TYPE Original Research

PUBLISHED 11 March 2024

DOI 10.3389/fenvs.2024.1341807

Identification of optimal
OPEN ACCESS CMIP6 GCMs for future typical
EDITED BY
Amit Awasthi,
University of Petroleum and Energy Studies,
meteorological year in major
India

REVIEWED BY
cities of Indonesia using
Abu Reza Md. Towfiqul Islam,
Begum Rokeya University, Bangladesh
Alok Sagar Gautam,
multi-criteria decision analysis
Hemwati Nandan Bahuguna Garhwal
University, India Vinayak Bhanage 1*, Han Soo Lee 1,2, Jonathan Salar Cabrera 1,3,
*CORRESPONDENCE
Vinayak Bhanage,
Tetsu Kubota 1, Radyan Putra Pradana 1,4, Faiz Rohman Fajary 1,5
[email protected] and Hideyo Nimiya 6
RECEIVED 21 November 2023 1
Transdisciplinary Science and Engineering Program, Graduate School of Advanced Science and
ACCEPTED 19 February 2024 Engineering, Hiroshima University, Hiroshima, Japan, 2Center for Planetary Health and Innovation
PUBLISHED 11 March 2024
Science (PHIS), The IDEC Institute, Hiroshima University, Hiroshima, Japan, 3Faculty of Computing, Data
CITATION Sciences, Engineering and Technology, Davao Oriental State University, Mati, Philippines, 4Center for
Bhanage V, Lee HS, Cabrera JS, Kubota T, Research and Development, Indonesian Agency for Meteorology Climatology and Geophysics (BMKG),
Pradana RP, Fajary FR and Nimiya H (2024), Jakarta, Indonesia, 5Atmospheric Science Research Group, Faculty of Earth Science and Technology,
Identification of optimal CMIP6 GCMs for future Institute Technology Bandung, Bandung, Indonesia, 6Graduate School of Science and Engineering,
typical meteorological year in major cities of Kagoshima University, Kagoshima, Japan
Indonesia using multi-criteria decision analysis.
Front. Environ. Sci. 12:1341807.
doi: 10.3389/fenvs.2024.1341807

COPYRIGHT Many studies often use a single global climate model (GCM) across multiple cities
© 2024 Bhanage, Lee, Cabrera, Kubota,
Pradana, Fajary and Nimiya. This is an open-
to develop future Typical Meteorological Year (TMY), without emphasizing city-
access article distributed under the terms of the specific selection of GCM. The present investigation employs the Analytical
Creative Commons Attribution License (CC BY). Hierarchy Process (AHP) to assess city-specific GCMs for generating future
The use, distribution or reproduction in other
forums is permitted, provided the original
TMY datasets across 29 Indonesian cities. Six GCMs from the coupled model
author(s) and the copyright owner(s) are intercomparison project phase 6 (CMIP6) were evaluated against Modern-Era
credited and that the original publication in this Retrospective Analysis for Research Applications (MERRA-2) to assess their
journal is cited, in accordance with accepted
academic practice. No use, distribution or
performance in simulating surface air temperature, precipitation, wind speed,
reproduction is permitted which does not and relative humidity. Six statistical measures were used to recognize the
comply with these terms. systematic biases. Further, AHP was applied to integrate these statistical
measures to calculate the city-specific total relative error for each
meteorological parameter. Results of total relative error show that TaiESM, 6-
Model Ensemble (6ME), NorMM, and MPI-HR were best for simulating surface air
temperature, precipitation, wind speed, and relative humidity in most cities,
respectively. TMY recognizes distinctive importance among meteorological
parameters. Thus, it is essential to reflect the parameter-specific importance
while selecting GCMs for future TMY. Hence, AHP was reapplied on total relative
errors accounting for differing weights of each meteorological parameter.
Outcomes show that TaiESM, 6ME, and MPI-HR were found suitable for
generating future TMY datasets in 18, 5, and 3 cities, respectively, while MPI-
LR, NorLM, and NorMM were recommended for Boven Digoel, Medan, and
Bengkulu cities, respectively. Using city-specific GCMs ensures precise and
cost-effective future TMY generation, assisting urban planners and
policymakers in designing environmentally sustainable buildings for
anticipated climatic changes.

KEYWORDS

analytical hierarchical process, CMIP6, Indonesia, total relative error, typical

meteorological year

Frontiers in Environmental Science 01 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

1 Introduction data. Depending upon several factors, such as parameterization

schemes, grid size, representation of the earth system, and initial
Climate change is the persistent alteration of global or local and boundary conditions, each GCM exhibits different
climate conditions over an extended period. Alterations in climate uncertainties. However, using the climatic data by integrating all
are mainly identified by measuring the variations in meteorological the GCMs may not always be suitable. The use of many models leads
factors, including temperature, precipitation, and other parameters to increased ambiguities due to low simulation accuracy and process
(Gleick, 1986; Zhang et al., 2011). According to the algorithms of ensemble members. Therefore, to generate the future
Intergovernmental Panel on Climate Change (IPCC) fifth TMY from individual GCMs, it is vital to compare each GCM or to
assessment report, from 1951 to 2012, the average global surface identify the relative performance of each GCM based on different
air temperature of the land and ocean has risen by 0.72°C (Stocker, criteria that suit the TMY generation approach.
2014). By the end of the 21st century, it will increase to 1.8°C–4°C In recent years, numerous studies have explored the intricacies
(Field, 2014). This report also applauds substantial greenhouse gas of climatic features in various regions, aiming to predict future
cuts, advocating for net zero emissions by mid-century to limit climate scenarios. These investigations involve a meticulous
global temperature increase to under 1.5°C. As reported by various evaluation of GCMs against observed or reanalysis data. For
authors, urbanization is anticipated to drive a future rise in urban air example, over Indian subcontinent, the exceptional performance
temperatures (Vinayak et al., 2021; 2022; Bhanage et al., 2023a). The of an ensemble comprising KIOST-ESM, MRI-ESM2-0, MIROC6,
urbanization process also amplifies greenhouse gas (GHG) NESM3, and CanESM5 for maximum temperature and E3SM-1-0,
emissions, making the urban environment one of the prior NESM3, CanESM5, GFDL-CM4, INM-CM5-0, and CMCC-ESM2
concerns in the present world order (Ambade et al., 2021; for minimum temperature has been reported (Rahman and Pekkat,
Gautam et al., 2021; Thapliyal et al., 2022). These predictions of 2024). In the case of Indonesia, the superiority of the multi-model
forthcoming changes in the climate bring noteworthy consideration ensemble in simulating extreme precipitation events was noted
to the research topics that highlight the paths to cut down on (Kurniadi et al., 2023). A recent research over Guatemala located
GHG emissions. in central American region noted that MRI-ESM-2.0 and IPSL-
Buildings now contribute 33% of GHG emissions and 40% of CM6A-LR GCMs as the top performer for simulating surface air
global energy use, projected to rise by 61% by 2,100 (Santamouris, temperature and precipitation (Rivera, 2023). Research carried out
2016; Teske et al., 2021). To restrict building energy consumption, in 2023, evaluated the performance 27 GCMs derived from
the design of a building should pass the building energy CMIP6 over Southeast Asian territory, identified TaiESM as a
consumption test to check whether the design of a building most exceptional performer (Pimonsree et al., 2023).
meets the criteria of energy savings. Building designs with While these studies primarily aimed to assess the GCMs’
accurate meteorological data, specifically Typical Meteorological proficiency in capturing large-scale spatial and temporal climate
Year (TMY) datasets, is essential in this testing. TMY dataset is a patterns, including regional temperature trends and precipitation
set of meteorological data representing the typical hourly weather variability, it is essential to emphasize that when evaluating GCMs
conditions for a specific location over 1 year. These datasets for constructing future TMY data, a more focus on the local-scale
represent local climate conditions for simulations. However, characteristics of the climate becomes imperative. Thus, it is crucial
future climate changes make historical and present TMYs to recognize that the approach adopted for regional-scale model
inadequate. Thus, creating TMYs considering future climate evaluations may not necessarily be suitable for the meticulous
possibilities is vital. The most adopted approach for preparing preparation of future TMY datasets.
future TMY datasets is morphing (Dias et al., 2020). The In past, numerous studies have been conducted to prepare the
morphing method mainly relies on delta change in monthly future TMY dataset from the GCMs (Haase et al., 2011; Nik, 2016;
means of the weather data to illustrate future weather; still, Huld et al., 2018; Troup et al., 2019; Yuan et al., 2022). However, to
various studies reveal that weather variability and extreme events generate the future TMY, the criteria used to select GCMs differ in
are not replicated by variations in monthly means (Dias et al., 2020). different studies. In 2011, after statistical evaluation of surface air
Apart from that, the morphing method also induces the effects of temperature and precipitation acquired from six GCMs, Chan
climate change independently between variables. Therefore, the (2011) utilized the MIROC6-2-Med to simulate future TMY data
most reliable methodology for constructing the future TMY over the entire Hongkong region. In 2022, Doutreloup et al. (2022)
dataset is to dynamically downscale the global climate models generated the future TMY dataset over Belgium using three climate
(GCMs). Compared to the morphing technique in a dynamical models, namely, BCC-CSM2-M, MPI-ESM 1.2, and MIROC6. Jiang
downscaling of GCMs, it captures a more detailed representation of et al. (2019) used the HadCM3 data to generate the future TMY over
physical processes, including atmospheric dynamics, land-surface eight different cities of the globe. Ji Wei Zhou (2022) used the
interactions, and local feedback mechanisms, contributing to a more Coordinated Regional Downscaling Experiment to estimate the
accurate representation of future climate conditions. future reference years over different Canadian cities. Bass et al.
The GCMs are numerical models that simulate the global (2018) created a TMY dataset for Maricopa, Arizona, using six
climate’s preceding, contemporary, and forthcoming conditions GCMs without evaluating their performance. Similarly,
(Alizadeh, 2022). The sixth coupled model intercomparison Manapragada et al. (2022) used the Canadian Earth System
project (CMIP6) provides a climatic dataset from more than Model to generate future TMY data over Delhi, India without
30 GCMs developed by various organizations. It is necessary to comparing different GCMs.
assess the proficiency of these GCMs in replicating essential climate However, there are some limitations to these studies especially
variables by comparing their outputs with observed or reanalysis while selecting the GCMs for the development of future TMY

Frontiers in Environmental Science 02 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

author’s knowledge, this study will be a pioneering effort that

considers the AHP-based multi-criteria decision analysis
(MCDA) approach to incorporate and aggregate the subjective
judgments associated with various meteorological and statistical
parameters that allow a more transparent decision process.
The paper is structured into six distinct sections. The first
section explores the research problem’s context, underscoring its
significance, and outlines the study’s objectives. The second
section offers comprehensive information about the study
area. The third section delves into the research design and
methodology employed in the study. The fourth section
presents the study’s outcomes, encompassing data analysis,
statistical insights, and other pertinent information related to
the findings. The fifth section is dedicated to interpreting and
comparing the results with previous studies. Finally, the sixth
section summarizes the key findings, addresses limitations, and
discusses the future scope of the study.

2 Study area
The present study has been conducted over twenty-nine major
cities located in Indonesia (Figure 1). The overall area of Indonesia
extends from the latitudes 11°S and 6°N and longitudes 95°E and
141°E. The elevation of the Indonesian region varies from 0 m to
3,505 m. Indonesia experiences a hot-humid tropical climate. The
FIGURE 1 surface air temperature in Indonesia stays relatively constant
Study Area representing (A) Climatic zones of Indonesia based on
throughout the year. The country has two seasons: the wet
Koppen -Geiger classification and (B) spatial variation in the elevation
and the location of 29 major cities in Indonesia. season, from October to April, and the dry season, from May to
September. Due to its tropical climate, the coastal plains experience
an average surface air temperature of around 28°C.
In comparison, the internal highland regions have a mean of
dataset. First, some of the above studies have used a unique GCM to 26°C, and the higher highland regions have temperatures around
develop the future TMY dataset over different cities. But, some 23°C. The country’s relative humidity is very high and fluctuates
studies recently have shown that the performance of GCMs in between 70% and 90%. Indonesia experiences ample annual
simulating the different meteorological variables and phenomena precipitation ranging from the driest regions of around 500 mm
varies concerning the cities (Bhanage et al., 2023b; VR, 2023). to more than 3,000 mm in the wettest regions (Kurniadi et al., 2023).
Second, many of the above-mentioned studies do not evaluate It highlights the spatial heterogeneity in the precipitation over
the performance of GCMs before adopting them for the Indonesia. Given the spatial variabilities in temperature and
development of future TMY dataset. Evaluating the performance precipitation across Indonesia, it is crucial to identify GCMs
of GCMs helps identify their strengths and weaknesses, allowing for specific to each city for generating future TMY datasets.
the selection of models that better capture the local climate
variability, which is essential for TMY dataset generation (Fildes
and Kourentzes, 2011; McSweeney et al., 2015; Ruane and 3 Materials and methods
McDermid, 2017; Chowdhury and Behera, 2019) Third, in
creating TMY data, it is imperative to recognize that different 3.1 Data
meteorological parameters have varying levels of importance. Yet,
the studies mentioned above did not consider the respective The CMIP6 program includes historical (1850-2014)
significance of these parameters while selecting an appropriate simulations based on observations and external forcings such as
GCM for future TMY development. volcanic eruptions, solar variability, changes in GHGs, and aerosol
To overcome these limitations, the principal objective of this concentrations (Eyring et al., 2016). In the CMIP6 stage, the
study is to develop a new framework for identifying GCMs tailored scenarios of representative pathways (RCPs) i.e., RCP 2.6,
to the local climatic characteristics of each of Indonesia’s 29 cities to RCP4.5, RCP 6.8 and RCP 8.5 of the CMIP5 have been apprised
generate future TMY datasets. This study’s distinctive feature lies in to shared socioeconomic pathways (SSPs) i.e., SSP1-2.6, SSP2-4.5,
its pioneering method for GCM selection, as it recognizes the diverse SSP4-6.0, and SSP5-8.5 respectively (Arnell et al., 2019; Gidden et al.,
levels of significance attributed to individual meteorological and 2019; Shiru and Chung, 2021). These SSP scenarios provide a
statistical parameters. Therefore, while identifying the city-wise framework for climate change research by describing plausible
optimal GCMs for preparing future TMY data, according to the alternative futures of human development and greenhouse gas

Frontiers in Environmental Science 03 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

TABLE 1 Description of selected GCM data.

Sr.No. GCM-ID Abbreviation Spatial resolution Period Institution

1 NorESM2 MM NORMM 1×1 Jan 1980–December 2014 Norwegian-Climate Centre, Norway

2 NorESM2 LM NORLM 2.5 × 2.5 Jan 1980–December 2014 Norwegian-Climate Centre, Norway

3 MPIESM 1–2 HR MPI-HR 1×1 Jan 1980–December 2014 Max Planck Institute of Meteorology, Germany

4 MPI-ESM 1–2 LR MPI-LR 2.5 × 2.5 Jan 1980–December 2014 Max Planck Institute of Meteorology, Germany

5 RCEC. TaiESM 1×1 Jan 1980–December 2014 Research Centre for Environmental Changes, Taiwan,
China
TaiESM1

6 AWI-CM- AWI 1×1 Jan 1980–December 2014 The Alfred Wegener Institute, Germany
1-1 MR

emissions. Each SSP scenario represents a different narrative about

future socioeconomic conditions and policies. The SSPs encompass
five different potentials: (a) a scenario characterized by growth and
equality with a focus on sustainability (SSP1); (b) a condition where
trends closely align with historical patterns (SSP2); (c) a world
noticeable by resurgent nationalism and fragmentation (SSP3); (d) a
scenario of intensifying global inequality; and (e) a state of rapid and
unrestrained growth in both energy consumption and economic
output (SSP5) (Shiru and Chung, 2021). The major goal of the
CMIP6 program is to enhance the accuracy and reliability of climate
models by incorporating improvements in emissions scenarios,
land-use representation, physical process modeling, and model
parameterization. This comprehensive approach allows for a
more robust exploration of future climate conditions (Eyring
et al., 2016). In the ongoing research, six historical GCMs were
chosen to evaluate surface air temperature, precipitation, wind
speed, and relative humidity. These GCMs were selected from
the CMIP6 dataset, spanning the time frame from 1980 to 2014.
The study utilizes MCDA-based statistical methods to perform a
comparative evaluation between the chosen GCMs and
reanalysis datasets.
The evaluation period from 1980 to 2014 was chosen because (i)
the period of 1980–2014 serves as a reference period for climate
agreement as it covers a large period of the recent past, (ii) the
growing GHGs and anthropogenic influences marked this period of
1980 onwards, and (iii) as compared to earlier period (<1980), very FIGURE 2
General flow of methodology adopted for identifying city-
few large-scale disturbances and the number of volcanic eruptions specific suitable GCMs for generating the future TMY dataset.
have increased between post-1980 period (Doocy et al., 2013). For
this evaluation, 6 GCMs were selected based on the following
criteria: (i) The GCMs must comprise the historical dataset of
1 n
surface air temperature, precipitation, relative humidity, and (6ME) GCMi (1)
n i1
wind speed from 1980 to 2014, (ii) the identical model should
incorporate the climatological data from 2015 to 2,100 under SSP Further, this data has been compared with the monthly
5–8.5 scenario at 6 hourly interludes, and (iii) the model must have a surface air temperature, precipitation, and relative humidity
forcing data from the r1i1p1f1 (first realization). Selection of the first data obtained from the Modern-Era Retrospective Analysis for
realization ensures a fair and unbiased comparison across all GCMs, Research Applications (MERRA), Version 2 (Bosilovich, 2015).
while the selection of the SSP 5–8.5 scenario is motivated by the This dataset is established by incorporating ground-satellite
chance it provides to gain insights into the potential extreme impacts observations and numerical model data. The spatial resolution
of climate change. Detailed information on chosen GCMs is of the MERRA2 dataset is 0.50 × 0.65. The flow of methodology is
provided in Table 1. illustrated in Figure 2. After the acquisition of the MERRA-2 and
Subsequently, to assess the integrated performance, the selected selected CMIP6 GCMs, they were regridded to 1 × 1. This
GCMs, multi-model ensemble dataset (6ME), was calculated using regridding process was performed to align the spatial
Eq. 1 where i = different shortlisted GCMs. resolution of the GCMs with the dataset, ensuring a consistent

Frontiers in Environmental Science 04 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

TABLE 2 Details of six statistical measures.

Statistical measure Formula

Correlation Coefficient (Oi −O)(M−
√

M)
√

Σ(Oi−O)
2 2
Σ(Mi −M)

Standard Deviation n
i1 (Mi−u)
2
n

y1 Ty
N
Mean Annual
N

n (Mi ) − (Oi )
Mean Bias Error 1

Mean Seasonal Cycle Amplitude (Mmax ) − (Mmin )

n
Root Mean Square Error i1 (Mi −Oi )2
n

Note: Mi = Model estimated values, Oi = Observed Values, n = Total no. of values in dataset, = The mean of the population (observed or Model), Mmax = mean warmest/wettest/windiest/most
humid month of model for years 1980–2014, Mmin = mean coldest/driest/least windy/least humid month of model for years 1980–2014, Ty = Annual surface air temperature/precipitation/
relative humidity/wind speed of the year y, N = Total no. of years.

spatial framework for comparison. Furthermore, city-wise simulated/predicted and reference datasets. The lower (higher)
statistical analysis has been conducted to evaluate the values of RMSE indicate the smaller (greater) error from the
performance of surface air temperature, relative humidity, simulated/modeled values.
precipitation, and wind speed derived from the MERRA-2 and
CMIP6 GCMs, including the 6ME.
3.3 Relative weights by the analytical
hierarchy process (AHP)
3.2 Statistical performance metrics
The quantitative assessment for the suitable GCMs for the
Several statistical performance metrics are generally utilized to generation of future TMY depends on multiple criteria, such as
evaluate the accuracy and reliability of the GCMs. Here, for each surface air temperature, precipitation, wind speed, and relative
city, by comparing the MERRA-2 and GCMs, including the 6ME, six humidity. Also, statistical measures such as bias in SA, MBE, RMSE,
statistical metrics were calculated to analyze surface air temperature, CC, and bias in SD were used to generate the total relative error. This
precipitation, relative humidity, and wind speed as follows (see also study employed the AHP method twice to ascertain the relative weights
Table 2 for their formula): assigned to individual statistical measures and the total relative error
The correlation coefficient (CC) depicts the strength of the calculated for each meteorological parameter.
association among the comparative dynamics of two variables; the Subsequently, the relative importance of each statistical measure
values of the CC range from −1 to 1. The values closer to 0 (1 or −1) (total relative error of every meteorological parameter) was established
show the weaker (stronger) association between the two variables. by administering questionnaires to a panel of 10 (8) experts. Detailed
Standard deviation (SD) elucidates the dataset’s variation or information regarding each questionnaire can be found in the
dispersion. A lower standard deviation indicates that the values in Supplementary Material S1. The first step of the AHP process is the
the data are closer to the mean of the dataset. In contrast, higher generation of the pairwise comparison matrix, where each criterion was
standard deviation values imply that the dataset’s values spread wider. compared to another using the Saaty Scale (Saaty, 1980) (see Table 3). In
Mean annual (MA) can calculate the sum of each mean annual this study, the scale ranges between 1 and 5 based on the number of
divided by the total number of years. It is mainly used to estimate the norms used to generate the total relative error and 1 and 4 for each
shifts in the average annual values of each meteorological variable meteorological parameter that can generate future TMY. The second
over the 29 cities. step includes dividing each value in the matrix (Cij) by the sum of
Mean bias error (MBE) is the positive or negative difference columns in the pairwise comparison matrix to produce the normalized
between the dataset acquired from the reference and CMIP6 GCMs. value (Xij), as shown in Eq. 2.
MBE outcomes reveal whether the GCM simulations are warm or
n
cold-biased compared with the reference data.
X ij C ij C ij (2)
Mean seasonal cycle amplitude (SA) mainly highlights the i1
difference between extremely hot and cold months that can be
used to assess the gravity of alterations in surface air temperature, The third step is to generate the weight (Wij) of each criterion,
precipitation, wind speed, and relative humidity over the 29 cities which is calculated by dividing the normalized value (Xij) by the total
considered for this study. number of criteria (n), as shown in Eq. 3.
Root mean square error (RMSE) commonly measures the n
disparity between values simulated/predicted by a GCM and the W ij X ij n (3)
reference. In other words, RMSE measures the error between i1

Frontiers in Environmental Science 05 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

TABLE 3 Saaty Scale of relative importance and its corresponding description.

Scale Judgment Description

1 Equal importance Two criteria contribute equal importance

3 Moderate The judgment slightly favors one over the other

5 Strong The judgment strongly favors one over the other

7 Very strong The judgment very strongly favors one over the other

9 Extremely important The judgment highly favors one over the other

2,4,6,8 Intermediate values When compromise needed

TABLE 4 Random index matrix per number of criteria in decision-making. correlations (where Xobs necessarily equaled 1). Furthermore,
relative error E*I,jis defined as
Number of criteria 2 3 4 5
RI 0.00 0.58 0.90 1.12 Ei,j − min Ei,j
E*I,j (7)
max Ei,j − min Ei,j

The last part of the AHP process is generating the consistency Then, for the calculation of total relative error, a weighted sum
ratio (CR). The CR is used to determine whether each criterion’s has been carried out concerning relative error estimated across all m
weight is acceptable. It is acceptable if the CR value is less than or equal metrics in Eq. 8
to 10% (0.10); otherwise, it is unacceptable. There are three sub- m
procedures to generate the CR. The first is to calculate the consistency EI,tot
*
J1 W j p E*I,j (8)
measure (CM). CM can be calculated by multiplying the pairwise
matrix with the weight (Wij) and dividing the result by the weighted to get the total relative error E*I,tot per model. Wj is the weight of
sum with the criterion weight. The consistency index (CI) is then each statistical metric j derived by using the AHP method. The
calculated using Eq. 4, where the λmax is the average of the CMs. Lastly, values obtained for E*I,tot are unitless. Finally, the models are
the CR is derived using Eq. 5, where the relative index (RI) value is ranked by their respective total relative error. Lower (higher)
shown in Table 4. values of the E*I,tot represents the higher (lower) suitability of the
GCMs for an individual meteorological parameter over a
CI (λmax – n) (n – 1) (4) specific city.
CR CIRI (5) In the past, it was noted that the relative importance of
meteorological parameters varies in the development of TMY
datasets (Cebecauer and Suri, 2015; Li et al., 2021). Based on the
methodology mentioned in earlier sections, weights for each
3.4 City-specific ranking of the GCMs climatic parameter were established for generating future TMY
datasets (Wtmy ). Further, to determine city-specific rankings of
For all the GCMs, the city-specific total relative error was GCMs for future TMY creation, E*I,tot Values were calculated for
calculated for each meteorological parameter. It was calculated by each meteorological parameter, and GCM for each city was
combining the five statistical measures computed in the above combined with the corresponding. Wtmy using Eq. 9. The
section. It reflects various attributes of model performance, such outcomes of Eq. 9 yields unitless ranking scores depicted by
as biases, CC, and temporal changes. This thorough assessment (E*I,rank ), enabling the determination of city-specific suitability of
enables a more extensive examination of GCMs. Calculating the GCM for the future development of TMY. The resulting non-
total relative error to individual models based on performance dimensional values, closer to 0 (1), indicate higher (lower)
enables an apparent evaluation and recognizes GCMs suitability of a GCM for developing future TMY datasets for a
demonstrating comparatively better or poorer performance. The specific city.
details of the methodology employed for the calculation of total
relative error are as follows: m
To determine the total relative error of the GCMs, we have E*I,rank J1 W tmy p E*I,total (9)
incorporated all statistical metrics and assigned varying
weights to each metric, which were obtained through the
AHP. For a given model i and metric j, we first defined an
error EI,j as 4 Results
Ei,j Xobs,j− Xi,j (6) 4.1 City-wise GCM performance

where Xobs and Xi are the observed and simulated ensemble mean In statistical assessment for surface air temperature,
metrics, respectively, and the application of Eq. 6 included precipitation, relative humidity, and wind speed, simulated

Frontiers in Environmental Science 06 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

FIGURE 3
City-wise variation of different statistical metrics computed for surface air temperature across 29 cities in Indonesia (A) MA, (B) MBE, (C) CC, (D)
RMSE, (E) SA, and (F) SD.

monthly climatic data derived from 6 GCMs and one ensemble were surface air temperature obtained from the MERRA-2 and GCMs,
compared with the MERRA-2 reanalysis data (1980-2014). Using including the 6ME, spans 290–302 K (Figure 3A). Based on the
different statistical measurements offers unique insights and helps reanalysis data, Sumenep (Wamena) City recorded the highest
evaluate different aspects of each GCM, allowing for a more (lowest) mean annual surface air temperature among all cities. In
robust analysis. the case of Sumenep (Wamena) City, the magnitude of MA surface
air temperature derived with MPI-LR (MPI-HR) exhibited close
4.1.1 Surface air temperature similarity when compared with the reanalysis dataset. The MBE of
Figure3A–F represents the six statistical measures calculated for surface air temperature calculated from six GCM and 6ME varied
the surface air temperature over 29 cities of Indonesia. Here, MA between −3–5 K (Figure 3B). Considering the 6ME, out of 29 cities,

Frontiers in Environmental Science 07 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

FIGURE 4
City-wise variation of different statistical metrics computed for precipitation across 29 cities in Indonesia (A) MA, (B) MBE, (C) CC, (D) RMSE, (E) SA,
and (F) SD.

20 (9) cities show warm (cold) biases in mean annual surface air values of CC range from 0.3 to 0.85. Regarding CC, the performance
temperature. Among the 29 cities analyzed using the 6ME, 20 exhibit of the 6ME was dominant in 20 cities. Further, the RMSE values
warm biases in mean annual surface air temperature, while 9 display computed by comparing each GCM and reanalysis dataset are
cold biases. However, for other GCMs, biases vary depending on illustrated in Figure 3D. For all the cities, the RMSE values range
their specific tendencies. Regarding mean bias error, among from 0.4 to 5.4 K. Considering the RMSE, 6ME (TaiESM)
29 cities, 6ME was deemed suitable for 2 cities. performance was better than that of 13 (9) cities. Figure 3E
On the other hand, TaiESM exhibited exceptionally well over illustrates the city-wise alterations in SA. The SA derived from
11 cities in Indonesia. Figure 3C illustrates the variation in CC six GCMs, including 6ME, fluctuates between 1.97 and 8.67 K.
estimated for surface air temperature across different cities. The Compared with the MERRA-2 dataset, other GCMs tend to

Frontiers in Environmental Science 08 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

FIGURE 5
City-wise variation of different statistical metrics computed for relative humidity across 29 cities in Indonesia (A) MA, (B) MBE, (C) CC, (D) RMSE, (E)
SA, and (F) SD.

underestimate the values of the SA. The SD estimated from six According to evaluations based on mean bias error, MPI-HR was
GCMs, 6ME, and reanalysis datasets falls within a range of the top-performing model among all the GCMs.
0.26–1.95 K (Figure 3F). Compared to MERRA-2, the SD values MBE for mean annual precipitation fluctuates
obtained from other GCMs and 6ME are generally lower. from −1,000–2,500 mm/year. A correlation-based analysis was
conducted for the monthly averaged precipitation data derived
4.1.2 Precipitation from the GCMs, 6ME, and MERRA-2 datasets from 1980 to
Figure 4A represents the city-specific MA for precipitation derived 2014. Figure 4C demonstrates the city-specific CC. Subsequently,
from 6ME, 6 GCMs, and reanalysis data. MA precipitation obtained the values of the CC vary between −0.14 and 0.77.
from all the sources lies within a range of 1,230–3,008 mm. Further, An analysis associated with CC indicates that the 6ME
MBE was calculated for average annual precipitation (Figure 4B). performed exceptionally well across 22 cities. Figure 4D displays

Frontiers in Environmental Science 09 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

FIGURE 6
City-wise variation of different statistical metrics computed for for wind speed over 29 different cities of Indonesia during 1980–2014 (A) MA, (B)
MBE, (C) CC, (D) RMSE, (E) SA, and (F) SD.

the RMSE range computed for mean monthly precipitation, ranging (5 cities). Across all the cities, the SD estimated from all the
from 3 to 8 mm/day. Among the various GCMs, the 6ME exhibits sources lies between 2 and 7 mm/day (Figure 4F). Subsequently,
the lowermost RMSE across all 21 cities, outperforming other the trend of the SD obtained from both TaiESM and the 6ME
models like TaiESM (MPI-HR) and AWI, demonstrating the closely resembles the actual SD trend observed in the
smallest RMSE values for 3 and 2 cities, respectively. reanalysis datasets.
In the case of precipitation, the SA ranges between 10 and
30 mm/day are depicted for each city (Figure 4E). Figure 4E 4.1.3 Relative humidity
depicts that for SA, NorMM demonstrated the best performance Figure 5A indicates the MA relative humidity. Across
over seven cities, followed by the MPI-HR (6 cities) and AWI 29 cities, the average annual relative humidity estimated from

Frontiers in Environmental Science 10 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

TABLE 5 Normalized matrix with the corresponding weights of each 4.2 Total relative error
criterion for the total relative error. The λmax and CI values are 5.32 and
0.07949, respectively.
Across 29 cities, the total relative error was calculated for
Criteria C1 C2 C3 C4 C5 Weights four meteorological parameters. It was calculated using the AHP
Bias in SA (C1) 0.11 0.09 0.14 0.06 0.21 0.12 method with different statistical measures. To obtain the weights
for different statistical measures, an online survey was
MBE (C2) 0.09 0.08 0.14 0.06 0.05 0.08
conducted in which more than four expert member opinions
RMSE (C3) 0.27 0.21 0.35 0.38 0.41 0.33 were considered through a survey sheet. Detailed information
CC (C4) 0.44 0.33 0.23 0.25 0.16 0.28
about the survey outline can be found in Supplementary
Material S1. The survey outcomes are depicted in Table 5
Bias in SD (C5) 0.09 0.29 0.14 0.25 0.17 0.19 with a CR value of 0.071. According to expert opinions, the
SUM 1.00 1.00 1.00 1.00 1.00 1.00 RMSE and CC are the most important, followed by bias SD, bias
in SA, and MBE, which are the least important. The total relative
error is computed using the weights. The total relative error
measures the closeness of simulated data to the reanalysis data.
all sources ranges from 70% to 92%. The MA relative humidity, The total relative error values near zero indicate the good
resulting from six GCMs, including the 6ME, exhibits a stronger performance of the GCMs. For surface air temperature, the
association with the MA relative humidity derived from total relative error values for all the GCMs, including the
MERRA-2 data. 6ME, vary between 0.1 and 0.8 across 29 major cities of
Figure 5B illustrates the MBE in average annual relative Indonesia (Figure 7A).
humidity fluctuates between −21.58% and 7.69%. For each city, The simulated surface air temperature obtained from TaiESM
Figure 5C represents the outcomes of the correlation analysis was determined to have the lowest total relative error among
conducted between the simulated and MERRA-2 datasets of 13 cities, making it the top-performing model. Following closely
monthly relative humidity. It shows that the CC varies behind, the MPI-LR GCM exhibited superior performance over
from −0.22 to 0.84. Based on the correlation analysis, the 6ME seven cities. In Wamena City, MPI-HR was identified as the model
was identified as the best performer for 19 cities. with the lowest total relative error (0.054) for surface air
Figures 5D, E represent the values of RMSE and SA, respectively. temperature.
The RMSE breadth is from 3.52% to 21.54%, while the SA varies On the contrary, NorMM exhibited the poorest performance,
from 6% to 52.37%. rendering it unsuitable for any individual city due to its
Figure 5F depicts the city-wise tendencies in the SD estimated unsatisfactory results. The city-wise total relative error for
from all the data sources. Among all the GCMs, The SD produced by precipitation is shown in Figure 7B, and the 6ME performed
NorMM, NorLM, and MPI-HR was found to be appropriate for six exceptionally well for precipitation across 15 cities. The following
cities each. 6ME, TaiESM, MPI-HR, and NorMM were found to be appropriate
models for precipitation in 9, 4, and 1 city, respectively. However,
4.1.4 Wind speed the worst performers for precipitation were found to be AWI, MPI-
Figure 6A demonstrates the variation in MA wind speed LR, and NorLM.
calculated using multiple GCMs, 6-ME, and a reanalysis The statistical evaluation of total relative error for relative
dataset. For 29 cities, it varies from 0.44 m/s to 6.66 m/s. humidity revealed that NorMM, MPI-HR, 6ME, TaiESM, AWI,
Figure 6B represents the MBE associated with the MA wind and the NorLM demonstrated their suitability across 6, 6, 5, 5, 2, and
speed derived from 6ME and GCMs fluctuating 2 cities in Indonesia, respectively (Figure 7C).
between −1.25 m/s to 4.98 m/s. Figure 6C provides the Regarding wind speed, the analysis shows that MPI-HR is
outcomes of correlation analysis—further, the values of CC suitable for 19 cities and is the best-performing model. As shown
range from −0.38 to 0.72. Analysis of the CC indicates that in Figure 7D, NorLM and the 6ME were found to be adequate for
among all cities, the performance of the 6ME was superior to 6 and 2 cities, respectively. In addition, regarding wind speed, AWI
16 cities, followed by the MPI-HR, which was found to be suitable and TaiESM were found to be appropriate over the Wamena and
over seven cities. Figure 6D shows the city-wise alterations in the Maluku Tenggara, respectively.
RMSE. The RMSE values span between 0.26 m/s to 5.22 m/s.
Based on the RMSE, MPI-HR was the performing GCM over
14 major cities in Indonesia. The fluctuations in seasonal cycle 4.3 City-specific ranking
amplitude of the mean monthly wind speed data are presented in
Figure 6E. Considering the SA, the performance of AWI is The generation of the future dataset of TMY is practically
exceptionally well over the five cities, whereas 6ME was found very complex, and it is tedious to downscale different
to be unfit for any location. In the case of SD, for all the cities, meteorological parameters from different GCMs for a specific
values vary from 0.08 m/s to 2.07 m/s (Figure 6F). Further, it has city, especially for non-climate scientists. Therefore, to identify
been observed that all the GCMs replicate the city-wise trend of the suitable city-specific GCM for future TMY generation, we
standard deviation derived from the MERRA-2 dataset. have used the AHP method again by implementing the weighting

Frontiers in Environmental Science 11 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

FIGURE 7
Total relative error depicting city-specific differences in the performance of each meteorological parameter (A) surface air temperature, (B)
precipitation, (C) relative humidity, and (D) wind speed.

factors associated with different meteorological variables, as ranking scores derived from AWI and MPI-LR range from 0.30 to
shown in Table 6. According to expert opinion, the most 0.75, whereas ranking scores derived from other GCMs,
important factors are temperature and relative humidity, including the 6ME, vary from 0.18 to 0.66. According to the
followed by wind speed and precipitation. The weights of city-specific ranking results, AWI is unsuitable for generating
these weather variables will be used to calculate the future TMY datasets for any city. These outcomes also reveal that
appropriate GCMs in each city. among 29 cities, TaiESM (6ME) can be used over 18 (5) cities to
Figure 8 exhibits the city-wise optimal GCM for the generate the future TMY dataset. GCMs like MPI-LR, NorLM,
generation of future TMY. The lower (higher) ranking score and NorMM can be applied to cities such as Boven Digoel,
indicates the higher (lower) suitability of the GCM for generating Medan, and Bengkulu. Moreover, for generating future TMY
future TMY data over specific cities. The city-specific ranking data, MPI-HR was identified as appropriate for Wamena, Medan,
scores for each GCM range from 0.2 to 0.7. For all the cities, the and Depati cities. The list of GCMs that are appropriate for

Frontiers in Environmental Science 12 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

TABLE 6 Normalized matrix with the corresponding weights of each number of cities in simulating the surface air temperature,
criterion for the meteorological variables. The λmax, CI, and CR values are
4.01, 0.00389, and 0.004, respectively. precipitation, relative humidity, and wind speed, respectively.
Additionally, our analysis associated with total relative error
Criteria C1 C2 C3 C4 Weights indicates that different GCMs’ performance varies with the cities
Surface Air Temperature (C1) 0.47 0.41 0.48 0.48 0.46 and meteorological variables. Similar findings have been reported in
prior studies, corroborating the outcomes of the total relative error
Precipitation (C2) 0.13 0.11 0.11 0.09 0.11
(Firpo et al., 2022; Hemanandhini and Vignesh, 2023). Furthermore,
Wind Speed at 10 m (C3) 0.17 0.18 0.18 0.19 0.18 the outcomes of total relative error, together with the results of the
Relative Humidity (C4) 0.23 0.29 0.23 0.24 0.25
studies mentioned above, highlight that the performance of GCMs is
inconsistent across cities or meteorological variables. This confirms
SUM 1.00 1.00 1.00 1.00 1.00 that identifying city-specific, suitable GCMs will yield more precise
and location-specific meteorological data. This, in turn, enables the
generation of TMY datasets that can accurately reflect the variability
generating future TMY data specific to each city is given in local climate.
in Table 7. To date, multiple authors have undertaken evaluations of the
performance of CMIP6 G GCMs within the Indonesian and
South Asian region context (Almazroui et al., 2020; Iqbal
5 Discussion et al., 2021; Liu et al., 2023). These evaluations have primarily
aimed to identify GCMs with high precision in simulating
In the present study, the analysis of statistical measures for each meteorological parameters. This collective effort has been
meteorological variable, obtained by comparing the meteorological instrumental in advancing our understanding of the
parameters derived from each GCM with reanalysis data, reveals a capabilities and limitations of these models in capturing the
substantial variation in the performance of individual GCMs for complex meteorological patterns of Indonesia and other parts
each meteorological parameter. Notably, a GCM may exhibit strong of the South Asian region.
performance for one statistical measure while showing weaker While previous assessments have laid the groundwork by
performance for another simultaneously. Therefore, expert establishing the overall performance of GCMs, we have honed in
judgment was used to allocate the distinct weights to each on the practicality of these models by considering the varying
statistical measure estimated for each meteorological variable levels of importance assigned to different cities within
during the computation of the total relative error. Notably, more Indonesia. This approach allows us to identify GCMs that are
emphasis was placed on normalizing RMSE than any other accurate on a broader scale and tailored to the unique needs of
statistical measure. A similar higher prominence on normalized specific cities. Subsequently, the present study takes a significant
RMSE was also corroborated by Srinivasa Raju and Nagesh Kumar, step forward by associating the importance of each
(2014) and Jose and Dwarakish, (2022). meteorological parameter with the total relative error,
The total relative error analysis demonstrates the city-specific pinpointing suitable GCMs for future TMY dataset generation
performance of GCMs for various meteorological parameters based on city-specific needs.
(Figure 7D). Among all the GCMs, TaiESM, 6ME, NorMM/MPI- The AHP methodology employed for GCM selection in
HR, and MPI-HR exhibit the best performance over a maximum forthcoming TMY data generation offers significant

FIGURE 8
City-specific ranking of the GCMs used for generating future TMY data.

Frontiers in Environmental Science 13 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

TABLE 7 List of optimal GCMs for generating the city-specific future TMY
data.
6 Conclusion
City Suitable GCM This study utilized an AHP-based MCDA approach to identify a
city-specific suitable GCM for generating future TMY datasets over
Jambi TAIESM
29 Indonesian cities. The evaluation process involved five statistical
Palembang 6ME measures: correlation coefficient, root mean square error, mean bias
Pontianak 6ME
error, standard deviation, and seasonal cycle amplitude. These metrics
collectively assessed the model’s efficacy in-depth, providing a
Balikpapan 6ME comprehensive understanding of its accuracy, bias, and capacity to
Aceh TAIESM capture crucial temporal variations. These statistical measures were
calculated for surface air temperature, precipitation, relative humidity,
Bengkulu NORLM
and wind speed by comparing GCMs with reanalysis data. Different
Medan NORMM weights were assigned to the statistical measures based on expert
Citeko TAIESM
opinions to determine the performance of each GCM in terms of
total relative error. These weights were integrated to compute the city-
Depati MPI-HR
specific performance of each GCM and every meteorological variable.
Pongtiku 6ME The total relative error outcomes provided an understanding of the
appropriate GCM for each meteorological parameter at each location.
Wamena MPI-HR
Next, the weights for individual meteorological parameters required to
Tangerang TAIESM estimate TMY were calculated based on the expert’s opinion. Then, total
Bogor TAIESM relative errors were integrated to identify the city-specific suitable GCM
for TMY generation, considering the respective weights. The major
Minahasa TAIESM
findings of this study are.
Semarang TAIESM
(i) GCM suitability outcomes derived for the individual
Lombok TAIESM
meteorological variables and future TMY generation show
Jayapura MPIHR that regionally best performing GCM need not be good at the
Kupang TAIESM city scale.
(ii) In terms of relative error estimated for individual meteorological
Sumba TAIESM
variables indicate that among all the GCMs, TaiESM, 6Model-
Sumbawa TAIESM Ensemble, NorMM/MPI-HR, and MPI-HR exhibit the best
performance over a maximum number of cities in simulating
Surabaya TAIESM
the surface air temperature, precipitation, relative humidity, and
Sumenep 6ME wind speed, respectively.
Ketapang TAIESM (iii) The evaluation of GCMs for the generation of the future TMY
dataset shows that among 29 cities, TaiESM (6ME) can be used
Kota TAIESM
over 18 (5) cities to generate the future TMY dataset. GCMs like
Indragiri Hulu TAIESM MPI-LR, NorLM, and NorMM can be applied to cities such as
Boven Digoel, Medan, and Bengkulu. Moreover, for generating
Jakarta TAIESM
future TMY data, MPI-HR was identified as appropriate for
Palu TAIESM Wamena, Medan, and Depati cities.
Maluku TAIESM (iv) The assessment also shows TaiESM as a repeatedly suitable
GCM for generating future TMY datasets across multiple
Boven Digoel MPI-LR
cities. This finding suggests the model’s robustness and
reliability in capturing the meteorological characteristics
required for TMY generation.
advantages:(i) It assesses the relative importance of each (v) The AWI model was identified as the poorest-performing
meteorological parameter during GCM selection for TMY data among all the GCMs, proving unsuitable for any of the cities
generation. (ii) the present approach also accounts for regional under consideration.
climate differences, reflecting the real-world variations in
meteorological conditions across cities in Indonesia. This This study did not comprehensively assess all potential aspects of
approach adds value to future climate modeling efforts by CMIP6 GCM performance in simulating the climate of Indonesian
tailoring GCM selection to the specific needs of each location. cities. Instead, it focused on key climatic parameters essential for
(iii) The research contributes to the practical utility of the future generating TMY data. Additionally, due to the unavailability of
TMY data generation process by streamlining the inclusion of long-term ground observation data, the performance of all the
necessary meteorological variables from the same GCM. This GCMs was evaluated against reanalysis data in this study. Despite
simplifies the dynamic downscaling process, making it more these limitations, the findings of this research present several
efficient and cost-effective. advantages. Initially, it assists in identifying GCMs tailored to

Frontiers in Environmental Science 14 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

specific cities that excel in representing local meteorological conditions. Conceptualization, Formal Analysis, Supervision, Writing–original
This enables a focused allocation of resources for the most suitable draft, Writing–review and editing. JC: Methodology,
GCMs, streamlining the downscaling process and enhancing efficiency. Writing–original draft. TK: Conceptualization, Project
This targeted strategy reduces overall time and costs and contributes to administration, Supervision, Writing–original draft. RP: Data
a more precise and cost-effective future TMY generation. Secondly, the curation, Investigation, Writing–original draft. FR: Data curation,
study enhances the production of accurate and reliable results by Writing–original draft. HN: Supervision, Writing–original draft.
guiding the selection of GCMs closely aligned with the specific
characteristics of the local climate. Thirdly, it optimizes resource
allocation by directing researchers towards downscaling GCMs Funding
better suited to the intricacies of local conditions.
AHP is a promising method for identifying suitable GCM to The author(s) declare that financial support was received for
generate future TMY datasets. However, it does have a limitation due the research, authorship, and/or publication of this article. This
to the subjectivity involved in choosing the weights of each research was conducted by the Climate Research Group for
meteorological variable based on expert judgments (Cabrera and Lee, the Development of Standard Weather Data as part of the
2020). Fortunately, this limitation can be addressed by the consistency Development of Low-Carbon Affordable Apartments in the
ratio threshold specified by Saaty (1980). In this study, each response was Hot-Humid Climate of Indonesia Project toward Paris
carefully assessed. Responses with inconsistent results (i.e., the CR > Agreement 2030; the Science and Technology Research
10%) were not included in the final evaluation, reducing the number of Partnership for Sustainable Development (SATREPS); and
acceptable respondents. Having fewer respondents might not capture a collaboratively supported by the Japan Science and Technology
broader perspective. While having many respondents in AHP is not Agency (JST, JP-MJSA1904), the Japan International Cooperation
strictly required, having more respondents can help capture a broader Agency (JICA), Hiroshima University, Kagoshima University, the
consensus. Nevertheless, by employing AHP, this research has provided Ministry of Public Works and Housing (PUPR) of Indonesia, and
valuable insights into selecting appropriate GCMs for making realistic the Meteorological, Climatological, and Geophysical Agency
projections of the future TMY under varying climate change scenarios. (BMKG) of Indonesia.
In the future, employing recommended city-specific GCMs for
generating future TMY datasets is critical for architectural researchers
and policymakers worldwide. This practice enhances the accuracy of Conflict of interest
climate projections, offering valuable insights for informed urban
planning and design decisions. Moreover, it facilitates the The authors declare that the research was conducted in the
implementation of sustainable building practices and plays a pivotal absence of any commercial or financial relationships that could be
role in shaping effective climate policies for the future. The approach construed as a potential conflict of interest.
tested in this study provides a universal method for selecting GCMs for
preparing future TMY datasets, applicable across the globe for
comprehensive climate modeling and planning. Publisher’s note
All claims expressed in this article are solely those of the authors
Data availability statement and do not necessarily represent those of their affiliated
organizations, or those of the publisher, the editors and the
The original contributions presented in the study are included in reviewers. Any product that may be evaluated in this article, or
the article/Supplementary Material, further inquiries can be directed claim that may be made by its manufacturer, is not guaranteed or
to the corresponding author. endorsed by the publisher.

Author contributions Supplementary material

VB: Conceptualization, Data curation, Formal Analysis, The Supplementary Material for this article can be found online
Investigation, Methodology, Software, Validation, Visualization, at: https://www.frontiersin.org/articles/10.3389/fenvs.2024.1341807/
Writing–original draft, Writing–review and editing. HL: full#supplementary-material

References
Alizadeh, O. (2022). Advances and challenges in climate modeling. Clim. Change 170, of black carbon over an urban region of East India. Urban Clim. 38, 100896. doi:10.
18. doi:10.1007/s10584-021-03298-4 1016/j.uclim.2021.100896
Almazroui, M., Saeed, S., Saeed, F., Islam, M. N., and Ismail, M. (2020). Projections of Arnell, N. W., Lowe, J. A., Bernie, D., Nicholls, R. J., Brown, S., Challinor, A. J.,
precipitation and temperature over the South Asian countries in CMIP6. Earth Syst. et al. (2019). The global and regional impacts of climate change under
Environ. 4, 297–320. doi:10.1007/s41748-020-00157-7 representative concentration pathway forcings and shared socioeconomic
pathway socioeconomic scenarios. Environ. Res. Lett. 14, 084046. doi:10.1088/
Ambade, B., Sankar, T. K., Panicker, A. S., Gautam, A. S., and Gautam, S. (2021).
1748-9326/ab35a6
Characterization, seasonal variation, source apportionment and health risk assessment

Frontiers in Environmental Science 15 frontiersin.org

Bhanage et al. 10.3389/fenvs.2024.1341807

Bhanage, V., Kulkarni, S., Sharma, R., Lee, H. S., and Gedam, S. (2023a). Li, H., Huang, J., Hu, Y., Wang, S., Liu, J., and Yang, L. (2021). A new TMY generation
Enumerating and modelling the seasonal alterations of surface urban heat and method based on the entropy-based TOPSIS theory for different climatic zones in
cool island: a case study over Indian cities. Urban Sci. 7, 38. doi:10.3390/ China. Energy 231, 120723. doi:10.1016/j.energy.2021.120723
urbansci7020038
Liu, S., Raghavan, S. V., Ona, B. J., and Nguyen, N. S. (2023). Bias evaluation in rainfall
Bhanage, V., Lee, H. S., Kubota, T., Pradana, R. P., Fajary, F. R., Arya Putra, I. D. G., over Southeast Asia in CMIP6 models. J. Hydrol. 621, 129593. doi:10.1016/j.jhydrol.
et al. (2023b). City-wise assessment of suitable CMIP6 GCM in simulating different 2023.129593
urban meteorological variables over major cities in Indonesia. Climate 11, 100. doi:10.
McSweeney, C. F., Jones, R. G., Lee, R. W., and Rowell, D. P. (2015). Selecting
3390/cli11050100
CMIP5 GCMs for downscaling over multiple regions. Clim. Dyn. 44, 3237–3260. doi:10.
Bosilovich, M. G. (2015). MERRA-2: initial evaluation of the climate. Greenbelt, MD, 1007/s00382-014-2418-8
United States: National Aeronautics and Space Administration, Goddard Space Flight
Nik, V. M. (2016). Making energy simulation easier for future climate–Synthesizing
Center.
typical and extreme weather data sets out of regional climate models (RCMs). Appl.
Cabrera, J. S., and Lee, H. S. (2020). Flood risk assessment for Davao Oriental in the Energy 177, 204–226. doi:10.1016/j.apenergy.2016.05.107
Philippines using geographic information system-based multi-criteria analysis and the
Pimonsree, S., Kamworapan, S., Gheewala, S. H., Thongbhakdi, A., and
maximum entropy model. J. Flood Risk Manag. 13, 1–17. doi:10.1111/jfr3.12607
Prueksakorn, K. (2023). Evaluation of CMIP6 GCMs performance to simulate
Cebecauer, T., and Suri, M. (2015). Typical meteorological year data: SolarGIS precipitation over Southeast Asia. Atmos. Res. 282, 106522. doi:10.1016/j.atmosres.
approach. Energy Procedia 69, 1958–1969. doi:10.1016/j.egypro.2015.03.195 2022.106522
Chowdhury, P., and Behera, M. R. (2019). Evaluation of CMIP5 and CORDEX Rahman, A., and Pekkat, S. (2024). Identifying and ranking of CMIP6 - global climate
derived wave climate in Indian Ocean. Clim. Dyn. 52, 4463–4482. doi:10.1007/s00382- models for projected changes in temperature over Indian subcontinent. Sci. Rep. 14,
018-4391-0 3076–3114. doi:10.1038/s41598-024-52275-1
Dias, J. B., da Graca, G. C., and Soares, P. M. M. (2020). Comparison of methodologies Rivera, P. (2023). Climate change projections in Guatemala: temperature and
for generation of future weather data for building thermal energy simulation. Energy precipitation changes according to CMIP6 models. Model. Earth Syst. Environ.,
Build. 206, 109556. doi:10.1016/j.enbuild.2019.109556 1–19. doi:10.1007/s40808-023-01881-5
Doocy, S., Daniels, A., Dooling, S., and Gorokhovich, Y. (2013). The human impact of Ruane, A. C., and McDermid, S. P. (2017). Selection of a representative subset of
volcanoes: a historical review of events 1900-2009 and systematic literature review. PLoS global climate models that captures the profile of regional changes for integrated climate
Curr. 5. doi:10.1371/currents.dis.841859091a706efebf8a30f4ed7a1901 impacts assessment. Earth Perspect. 4, 1–20. doi:10.1186/s40322-017-0036-4
Eyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., et al. (2016). Saaty, T. L. (1980). The analytical hierarchy process, planning, priority. USA: Resour.
Overview of the coupled model intercomparison project phase 6 (CMIP6) experimental Alloc. RWS Publ.,
design and organization. Geosci. Model Dev. 9, 1937–1958. doi:10.5194/gmd-9-1937-
Santamouris, M. (2016). Cooling the buildings–past, present and future. Energy Build.
2016
128, 617–638. doi:10.1016/j.enbuild.2016.07.034
Field, C. B. (2014). Climate change 2014: impacts, adaptation and vulnerability:
Shiru, M. S., and Chung, E.-S. (2021). Performance evaluation of CMIP6 global
working Group II contribution to the fifth assessment report of the intergovernmental
climate models for selecting models for climate projection over Nigeria. Theor. Appl.
panel on climate change. Cambridge, United Kingdom: Cambridge University Press.
Climatol. 146, 599–615. doi:10.1007/s00704-021-03746-2
Fildes, R., and Kourentzes, N. (2011). Validation and forecasting accuracy in models
Srinivasa Raju, K., and Nagesh Kumar, D. (2014). Ranking general circulation
of climate change. Int. J. Forecast. 27, 968–995. doi:10.1016/j.ijforecast.2011.03.008
models for India using TOPSIS. J. Water Clim. Chang. 6, 288–299. doi:10.2166/
Firpo, M. A. F., Guimarães, B., Dantas, L. G., Silva, M. G. B. da, Alves, L. M., wcc.2014.074
Chadwick, R., et al. (2022). Assessment of CMIP6 models’ performance in simulating
Stocker, T. (2014). Climate change 2013: the physical science basis: working Group I
present day climate in Brazil. Front. Clim. 170. doi:10.3389/fclim.2022.948499
contribution to the fifth assessment report of the intergovernmental panel on climate
Gautam, S., Sammuel, C., Bhardwaj, A., Shams Esfandabadi, Z., Santosh, M., Gautam, change. Cambridge, United Kingdom: Cambridge University Press.
A. S., et al. (2021). Vertical profiling of atmospheric air pollutants in rural India: a case
Teske, S., Pregger, T., Simon, S., Naegler, T., Pagenkopf, J., Deniz, Ö., et al. (2021). It is
study on particulate matter (PM10/PM2.5/PM1), carbon dioxide, and formaldehyde.
still possible to achieve the paris climate agreement: regional, sectoral, and land-use
Meas. J. Int. Meas. Confed. 185, 110061. doi:10.1016/j.measurement.2021.110061
pathways. Energies 14, 2103. doi:10.3390/en14082103
Gidden, M. J., Riahi, K., Smith, S. J., Fujimori, S., Luderer, G., Kriegler, E., et al. (2019).
Thapliyal, J., Bhattacharyya, M., Prakash, S., Patni, B., Gautam, S., and Gautam, A. S.
Global emissions pathways under different socioeconomic scenarios for use in CMIP6: a
(2022). Addressing the relevance of COVID–19 pandemic in nature and human socio-
dataset of harmonized emissions trajectories through the end of the century. Geosci.
economic fate. Stoch. Environ. Res. Risk Assess. 36, 3239–3253. doi:10.1007/s00477-022-
Model Dev. 12, 1443–1475. doi:10.5194/gmd-12-1443-2019
02191-5
Gleick, P. H. (1986). Methods for evaluating the regional hydrologic impacts of global
Troup, L., Eckelman, M. J., and Fannon, D. (2019). Simulating future energy
climatic changes. J. Hydrol. 88, 97–116. doi:10.1016/0022-1694(86)90199-x
consumption in office buildings using an ensemble of morphed climate data. Appl.
Haase, M., Andresen, I., and Hestnes, A. G. (2011). The development of future Energy 255, 113821. doi:10.1016/j.apenergy.2019.113821
weather data files for Norway. Int. J. Clim. Chang. Impacts Responses 2, 1–24. doi:10.
Vinayak, B., Lee, H. S., Gedam, S., and Latha, R. (2022). Impacts of future
18848/1835-7156/cgp/v02i03/37327
urbanization on urban microclimate and thermal comfort over the Mumbai
Hemanandhini, S., and Vignesh, R. L. (2023). Performance evaluation of metropolitan region, India. Sustain. Cities Soc. 79, 103703. doi:10.1016/j.scs.2022.
CMIP6 climate models for selecting a suitable GCM for future precipitation at 103703
different places of Tamil Nadu. Environ. Monit. Assess. 195, 928. doi:10.1007/
Vinayak, B., Lee, H. S., and Gedem, S. (2021). Prediction of land use and land
s10661-023-11454-9
cover changes in Mumbai city, India, using remote sensing data and a multilayer
Huld, T., Paietta, E., Zangheri, P., and Pinedo Pascua, I. (2018). Assembling typical perceptron neural network-based Markov chain model. Sustain 13, 471. doi:10.
meteorological year data sets for building energy performance using reanalysis and 3390/su13020471
satellite-based data. Atmos. (Basel) 9, 53. doi:10.3390/atmos9020053
Vr, L. (2023). Performance evaluation of CMIP6 climate models for selecting a
Iqbal, Z., Shahid, S., Ahmed, K., Ismail, T., Ziarh, G. F., Chung, E.-S., et al. (2021). suitable GCM for future precipitation at different places of Tamil Nadu. Environ. Monit.
Evaluation of CMIP6 GCM rainfall in mainland Southeast asia. Atmos. Res. 254, Assess. 195, 928. doi:10.1007/s10661-023-11454-9
105525. doi:10.1016/j.atmosres.2021.105525
Yuan, J., Huang, P., and Chai, J. (2022). Development of a calibrated typical
Jose, D. M., and Dwarakish, G. S. (2022). Ranking of downscaled CMIP5 and meteorological year weather file in system design of zero-energy building for
CMIP6 GCMs at a basin scale: case study of a tropical river basin on the South performance improvements. Energy 259, 125031. doi:10.1016/j.energy.2022.125031
West coast of India. Arab. J. Geosci. 15, 120–123. doi:10.1007/s12517-021-09289-0
Zhang, X., Alexander, L., Hegerl, G. C., Jones, P., Tank, A. K., Peterson, T. C.,
Kurniadi, A., Weller, E., Kim, Y.-H., and Min, S.-K. (2023). Evaluation of coupled et al. (2011). Indices for monitoring changes in extremes based on daily
model intercomparison project phase 6 model-simulated extreme precipitation over temperature and precipitation data. Wiley Interdiscip. Rev. Clim. Chang. 2,
Indonesia. Int. J. Climatol. 43, 174–196. doi:10.1002/joc.7744 851–870. doi:10.1002/wcc.147

Frontiers in Environmental Science 16 frontiersin.org