General Circulation Models (GCMs) are the primary tools for simulating climate change at the global scale, but systematic biases in their outputs limit their ability to reproduce local climatic conditions. This study aimed to correct temperature biases in CANESM2 model outputs using a Nonlinear AutoRegressive with eXogenous inputs (NARX) neural network at Yazd Synoptic Station from 2006 to 2020. The proposed NARX approach uses GCM-simulated temperatures as exogenous inputs and lagged observed temperatures as autoregressive inputs. This allows the network to learn nonlinear relationships and temporal dependencies between simulated and observed temperature series, improving systematic bias correction in climate model outputs. Monthly GCM temperature outputs and corresponding observations were used to train the NARX model. Its performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Results showed that CANESM2 had considerable errors in reproducing monthly temperatures at Yazd, with MSE, RMSE, and MAE values of 35.83, 5.99, and 5.45, respectively. After NARX correction, these metrics fell to 5.21, 2.28, and 1.84, respectively, with MSE decreasing by more than 85%. Error histograms and response plots indicated good generalization and no overfitting. Monthly and seasonal analyses showed that NARX effectively corrected systematic biases in CANESM2 outputs, particularly during the cold season, and substantially reduced the dispersion of prediction errors. Overall, NARX offers an effective framework for correcting biases in climate-model temperature outputs and has considerable potential to improve climate data in arid regions such as central Iran.
Asadi Zarch, M. A. & Ghaznavi, S. (2026). Evaluation of CANESM2 Temperature Bias Correction Using NARX (Case Study: Yazd Synoptic Station, Central Iran). Desert Management, 14(1), 103–124. https://doi.org/10.22034/jdmal.2026.2096224.1535
MLA
Asadi Zarch, M. A., & Ghaznavi, S. "Evaluation of CANESM2 Temperature Bias Correction Using NARX (Case Study: Yazd Synoptic Station, Central Iran)", Desert Management, 14, 1, 2026, 103–124. doi: 10.22034/jdmal.2026.2096224.1535
HARVARD
Asadi Zarch, M. A., & Ghaznavi, S. (2026). 'Evaluation of CANESM2 Temperature Bias Correction Using NARX (Case Study: Yazd Synoptic Station, Central Iran)', Desert Management, 14(1), pp. 103–124. doi: 10.22034/jdmal.2026.2096224.1535
CHICAGO
Asadi Zarch, M. A. & Ghaznavi, S., "Evaluation of CANESM2 Temperature Bias Correction Using NARX (Case Study: Yazd Synoptic Station, Central Iran)." Desert Management, 14 1 (2026): 103–124, doi: 10.22034/jdmal.2026.2096224.1535
VANCOUVER
Asadi Zarch M. A., Ghaznavi S. Evaluation of CANESM2 Temperature Bias Correction Using NARX (Case Study: Yazd Synoptic Station, Central Iran). Desert Management. 2026;14(1):103–124 (In Persian). doi: 10.22034/jdmal.2026.2096224.1535
