Downscaling of Air Temperature Data of CMIP6 Climate Models through MODIS-Based Land Surface Temperature: A case study of Yazd Province

ABSTRACT
In climate change research, the downscaling of General Circulation Model (GCM) data from coarse grid scales to finer spatial resolutions or point-based units is crucial. This study presents a method for downscaling monthly air temperature data from CMIP6 the GCMs by exploiting the empirical relationship between the near-surface air temperature of Yazd Province and the Land Surface Temperature (LST) derived from the MODIS sensor. Temperature data from three GCMs, TaiESM1, ACCESS-CM2, and CanESM5, with spatial resolutions of 105 × 118 km², 138 × 175 km², and 310 × 270 km², respectively, were downscaled to generate monthly temperature maps of Yazd Province at a spatial resolution of 1 × 1 km². The results indicated that the proposed downscaling approach outperformed conventional techniques such as the delta method. Moreover, the CanESM5 model exhibited higher accuracy than the ACCESS-CM2 and TaiESM1 models when using this approach. Subsequently, the mean monthly and annual air temperatures of Yazd Province were simulated using the TaiESM1, ACCESS-CM2, and CanESM5 models under three Shared Socioeconomic Pathway (SSP) scenarios, SSP1-2.6, SSP2-4.5, and SSP5-8.5, for the period 2015–2100. The findings reveal an upward trend in the mean annual temperature of Yazd Province over the coming decades, with an estimated increase of 15–35% (average 22%) relative to the baseline period. Among the examined CMIP6 models and SSP scenarios, the greatest projected temperature rise occurred in the CanESM5 model under the SSP5-8.5 scenario.
Extended Abstract
Introduction
Climate change, driven by the increasing concentration of greenhouse gases, particularly carbon dioxide, in the atmosphere, alters key climatic variables such as air temperature, precipitation, wind speed, and the amount of solar radiation reaching the Earth's surface. Accordingly, growing attention has been directed toward climate change in recent years because of its profound economic and social repercussions associated with extreme climatic events.
In this context, projections from climate models, commonly referred to as General Circulation Models (GCMs) or Atmospheric–Oceanic General Circulation Models (AOGCMs), provide a basis for assessing the future impacts of climate change on variables such as air temperature and precipitation. One of the essential steps in applying GCM data for climate parameter projection (e.g., air temperature) is the downscaling process, which transforms coarse-resolution outputs into finer spatial grids or point-scale data.
Since most downscaling techniques depend on ground-based observations (e.g., air temperature records from meteorological stations), their main limitation lies in the sparse distribution and limited number of ground reference points. For instance, when downscaling air temperature, only data from meteorological stations within the study area can be used, which restricts the validity of the results to those specific locations. In contrast, the method proposed in this study incorporates ground data across the entire study area on a pixel-by-pixel basis, as described in the following section.
In this study, air temperature data from CMIP6 climate models (TaiESM1, ACCESS-CM2, and CanESM5) were used to project air temperature maps for Yazd Province, located in the central Iranian Plateau and forming part of the Kavir Desert region. This region experiences hot summers and cold winters, where air temperature plays a critical role in environmental dynamics and the management of natural resources in arid lands.
Methodology
The study area is Yazd Province, which covers approximately 73,000 square kilometers. It is located between latitudes 29.5° and 33.5° N and longitudes 52.5° and 56.5° E. The datasets employed in this study comprise monthly air temperature records from 19 meteorological stations across Yazd Province, MODIS-derived Land Surface Temperature (LST) maps with a spatial resolution of 1 km, and air temperature outputs from several CMIP6 climate models featured in the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6), including TaiESM1, ACCESS-CM2, and CanESM5.
These models have spatial resolutions of 0.942° × 1.25°, 1.875° × 1.25°, and 2.813° × 2.791°, corresponding to grid cell dimensions of approximately 105 × 118 km², 175 × 138 km², and 270 × 310 km², respectively. Each grid cell of the TaiESM1, ACCESS-CM2, and CanESM5 models covers approximately 12,400 km², 23,600 km², and 83,700 km², respectively. Each model incorporates three carbon dioxide emission scenarios, SSP1-2.6, SSP2-4.5, and SSP5-8.5, which were used to simulate monthly air temperature under future climate conditions.
Accordingly, monthly air temperature data from CMIP6 climate models were extracted for the periods 1850–2014 (baseline) and 2015–2100 (future) under the aforementioned emission scenarios. A regression analysis was conducted to examine the relationship between monthly air temperature and LST data from the 19 meteorological stations, thereby establishing the air temperature–LST regression models for the study area. These relationships were subsequently applied to produce air temperature maps and corresponding dimensionless monthly temperature maps of the study area at a spatial resolution of 1 × 1 km². These maps represent the ratio of each pixel’s monthly temperature to the mean air temperature of the entire region. The dimensionless maps were then used to downscale the coarse-resolution temperature outputs from CMIP6 models into monthly air temperature maps with a 1 × 1 km² pixel size.
Results and Discussion
The results of this study indicate that the proposed downscaling method for CMIP6 air temperature data performs with varying efficiency across the three climate models with different spatial resolutions. Specifically, the CanESM5 model achieved higher accuracy compared with the ACCESS-CM2 and TaiESM1 models. As noted earlier, each grid cell of the CanESM5, ACCESS-CM2, and TaiESM1 models covers approximately 12,400 km², 23,600 km², and 83,700 km², respectively, with spatial resolution decreasing in the same order. Therefore, it can be concluded that the proposed downscaling method tends to yield higher accuracy for climate models with coarser spatial resolution.
Among the CMIP6 climate models examined, BCC-CSM1.1, BNU-ESM, CanESM2, FIO-ESM, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC-ESM, MIROC-ESM-CHEM, and NorESM1-M exhibit relatively coarse grid sizes. Therefore, these models are recommended for applying the proposed method to downscale their air temperature data. The observed variations among model outputs can also be attributed to inherent uncertainties in climate modeling, arising from differences in the underlying assumptions, input data, and auxiliary variables used for simulation. Nevertheless, further research is recommended to quantify and reduce these model uncertainties.
According to the final results, the mean annual air temperature of Yazd Province is projected to exhibit an upward trend over the coming decades, with an increase of 15–35% (average 22%) relative to the baseline period. Among the three analyzed climate models, the greatest temperature increase was projected by the CanESM5 model, whereas the highest warming occurred under the SSP5-8.5 scenario.
 
Conclusion
This study introduced a method for downscaling large-scale air temperature data from CMIP6 climate models using region-specific monthly air temperature–LST regression relationships. The proposed approach was successfully applied to downscale data from three CMIP6 models: TaiESM1, ACCESS-CM2, and CanESM5. Each model’s grid cell covers approximately 12,400 km², 23,600 km², and 83,700 km², respectively. Using this method, the final monthly air temperature maps of the study area were produced at a spatial resolution of 1 × 1 km².
The results also indicated that the proposed approach performs more accurately for CMIP6 climate models with coarser spatial resolution compared to those with finer grids. This method is recommended for downscaling CMIP6 air temperature data in regions with similar climatic and topographic characteristics. However, additional investigations are required before applying this approach to regions with different climatic and surface conditions.
Moreover, the main limitations of the proposed method are its time-consuming nature and lack of user-friendly implementation. These challenges are expected to be addressed through the development of automated algorithms or software tools based on this approach, capable of connecting directly to satellite databases to retrieve LST values for the study area.
 
Funding
There is no funding support.
 
Authors’ Contribution
Authors contributed equally to the conceptualization and writing of the article. All of the authors approved thecontent of the manuscript and agreed on all aspects of the work declaration of competing interest none.
 
Conflict of Interest
Authors declared no conflict of interest.
 
Acknowledgments
We are grateful to all the scientific consultants of this paper.