduced at remote stations, particularly in the east and south of the country. Despite this, the use of ML techniques in conjunction with satellite data to predict AT in Morocco provides an interesting alternative, which could be a useful tool for climate scientists as well as policy makers.

DOI:
10.1007/s00704-023-04759-9

ISSN:
0177-798X