1 of 25

Evaluation of BATS and CLM Land Surface Schemes in RegCM5 for Temperature Simulation over Bangladesh

Presented by

Jannatul Ferdous Jerin

2 of 25

CLM4.5 showed superior temperature correlation, especially in winter, and better performance in spring and winter based on key statistical indices. However, the BATS scheme generally produced lower temperature biases across most seasons (Moghadam et al. 2024)

CLM4.5 significantly amplified warm biases, leading to overestimations of up to 14°C. BATS produced lower temperature biases and a more accurate representation of seasonal peaks. Consequently, for temperature simulation over West Africa, BATS scheme outperformed CLM4.5 (Kouassi et al. 2022).

CLM3.5 outperformed BATS by generating temperatures closer to observations, reducing warm biases, especially over the Amazon. Accurately reproduced the magnitude of interannual temperature anomalies, while BATS tended to over-amplify variability America (Llopart et al., 2017).

Literature Review

3 of 25

No prior evaluation of BATS-CLM land surface schemes in RegCM for Bangladesh.

Existing studies focus more on precipitation, not temperature.

No comparative analysis of scheme performance for local temperature simulation.

Lack of data for recent extreme temperature event modeling.

No validated model configuration for reliable temperature projections of Bangladesh.

Research Gap

4 of 25

Research Objective

This study is guided by the following specific objectives:

To conduct historical period simulations over the Bangladesh using RegCM5 configured with the BATS and CLM land surface schemes.

To evaluate the performance of the uncorrected BATS and CLM simulations in replicating observed gridded temperature data across Bangladesh on annual, seasonal, and monthly timescales.

To apply a statistical bias correction technique to the raw model outputs to minimize systematic errors.

To perform a comparative analysis of the bias-corrected outputs from both schemes to determine which LSS most realistically simulates temperature over the study area.

5 of 25

Research Methodology

Spatial Distribution Analysis

Annual
Seasonal
Monthly

Temporal Pattern Analysis

Dhaka
Chattogram
Barisal
Khulna
Sylhet
Rajshahi
Rangpur
Mymensingh

Statistical Metrics evaluation

6 of 25

Data Source

Boundary and Initial Conditions: The ERA-Interim reanalysis dataset from the ECMWF was used to provide initial and lateral boundary conditions required to run the RegCM5 for the period 2015–2016.

Validation Data (Gridded): The Climatic Research Unit (CRU TS) gridded dataset provided a benchmark for the spatial validation of the simulated temperature.

Validation and Bias Correction Data (Station): Observational data from several stations across Bangladesh, obtained from the Bangladesh Meteorological Department (BMD), was used for point-based validation and to calculate the parameters for statistical bias correction.

7 of 25

Domain

Centered over Bangladesh

20°N to 27° N latitude

87°E to 93° E longitude.

8 of 25

Spatial Analysis

9 of 25

Annual Mean Temperature Analysis

Systematic Warm Bias: Both BATS and CLM overestimate annual mean temperatures compared to CRU observations.

Bias Magnitude: CLM shows stronger warming (27–29°C) than BATS (26–28°C) against observed 25–26°C.

10 of 25

Widespread Bias: The warm bias is consistent across all three columns (a, b, c), indicating a regional, not local, issue.

CLM's Stronger Bias: CLM shows a slightly larger deviation from observations compared to BATS.

Practical Implication: Results confirm that region-specific model tuning and bias correction are necessary before application.

11 of 25

Seasonal Mean Temperature Analysis

Pre-monsoon: As the season with highest temperature, pre-monsoon season is analyzed

Bias Severity: CLM shows larger overestimation (29-31°C) compared to BATS (28-29°C).

Observed Range: CRU data shows seasonal range (25-29°C) with clear spatial gradient.

Gradient Failure: Both models fail to capture the NE-SW temperature gradient observed in CRU data.

Spatial Inaccuracy: Worst performance in northwestern/central regions.

12 of 25

Spatial Gradient Failure: Both BATS and CLM fail to accurately represent the northeast-southwest temperature gradient observed in CRU data.

Northeastern warm Bias: The models consistently overestimate temperatures in the northeastern highland regions, failing to capture the area's cooler climatic conditions.

Uniform Overestimation Pattern: Both schemes exhibit lack in the regional variability seen in observations.

13 of 25

Monthly Mean Temperature Analysis

April Peak Heat Analysis: April, being the hottest month in Bangladesh, was selected to evaluate model performance under extreme temperature conditions.

CLM's Superior Performance: The CLM scheme provided better results, closely matching the observed intensity and spatial pattern of peak temperatures.

BATS Underestimation: The BATS scheme consistently underestimated temperatures during this critical high-heat period.

14 of 25

While CLM accurately simulates the magnitude of extreme monthly temperatures, it fails to reproduce the observed spatial gradient across Bangladesh.

The BATS scheme systematically underestimates temperatures, showing a persistent cold bias.

Both models struggle to represent the northeast-southwest temperature gradient, indicating a limitation in capturing regional climatic controls.

This analysis confirms CLM's stronger capability in simulating extreme temperature events compared to BATS.

15 of 25

Temporal Analysis & Bias Correction – BATS

16 of 25

BATS shows smaller temperature bias but struggles to capture temporal variability at some stations.

Chattogram station shows strongest agreement, suggesting better model performance in coastal zones.

Statistical adjustment significantly improves results, though temporal pattern issues persist inland.

Fails to fully represent temperature evolution at several stations, indicating need for localized calibration.

17 of 25

Both raw and bias-corrected BATS outputs fail to capture the observed temporal variability

Statistical post-processing does not improve the model's ability to capture temperature patter over the simulation period

Underestimates temperature in Sylhet station, unlike all the other stations

18 of 25

Temporal Analysis & Bias Correction – CLM

19 of 25

CLM consistently overestimates temperatures.

Successfully captures the temporal patterns at Dhaka, Barishal, Chattogram, and Khulna stations.

Similar to BATS, CLM demonstrates its best performance at the Chattogram station

20 of 25

CLM demonstrates better capture of temporal variability compared to BATS at Mymensingh, Rajshahi, and Rangpur stations.

CLM overestimates temperatures at all stations except Sylhet

Similar to BATS, underestimates temperature at the Sylhet station

21 of 25

Station	BATS (MSE)	CLM (MSE)	BATS (MAE)	CLM (MAE)	BATS (RMSE)	CLM (RMSE)	BATS (R²)	CLM (R²)
Chattogram	0.61	0.54	0.65	0.60	0.78	0.73	0.93	0.94
Sylhet	1.95	1.28	1.16	0.94	1.40	1.13	0.83	0.89
Dhaka	1.42	3.99	1.05	1.58	1.19	2.00	0.89	0.68
Khulna	2.53	3.70	1.20	1.54	1.59	1.92	0.83	0.74
Barishal	2.24	6.01	1.17	2.10	1.50	2.45	0.84	0.57
Rajshahi	3.04	7.75	1.47	2.30	1.74	2.78	0.85	0.63
Rangpur	3.91	7.49	1.67	2.27	1.98	2.74	0.78	0.59
Mymensingh	3.79	8.38	1.65	2.45	1.95	2.90	0.75	0.46

Statistical Metrics Comparison

BATS demonstrates superior overall performance across most stations with lower errors (MSE, MAE, RMSE) and higher correlation (R²).

CLM excels regionally, delivering the best results specifically for Chattogram and Sylhet.

Model effectiveness is location-specific, depending on local climatic and geographical conditions.

22 of 25

Key Finding	BATS Model	CLM Model
Spatial Performance	Relatively smaller warm bias across most areas.	Substantial warm bias, especially in the northwestern and central regions.
Temporal Pattern Capture	Poorer performance in capturing temporal variability.	Superior in capturing variations, despite overestimation.
Extreme Temperature Representation	Struggled to capture extreme temperature magnitudes.	Better at capturing extreme temperatures during critical periods.
Station-Specific Performance	Limited temporal pattern capture in regional stations (Mymensingh, Rangpur, Rajshahi, Sylhet).	Similar challenges in regional stations, with some skill at major stations.
Bias Correction Efficacy	Effective in mitigating warm bias and improving alignment.	Highly effective in reducing warm bias and improving alignment.

Key Findings of the Study

23 of 25

Limitations

Short Simulation Period: The study was limited to a two-year period (2015-2016) due to computational constraints. This is too short to assess model performance across different years with natural climate variability.

Lack of Data: The validation was constrained by a lack of high-resolution, continuous data for the recent years to assess the current climate analysis.

Missing Heatwave Analysis: A significant consequence was the inability to perform an in-depth analysis of extreme heat due to time computational constrain.

24 of 25

Reference

Llopart, M., da Rocha, R.P., Reboita, M. and Cuadra, S., 2017. Sensitivity of simulated South America climate to the land surface schemes in RegCM4. Climate Dynamics, 49(11), pp.3975-3987.

Kosari Moghadam, L., Ghahreman, N., Babaeian, I. and Irannejad, P., 2024. Sensitivity study of RegCM4. 7 model to land surface schemes (BATS and CLM4. 5) forced by MPI-ESM1. 2-HR in simulating temperature and precipitation over Iran. Theoretical and Applied Climatology, 155(9), pp.8515-8532.

Kouassi, A.A., Kone, B., Silue, S., Dajuma, A., N’datchoh, T.E., Adon, M., Diedhiou, A. and Yoboue, V., 2022. Sensitivity study of the RegCM4’s surface schemes in the simulations of West Africa climate.