Spatial and Temporal Analysis of Climatic Zones in Kazakhstan Using Google Earth Engine

Abstract

Kazakhstan, located in Central Asia, is experiencing faster warming than the global trend, making it an important region regarding the study of how climate change is affecting climatic zones. This research aims to identify projected shifts in Köppen–Geiger climate zones under high-emission Shared Socioeconomic Pathway (SSP) 5-8.5 climate scenarios. The Köppen–Geiger climate classification system is a practical tool that effectively captures climate types based on just two variables: temperature and precipitation. Monthly temperature and precipitation data from Climatic Research Unit (CRU,) ERA5-Land, and Coupled Model Intercomparison Project Phase 6 (CMIP6) ensembles from 1951 to 2100 were used to generate climatic zone maps. CMIP6 models were evaluated against meteorological station data and ERA5-Land, with bias metrics used to identify the best-performing models for temperature and precipitation in Kazakhstan. Based on these results, two inter-model datasets were developed and used to generate Köppen–Geiger climate maps for high-emission scenarios for the 2061–2100 time period. This research resulted in two key outcomes. First, to facilitate this analysis, a Google Earth Engine (GEE) application was developed as an open accessible tool for dynamic visualization of Köppen–Geiger climate maps. Second, projected maps based on CMIP6 SSP5-8.5 scenario projections indicate that southern Kazakhstan may shift to BSh (Hot Semi-Arid) and Csa (Mediterranean) climates, and the southwest region of the country is projected to shift to a BWh (Hot Desert) climate. These projected Köppen–Geiger climate maps contributed to climate adaptation efforts by identifying regions at risk of desertification and aridification. This study provides a comprehensive analysis of climate zone transformations in Kazakhstan and offers a practical scalable geovisualization tool for monitoring climate change impacts. This allows users easy access to climate-related information and insights into data processing procedures.

1. Introduction

Prominent reports indicate that the territory of Kazakhstan is experiencing environmental emergencies with land desertification being the primary issue. Issanova et al. (2020) report that Kazakhstan is experiencing intensive land degradation [1]. Another study by Hu et al. (2020) suggests that approximately 76.1% of Kazakhstan is sensitive to desertification [2]. Other works support these findings, highlighting that various regions of Kazakhstan are vulnerable to desertification and aridification. For example, Amirkhanov et al. (2025) also discuss the increasing vulnerability to these processes in the West and East of the country [3]. According to the Annual Bulletin on Climate Change and Condition Monitoring, the National Hydrometeorological Service of Kazakhstan “Kazhydromet” stated that—based on data for 2024—climate warming and air temperature rise in Kazakhstan is rising faster than the global trend. The global warming trend rate is 0.19 °C degrees for every ten years, while the rate of mean annual temperature increase in Kazakhstan is 0.36 °C. Each subsequent decade is becoming warmer, with 2023 being a record year since 1941, exceeding the climatic norm (1961–1990) by 2.58 °C degrees. While there is a rise in temperature, precipitation patterns show insignificant change [4]. Kazakhstan is located in Central Asia and it experiences an extreme continental climate with hot summers and cold winters. The most common causes of emergencies in Kazakhstan are strong winds, floods, mudflows in some regions, abnormal cold, abnormal heat, drought, heavy rainfall, blizzards, ice, and hail [5].

Kazakhstan’s vast territory and distinct seasons make it an ideal case study for assessing spatiotemporal climate variations using geospatial technologies. While meteorological data exists, its complexity limits accessibility for non-specialists, necessitating intuitive visualization tools. It is much more insightful to show how the climate zones change with time. For better understanding of how climate change affected Kazakhstan, this study has significance in terms of visualizing and analyzing changes in the boundaries of climate zones in the region. Therefore, this research aims to provide a comprehensive examination of climate classes by using the Köppen–Geiger classification system and mapping climatic zones using Google Earth Engine (GEE).

The Köppen–Geiger classification system was adopted due to its widespread use and reliance on temperature and precipitation thresholds, making it ideal for climate change studies. It categorizes the climate into five main classes based on temperature distribution and subdivides each main class by adding precipitation values [6]. There are scientific works that present Köppen climate maps in various resolutions [6,7,8,9] and for different regions [10,11,12]. However, a research gap exists in the development of algorithms for generating such maps. The first climate maps were introduced in the 19th century [13,14,15,16,17] and there are still ongoing efforts to reproduce Köppen climate maps using the most up-to-date data and modern technology. An example of such maps can be found at https://services.arcgis.com/nzS0F0zdNLvs7nc8/arcgis/rest/services/IPCC_Scenarios_2001_to_2025__separate_time_slices__WFL1/FeatureServer (accessed on 30 October 2025) as ArcGIS (https://www.arcgis.com/) feature layers. Although there are efforts being made to reproduce Köppen maps using different software platforms, only a limited number of tools are available that can produce dynamic and customizable climatic maps (for example, https://koppen.earth/ (accessed on 30 October 2025)) [9]. While there are already efforts being made in this direction, they lack flexibility and interactivity (e.g., upload their own data for a particular area of interest, choose specific time period, calculate areas of climate zones).

To bridge the gap, this research seeks to contribute to the understanding of regional climate dynamics in Kazakhstan by answering the following research questions: How have climate zones in Kazakhstan changed over time and how can we project future climate zone shifts under the Shared Socioeconomic Pathways (SSPs), specifically fossil-fueled development SSP5-8.5 scenarios [18]? Furthermore, how can modern technologies, specifically GEE, be leveraged to efficiently generate climate maps with minimal resources by utilizing the full potential of available datasets?

This research is focused on cartographic methods rather than meteorological or other perspectives. Cartographic visualization plays a vital role in supporting policymakers, researchers, and enhancing public awareness by providing clear insights into the effects of climate change. Unlike statistical analyses, it can be effectively interpreted without requiring specialized knowledge of visualization methods.

Using GEE’s cloud-based geospatial platform, the analysis of yearly and seasonal variations in temperature and precipitation was performed for further visualization of Köppen–Geiger climate maps. For performing such analysis, a unique and universal code script was developed, and is capable of processing any time series data with monthly temperature and precipitation information in GEE. As a result, a GEE-based tool that generates dynamic Köppen–Geiger climate maps was developed, enabling customizable visualization of past and future climate scenarios for Kazakhstan. What makes this tool unique is its comprehensive ability to automatically generate Köppen–Geiger climate maps from multiple sources, including ERA5-Land, CRU, and CMIP6 data, for both global and region-specific datasets. For ERA5-Land and CRU, users can generate maps globally or for specific regions, while CMIP6 allows for either ensemble or individual model data with preselection of a region due to computational constraints. The tool also enables users to upload their own validated data, provided it follows necessary preprocessing steps, including unit and temporal conversions, as detailed further in the guide.

Unlike other existing tools, this developed GEE-based tool is designed to be accessible to users with minimal programming experience. End-users can generate Köppen–Geiger climate maps through an intuitive and user-friendly interface using predefined climate datasets. The end-user interface enables users to generate climate maps for any time period, with an integrated feature that calculates the area of each Köppen–Geiger zone. Furthermore, the tool supports the generation of future climate maps, with options to select SSP2-4.5 or SSP5-8.5 scenarios, providing flexibility for diverse climate projections. If users wish to apply the workflow to their own area of interest or integrate custom datasets, the tool does not need to be rebuilt from scratch; instead, only minor modifications to a limited number of script parameters are required.

This tool is openly available and both the final user interface and backend code are fully accessible for further development and customization. All scripts are accompanied by step-by-step instructions, supporting reproducibility and facilitating broader adoption by the cartographic and climate research communities. This emphasis on usability, accessibility, and reproducible cartographic workflows distinguishes this study from previous approaches, which often require advanced programming skills or extensive manual processing to reproduce or update Köppen–Geiger climate maps.

This study primarily focuses on the SSP5-8.5 high-emission scenario to explore future climate zone changes. While the SSP2-4.5 scenario is also mentioned, it is not the focus of this research, but rather serves to highlight that the tool used can generate climate zone maps for both scenarios.

In summary, the availability of such a dynamic application, which generates customizable climate zone maps by allowing users to adjust their area of interest, choose climate data layers (time periods and CMIP6 scenarios), view areas of climate zones in square kilometers, and upload validated data for small regions, provides individuals with the opportunity to broaden their knowledge and expand research frameworks. This tool enhances the ability to explore and analyze Köppen–Geiger climate zones in a more flexible, accessible, and detailed manner, offering a significant advancement over existing conventional static tools.

2. Materials and Methods

This study focuses exclusively on Kazakhstan, but the developed script is designed for global, national, and subnational applications as well. Users can adapt the script to other regions by modifying a single line of code, making it flexible for diverse climatic analyses. For conducting this research, GEE (https://earthengine.google.com/) (accessed on 30 October 2025) cloud-based computational service was used as software for several reasons [19]. First of all, analysis of climate data implies processing of time series data. In this study, data from the period 1950 to 2100 was analyzed. GEE has the capacity to process vast amounts of data (e.g., parallel computation of large geospatial datasets, time series analysis), which is a significant advantage for research considering the vast territory of Kazakhstan, covering 2.7 million square kilometers. Secondly, GEE hosts curated climate datasets (e.g., ERA5-Land, CMIP6, CHIRPS), along with satellite data (e.g., Landsat, Sentinel, MODIS), and other environmental data such as soil moisture, vegetation indices, and land use/cover information. Apart from that, it is possible to upload your own data for analysis, which makes the platform exhaustive to new data. Lastly, GEE service allows free access to its full functionalities for academic researchers, thus giving an opportunity to conduct comprehensive research. Moreover, the scripting environment facilitates automation of data processing and improves reproducibility of the workflows.

Various data sources are available nowadays; however, the most common are climate reanalysis and model data. The selection of data was determined by specific criteria that they had to meet. Climate classification requires long-term data, with a minimum of two variables: temperature (temperature of the surface above 2 m) and precipitation (both rain and snow) values. Taking into account the tool interface, a selection was made from the dataset accessible in the GEE database. Thus, ERA5-Land reanalysis data [20] and CMIP6 (Coupled Model Intercomparison Project Phase 6) models [21] were utilized in this study, focusing specifically on temperature and precipitation parameters. Additionally, the CRU (Climatic Research Unit) dataset [22] was adjusted to ensure compatibility with the GEE interface and subsequently imported as an asset. While the CRU data were originally produced by the University of East Anglia [23], they were downloaded from the Climate Knowledge Portal of the World Bank. The dataset, initially in NetCDF format, was converted to GeoTIFF because GEE does not support the NetCDF format. The converted data was then exported as an asset to GEE for further analysis.

Table 1. Data and its characteristics used in the research.

Name	Variable	Spatial Resolution, Meters	Temporal Resolution	Unit	Time Period
ERA5-Land	Temperature	11,132	Monthly	Kelvin	1951–2023
	Precipitation			Meter
CRU	Temperature	55,659	Monthly	Celsius	1901–2020
	Precipitation			Millimeter
CMIP6	Temperature	27,830	Daily	Kelvin	1950–2100
	Precipitation			kg/m2/s

presents utilized data and its characteristics used in this research.

In this study, datasets with differing spatial and temporal resolutions were not combined. Consequently, a single dataset was used for generating climate maps at a time. When an inter-model dataset is employed, two model variables are integrated. However, since the integration in this case is performed within the CMIP6 dataset, where model parameters are consistent, no adjustments for resolution are necessary. The spatial resolution of the output climate maps matches the resolution of the input dataset used for generating them.

The Köppen–Geiger climate classification system is based on monthly precipitation and mean temperature, as well as their distribution throughout the year. It consists of five main climate groups, which are classified mostly by temperature: tropical-class A, arid-B, temperate-C, continental-D, polar-E. Then each main class is divided into subclasses that reflect specific precipitation patterns [6]. This system helps identify regional climate characteristics and their potential impacts on vegetation and ecosystems [7].

To analyze changes in Köppen–Geiger climate classifications, this study began with developing a tool that produces these maps. For that, the methodology lies in developing a code script in GEE that generates classified climate maps for any region and any time period using available data, including monthly mean temperature and precipitation amounts. The process flow is illustrated in Figure 1, and each step is explained in detail in the following subchapters.

To address code harmonization and optimization, the script is logically divided into four parts.

2.1. Data Preprocessing

This part of the code is adapted to harmonize input data. Since there is data of different types (different spatial and temporal resolutions, units), this part of the script transforms them into a single format. GEE uses the concept of ImageCollection [24] for time series data, where an Image has bands. In this study, an ImageCollection is utilized, containing yearly images, each with 12 bands for temperature and 12 bands for precipitation, corresponding to each month from January to December. The standard data form is set as monthly mean temperature in degrees Celsius and precipitation amounts in millimeters. This step is crucial to ensure that data is incorporated properly; all bands are standardized and ready for the next phase. The rest of the script is automated and does not require any change. Therefore, end-users only enter their input data and modify parameters such as time period, units and geographic location. Consequently, users can reuse the final script with minimal modifications to generate classified maps.

All datasets have different spatial resolutions. The datasets were not combined in this research, therefore there was no need to sample data to a single unified resolution. This is due to the approach of creating GEE applications based on each dataset individually, taking into account the computing capabilities of the Earth Engine platform.

2.2. Precalculation

This part of the code is designed to calculate aggregate metrics such as total precipitation, minimum and maximum temperatures, and other related climatic parameters. For classification, it is not enough to have only monthly temperature and precipitation data. There are climatic metrics to be calculated before the classification step; their defining criteria were described in the work by Peel et al., 2007 [7].

There are certain conditions that need to be defined in order to comply with specific zones. For instance, to define specific climate zones, such as desert and semi-arid regions, certain criteria must be satisfied. In particular, the calculation of the arid threshold is essential for the classification of these climates.

Table 2. Climatic metrics needed for classification.

Raw Data	Derived Data	Comments
Monthly mean Temperature	Average Annual Temperature
	Minimum of monthly mean Temperature
	Maximum of monthly mean Temperature
	Number of months with mean Temperature > 10 °C
Monthly Precipitation	Total Annual Precipitation
	Minimum monthly Precipitation
	Maximum monthly Precipitation
	Minimum monthly Precipitation for winter months	October to March in the Northern Hemisphere
	Maximum monthly Precipitation for winter months
	Minimum monthly Precipitation for summer months	April to September in the Northern Hemisphere
	Maximum monthly Precipitation for summer months
	Summer precipitation ratio	Represents ratio of summer precipitation to total precipitation. Summer ratio = precipitation during summer/total precip
	Arid threshold	Represents calculated aridity threshold based on temperature and summer precipitation ratio. The formula is as follows: Arid threshold = Average temperature × 20 + (140 if summer ratio > 0.3) + (140 if summer ratio > 0.7).
	Monsoon	Calculated monsoon minimum threshold based on precipitation. The formula is as follows: 100 − (total precip/25).

demonstrates all calculated climatic metrics which are important for further steps in the classification section.

2.3. Classification

In this section of the code, a function applies to a previously preprocessed data collection and subsequently classifies bands using the Köppen–Geiger classification scheme. Categorization of climatic zones follows certain rules of the Köppen–Geiger system: first defining the main classes, then classifying the subclasses. The classification process begins with Class B, as the criteria defining Class B also encompass the criteria for the other classes [7]. The classification algorithm is hybrid, applying a rule-based system followed by binary, and with implicit branches of a decision tree. Rule-based logic prevails in the code to classify different climate types. For instance, the code line below is a rule to identify specific conditions for class B.

var B = maxtemp.gt(10).and(aridthreshold.gt(precip));

Within the rule-based system, binary classification is used extensively. Each condition or rule leads to a true or false outcome, which is characteristic of binary classification. For example, the code line below is a binary condition that evaluates to true or false.

var W = precip.lt(aridthreshold.multiply(0.5));

While the code does not explicitly create a decision tree structure, the enclosed conditions can be seen as nodes and branches of a decision tree. The code below is an example where each node represents a condition and each branch represents the outcome of that condition.

var BW = B.and(W);

var BWh = BW.and(h);

2.4. Visualization Configuration

This section of the code is the final part of the entire script and it is tailored to visualize output results. The outcome of this work is a GEE application where users can generate classified climatic maps on the fly. The visualization configuration includes interface settings (Figure 2) such as overview (on the left), legend (on the right) and the map itself.

End-users can modify the time period of visualization and produce climatic maps online. Additionally, for better user experience, automated pixel area calculation is enabled, so that when maps are produced, it computes the area of each zone presented for the customized time and region in square kilometers.

2.5. Code Repository and Metadata

The entire code script is openly accessible at the GitHub (https://github.com/) repository link provided, https://github.com/yessimkhanova/koppen_maps (accessed on 1 January 2026). The repository contains script files containing codes annotated with comments detailing the objective of each code line and block. Furthermore, the repository features a README.md document with comprehensive metadata, elaborating the specific datasets utilized, as well as providing links to applications generated for each dataset. Additionally, a LICENCE file is provided, allowing users to properly manage copyright and usage permissions.

2.6. Evaluation of CMIP6 Climate Models Against Meteorological Stations

This work evaluated 32 (out of 34) climate models of the CMIP6 ensemble [21] using meteorological observations provided by the National Hydrometeorological Service of Kazakhstan “Kazhydromet”. Meteorological records of 34 stations (map of stations is shown in Figure 3) from 1961 to 2014 served as a reference for assessing each model’s reliability to determine the most accurate model for temperature and precipitation in the area of Kazakhstan.

Statistical error metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Bias Error (MBE) were calculated to evaluate model performance. Pixel values of model data were extracted from the CMIP6 dataset in GEE corresponding to the same locations and time as the meteorological stations, ensuring direct comparison with observed data.

For each station, the three models with the lowest RMSE values were identified, indicating the most accurate models for that location. Scores were assigned to these models based on their ranking: the model with the lowest RMSE value received 3 points, the second-lowest received 2 points, and the third-lowest received 1-point, and the remaining models that did not meet the criteria were not awarded any points. Given the total of 34 stations and 32 models, the highest possible score a model can achieve is 102. These scores were then summed up across all stations to provide an overall performance ranking.

Table 3. Best scored models based on the valuation of CMIP6 climate models against meteorological stations.

Best-Scoring Models Based on Various Error Metrics for Temperature
RMSE	MAE	MBE
GISS-E2-1-G: 83	GISS-E2-1-G: 81	GFDL-CM4: 60
MIROC-ES2L: 55	INM-CM5-0: 35	MPI-ESM1-2-LR: 42
EC-Earth3-Veg-LR: 19	MIROC-ES2L: 23	INM-CM4-8: 30
Best-scoring models based on various error metrics for precipitation
RMSE	MAE	MBE
MIROC-ES2L: 58	MIROC-ES2L: 48	KACE-1-0-G: 57
FGOALS-g3: 33	GISS-E2-1-G: 35	IPSL-CM6A-LR: 35
UKESM1-0-LL: 23	FGOALS-g3: 34	FGOALS-g3: 24

reveals the most scored models for both temperature and precipitation data, revealing the most consistently accurate models in representing the observed climate data. Additionally, models with the best scores for MAE and MBE are also provided to offer a comprehensive evaluation of model performance.

Metrics for all models and the code for calculation is available on the GitHub repository at https://github.com/yessimkhanova/cmip6_models_evaluation (accessed on 1 January 2026).

In this study, based on RMSE scores of evaluations of CMIP6 data against station records, temperature and precipitation variables were combined and the result was referred to as Inter-Model Dataset 1. Specifically, a map was generated (Inter-Model Dataset 1) by integrating temperature data from the GISS-E2-1-G model [25] with precipitation data from the MIROC-ES2L model [26].

2.7. Evaluation of CMIP6 Climate Models Against ERA5-Land Climate Reanalysis

The evaluation of CMIP6 models in comparison with the reanalysis of the ERA5-Land and climate was carried out in the GEE using a pixel approach. Monthly precipitation and mean monthly temperature data for the period from 1951 to 2014 were compared. The ERA5-Land data has been recalculated to match the coarser resolution of the CMIP6 models (27,830 m per pixel).

Model performance was first assessed at the pixel level using Mean Bias, Mean Absolute Bias, Root Mean Square Error and the Standard Deviation of Bias to quantify deviations from ERA5-Land reanalysis data. The spatial distribution of these statistical metrics is illustrated using maps in Figure 4.

The CESM2 [27] model generally underestimates climate variables compared to ERA5-Land across most of Kazakhstan, while it tends to overestimate them in the southeast.

Following this pixel-based evaluation, GEE’s reducers were applied to aggregate the biases to the country level. Specifically, a reduceRegion function [28] was used to compute average biases within Kazakhstan’s geographical boundaries. The use of GEE’s reducers ensured efficient and accurate country aggregation. This approach provided a clear country-level assessment while preserving spatial details from the pixel-based analysis.

The aggregated results revealed that CESM2 showed lower biases for temperature, while BCC-CSM2-MR [29] demonstrated lower biases for precipitation (

Table 4. Bias metrics for CMIP6 climate models based on the evaluation against ERA5-Land.

Metrics	CESM2 (Temperature)	BCC-CSM2-MR (Precipitation)
Mean Bias	−0.2485	−6.6857
Mean Absolute Bias	0.5433	7.8547
Standard Deviation of Bias	0.6782	9.5969
RMSE	0.5843	11.7222

). Metrics for all models and the code for calculating metrics are available on the GitHub repository at https://github.com/yessimkhanova/cmip6_models_evaluation (accessed on 1 January 2026).

In this study, based on bias metrics of evaluation of CMIP6 data against ERA5-Land dataset, temperature and precipitation variables were combined and the result was referred to as Inter-Model Dataset 2. Specifically, a map was generated (Inter-Model Dataset 2) by integrating temperature data from the CESM2 model with precipitation data from the BCC-CSM2-MR model.

3. Results

3.1. Climate

Figure 5 shows the annual temperature trend for the territory of Kazakhstan based on observed data for 34 meteorological stations. The graph shows an increasing trend in temperature across all stations. To see annual statistics for each observation station, see the GitHub repository at https://github.com/yessimkhanova/annual_meteo_statistics_KZ (accessed on 1 January 2026).

Table 5. Temperature trend analysis for meteorological stations (1961–2014).

Station	Mann–Kendall Trend	Sen’s Slope Value Per Decade
Aksai	increasing	0.320513
AksuAyuly	increasing	0.246528
Aktobe	increasing	0.339286
Astana	increasing	0.355856
Atbasar	increasing	0.20679
Atyrau	increasing	0.328704
AuylTuraraRyskulova	no trend	0.075521
Besoba	increasing	0.177083
Blagoveshenka	no trend	0.154762
FortShevchenko	increasing	0.278571
Ganyushkino	increasing	0.236111
Karaganda	no trend	0.166667
Karasu	increasing	0.224806
Karaulkeldi	increasing	0.30303
Kordai	increasing	0.278571
Kurshim	no trend	0.15942
Kyzylorda	increasing	0.46875
Leninogorsk	no trend	0.120098
Moiynkum	increasing	0.278646
Pavlodar	increasing	0.206439
Petropavlovsk	increasing	0.26087
Ruzayevka	increasing	0.166667
Shalkar	increasing	0.172222
Shemonaikha	increasing	0.345833
Shymkent	increasing	0.244048
Taipak	increasing	0.373932
Taiynsha	increasing	0.240741
Taraz	increasing	0.305556
Uralsk	increasing	0.346591
Urzhar	no trend	0.077586
UstKamenogorsk	no trend	0.19697
Vozvyshenka	increasing	0.205357
Zharyk	no trend	0.16358
Zhezkazgan	increasing	0.270202

presents the temperature trend analysis for meteorological stations from 1961 to 2014, showing the Sen’s slope for each station per decade along with the Mann–Kendall test results for trend significance. The Mann–Kendall test was applied to assess the statistical significance of the temperature trends, revealing significant trends at some stations, while showing statistically insignificant trends at others. These trends are further illustrated by the Sen’s slope estimator, which indicates the rate of temperature change per decade for each station. On average, temperatures increased by 0.47 °C per decade, with the sharpest increases observed in Kyzylorda, Taipak, and Astana, while stations in East Kazakhstan (Ust Kamenogorsk, Kurshim, Leninogorsk, and Urzhar) exhibited almost no significant temperature trends.

The map of Sen’s slope values, visualized per decade in Kazakhstan, is shown in Figure 6. The map illustrates the spatial distribution of temperature trends across Kazakhstan from 1961 to 2014. The figure reveals that the highest temperature increases per decade are observed in the southern region, while the western part of the country also exhibits significant warming trends. The most notable increase in temperature in the northern area is observed in Astana. In contrast, the eastern part of Kazakhstan, including cities such as Ust Kamenogorsk, Kurshim, Leninogorsk, and Urzhar, shows no significant temperature increase during the observed period.

3.2. Köppen–Geiger Climate Maps

The following “Köppen–Geiger maps generator” applications were deployed using the App feature in GEE based on the available data (

Table 6. Google Earth Engine applications.

	Köppen–Geiger Maps Based on
1	CRU data.
2	ERA5-Land data.
3	CMIP6 ensemble, mean of 32 models, historical and scenario data.
4	Inter-Model Dataset 1, GISS-E2-1-G (temperature data) and MIROC-ESL2L (precipitation data) models, historical and scenario data.
5	Inter-Model Dataset 2, CESM2 (temperature data) and BCC-CSM2-MR (precipitation data) models, historical and scenario data.

).

All the links to the applications are provided in the aforementioned GitHub repository or see the applications gallery of GEE users, in this case, at https://kalamkas.users.earthengine.app/ (accessed on 1 January 2026).

As indicated in Table 1, three datasets are used to produce climatic maps. These maps are generated from 1901 to 2020 using CRU data, and from 1951 to 2023 using ERA5-Land data. The CMIP6 dataset differs from other climate data in that it includes both historical and projected data from an ensemble of climate models. For the projected data, the CMIP6 ensemble applies SSPs specifically medium pathway SSP2-4.5 and fossil-fueled development SSP5-8.5 scenarios. Therefore, climate maps were generated based on the mean value of the monthly temperature and precipitation amount of 32 models for historical data and SSP2-4.5 and SSP5-8.5 scenarios.

Generated Köppen–Geiger climate maps of Kazakhstan (presented in Figure 7) show that the overall shift is likely to happen from the southwest to the northeast by expanding mainly the following zones: BSk (Cold semi-arid), BWk (Cold desert), and Dfa (Humid continental with hot summer). This implies that climate change impacts on Kazakhstan’s zones may lead to the climate becoming warmer.

The tool can generate climate maps for both projected scenarios. In this study, however, the focus is on the maps projected for the worst-case SSP5-8.5 scenario, while maps for the SSP2-4.5 scenario are also possible. Future maps under the SSP5-8.5 scenario, generated using two approaches: one based on the CMIP6 ensemble mean and the other combining temperature and precipitation data from different models (referred to as Inter-Model Dataset 1 and 2), reveal significant shifts in Kazakhstan’s climate zones by the end of the century, with new zones emerging (see Figure 8, Figure 9 and Figure 10). These include BSh (Hot Semi-Arid) and Csa (Mediterranean) in the south, and BWh (Hot Desert) in the southwest.

Table 7. Area of climate zones in km².

Zone	CMIP6 Ensemble		Inter-Model Dataset 1		Inter-Model Dataset 2
	Present 1981–2010	Future, SSP5-8.5 2061–2100	Present 1981–2010	Future, SSP5-8.5 2061–2100	Present 1981–2010	Future, SSP5-8.5 2061–2100
BSk: Cold Semi-Arid	1,077,268.17	1,318,337.59	986,789.48	1,159,601.18	1,090,490.81	1,440,459.87
BSh: Hot Semi-Arid	-	58,728.19	-	18,158.21	-	88,002.89
BWk: Cold Desert	534,297.94	771,158.38	570,050.93	717,690.98	593,667.47	774,695.25
BWh: Hot Desert	-	52,609.98	-	12,952.57	-	144,194.31
Dfa: Humid Continental Hot Summer	322,448.31	502,516.34	24,471.64	425,400.21	-	38,034.49
Dfb: Humid Continental Mild Summer	735,069.28	31,041.7	493,760.34	46,851.59	20,507.46	-
Dfc: Subarctic	51,704.70	19,942.85	24,981.27	11,457.42	5331.50	-
Dsa: Dry Hot Summer	36,268.09	19,870.81	315,091.27	293,666.97	425,907.11	271,798.46
Dsb: Dry Mild Summer	6640.94	-	223,774.16	4436.63	234,627.82	3036.07
Dsc: Dry Cold Summer	3524.56	3524.56	7789.96	5686.65	11,420.05	8312.50
Dwa: Humid Continental Hot Summer	-	-	-	64,455.01	85,811.16	15,193.62
Dwb: Humid Continental Mild Summer	-	-	91,065.33	9306.95	299,408.60	8574.03
Dwc: Subarctic Hot Summer	-	-	34,066.55	8377.65	24,756.80	2918.81
Csa: Mediterranean	32,577.82	33,931.96	6343.16	24,714.34	-	10,615.58
Cwa: Humid Subtropical	4466.81	2255.52	26,082.51	11,161.53	19,649.90	11,160.48
ET: Tundra	9651.26	503.27	9651.26	503.27	5417.69	-
EF: Ice Cap	503.27	-	503.27	-	-	-

shows the areas of climate zones shown on the maps in Figure 8, Figure 9 and Figure 10, providing a quantitative overview.

4. Discussion

4.1. Challenges and Limitations

In the context of this work, the goal is to share the code and all related steps for its implementation, ensuring that the tool is openly accessible to the community. GitHub serves as an ideal platform to facilitate this openness and collaboration. Utilizing GitHub is a crucial step in preserving the work and addressing the challenges associated with managing all GEE applications. Although all developed applications perform similar tasks, the limitations of GEE computational power and memory usage limit [30] do not allow the building of one single tool. This is related to the complexity of processing different data sources, long time period, and computation functions. Therefore, having a GitHub repository that provides links to all the applications is essential. Moreover, GEE provides a “Gallery” feature where you can access all your applications.

Kazakhstan lacks proprietary climate models exclusively designed for the area. Therefore, it is vital to have rich ground measurements for a long time period in order to evaluate available models for identifying the best fit model for the area. There are a total of 347 meteorological stations across Kazakhstan [31]. However, for this research, data from 34 stations were used and provided consistent and complete monthly records. Some stations had data gaps, with missing values for certain months or years, which made them unsuitable for inclusion in the analysis. Therefore, only the 34 stations with complete and uninterrupted monthly data were selected for this study. Station proximity map (Figure 11) illustrates the coverage of the stations and distances from nearest station, showing that in some areas the stations are dense, while other parts of the region are not covered. Considering the vast size of Kazakhstan (2.7 million square kilometers) the number of available stations is very limited. Furthermore, the stations are not evenly distributed across the country and fail to adequately represent regions with diverse topography, such as mountainous areas. This uneven station distribution significantly impacts the assessment of climate models, as it makes it difficult to accurately compare model outputs against station data. Consequently, the accuracy of the climate model evaluation is affected, as some regions may be underrepresented, leading to potential biases in the results.

There is a temporal mismatch in the evaluation periods of the data: the meteorological station data covers the period from 1961 to 2014, while the ERA5-Land data spans from 1951 to 2014. This discrepancy is due to the availability of the respective datasets. ERA5-Land extends further back in time, but the meteorological station records for Kazakhstan are not available for the same time period. Even when station data is available, it is inconsistent, with gaps in certain months and years. This temporal mismatch applies generally to all datasets used in this work, including those for producing Köppen–Geiger climate maps. To maximize the use of the available data, it was decided to use the datasets within their respective time spans. For example, the CRU data is available from 1901 to 2020, and it was utilized for the full period available to generate climate zone maps.

The performance of CMIP6 models against ERA5-Land reanalysis data and meteorological station observations was evaluated to identify the best-performing models. However, no bias adjustment was applied to the model outputs in this study. The sparse distribution of available meteorological station data used in this study makes it unfeasible to use these stations for bias correction purposes. As an alternative, the ERA5-Land dataset was considered for evaluation. However, the evaluation results showed that ERA5-Land does not provide a reliable reference for improving model performance in the region of Kazakhstan. When examining the difference in the monthly mean temperature of model GISS-E2-1-G relative to ERA5-Land over a 64-year period (1951–2014) in Kazakhstan showed a mean difference of 0.94 °C, with the minimum standard deviation of these monthly differences at 1.95 °C and the maximum at 4.24 °C. These values suggest that applying a simple bias correction does not enhance the model’s accuracy when ERA5-Land is used as the reference dataset. Therefore, bias correction was not applied, as it would not lead to a meaningful improvement in the model’s performance. Consequently, the output Köppen–Geiger climate maps are based on the raw model data, which may result in the misclassification of climate zones. This represents a limitation of the study, as the bias correction was not performed. Although a bias-corrected CMIP6 dataset exists by Zhongfeng et al. (2024), its resolution of 1.25° × 1.25° is only suitable for global-scale examination, not regional ones [32]. Kazakhstan spans latitudes between 40° and 55°. At this resolution, the dataset would represent the entire country with only 15 grid cells (or “pixels”), which is far too few to capture the regional climate variations accurately. The main focus of this research is to provide visualization tools for climate change, rather than assessing the accuracy of the data used. However, as these tools can be easily customized to any data source, and if anyone is interested in using the aforementioned dataset, it is possible to do so, but this is only advisable at a larger (at least continental) scale due to its reduced resolution.

This study applied two approaches to generate projected Köppen–Geiger climate maps. The first method utilized the CMIP6 ensemble mean, where temperature and precipitation from 32 models were averaged to provide a weighted climate projection. The second approach combined temperature from one model and precipitation from another, and was represented by two datasets in this study: inter-model dataset 1 and inter-model dataset 2. Inter-model dataset 1 combines temperature from model GISS-E2-1-G and precipitation from model MIROC-ES2L, while inter-model dataset 2 combines temperature from model CESM2 and precipitation from model BCC-CSM2-MR. Although this alternative approach offers a new perspective, it introduces significant challenges. Climate models simulate interdependent physical processes, where temperature and precipitation are linked within the same model. Taking temperature from one model and precipitation from another can break this connection, leading to unrealistic combinations [33]. This approach could distort projected Köppen–Geiger classification. In contrast, the first approach of using the CMIP6 ensemble mean ensures more reliable results by averaging outputs from multiple models. This method minimizes biases and maintains the consistent relationship between temperature and precipitation. While the multi-model ensemble approach remains widely used, exploring the combination of variables from different models offers an interesting direction for further research, though it may come with certain limitations. Another method developed by the authors for generating future maps of climate zones in Kazakhstan is outlined in a separate study [34], where a modal map of the Köppen–Geiger climate was created. This approach provides an alternative perspective to the methods used in the present study.

Inter-Model Dataset 2 predicts faster aridification compared to Inter-Model Dataset 1 and the CMIP6 ensemble mean. While Inter-Model Dataset 1 and the CMIP6 ensemble mean suggest that the Dfa (Humid Continental Hot Summer) zone is likely to cover a significant area in the country, Inter-Model Dataset 2 projects it to be limited to a smaller area in northern Kazakhstan. The BWh (Hot Desert) and BSh (Hot Semi-Arid) zones are predicted in all three models; however, the Inter-Model Dataset 2 predicts these zones may occupy a larger area than the other models.

The area of Kazakhstan is experiencing significant temperature increases with a stable precipitation pattern [4]. Accordingly, the projected Köppen–Geiger maps show transition from cold to warmer climate types, specifically shifts from BSk (cold semi-arid) to BSh (hot semi-arid), BWk (cold desert) to BWh (hot desert) and Dsa (continental hot) to Csa (Mediterranean hot). These transitions represent the temperature sensitivity of the Köppen–Geiger classification to a temperature fluctuation. Temperature and precipitation factors in the classification system are equally important, but statistically insignificant changes in precipitation support the findings of the study that temperature is a key factor in the projected climate zone shifts in Kazakhstan. Temperature-driven projected climate shifts are aligned well with the observed warming trends in Kazakhstan. However, high reliability of the classification system on precipitation amount should be taken into account to avoid misclassification of climate type. This might happen in regions where temperature changes are more distinct than precipitation changes. Rapid temperature increase in Kazakhstan presents a challenge of adequate classification of climate by the Köppen–Geiger system as it may not fully capture the climate dynamics. In addition, the system simplifies classification by integrating only temperature and precipitation; therefore, it may not properly capture local variations in other variables such as moisture and altitude.

Results of the research align closely with previous studies showing that temperature is the primary driver of climate zone shifts in Kazakhstan. Future projections by Beck et al., (2023) indicate increasing aridity and transitions from semi-arid to arid climates under warming scenarios [9]. Analysis by Hu & Han (2022) shows a northward expansion of desert and arid zones across Central Asia [35] and the same trend is happening at the national scale within Kazakhstan itself which is described in the work by Bissenbayeva et al. (2025) [36]. Regional study by Fallah et al. (2024) using CMIP5 and CMIP6 ensembles demonstrates that warming dominates Köppen–Geiger climate transitions in Central Asia, with precipitation playing a secondary role [37]. Song et al. (2025) shows that temperature increases alone can trigger Köppen–Geiger class changes in mid-latitude continental regions [38]. These findings are in agreement with prior work in this area and provide national-scale confirmation that projected new Köppen–Geiger zone changes in Kazakhstan are largely temperature-driven.

Under SSP5-8.5, southern Kazakhstan is likely to face near-total desertification (BWh/Csa zones) by 2100, while it is expected that at least 60% of the country’s land area is projected to experience a change in climate zone. These changes highlight the urgent need for climate action to mitigate their potential impacts. If no action is taken, the southern region of Kazakhstan may face increasing desertification, making immediate adaptation measures essential. The projected expansion of the BSk (Cold Semi-Arid) and BWk (Cold Desert) zones indicates ongoing aridification, which poses a serious threat to rain-fed agriculture and regional livelihoods.

This study did not include the generation of maps based on models that demonstrate stronger performance under climatic conditions similar to those observed in Kazakhstan. Future research could extend this work by conducting comprehensive analysis to identify and evaluate models that perform well in regions with analogous climate types.

4.2. Implications

Understanding regional climatic patterns is essential for sustainable development, resource management, and climate change adaptation. Climate maps are more insightful for showing how the climate zones change with time and location. Therefore, research results and further developed tools have an importance in realization of both scientific and practical ideas. The tool developed in this study has the potential to facilitate the monitoring of vegetation changes and contribute to biodiversity conservation efforts. Given that the tool generates future Köppen–Geiger climate maps for any region, it can assist in identifying areas at risk of species loss and land degradation. However, a thorough assessment of data accuracy and reliability is necessary prior to application. Additionally, the developed tool might assist lecturers for educational purposes in schools or universities to demonstrate how climate maps visualize and show the changes between certain time periods. Moreover, opening up the code script is compliant to the Findable, Accessible, Interoperable and Reusable (FAIR) concept [39]. Accessibility of the scripts and its documentation encourage researchers in replicating or extending this work. Reproducibility of the computation promotes validation of the findings and additionally, collaborative research initiatives.

The future Köppen–Geiger climate projections for Kazakhstan under the SSP5-8.5 scenario indicate that the southern and southwestern regions are particularly vulnerable to climate change. Therefore, proactive measures, including climate adaptation strategies and policy development, are essential to mitigate the risk of desertification in these areas.

5. Conclusions

The primary objective of this study was to analyze historical changes in climate zones across Kazakhstan and their potential future shifts using projected Köppen–Geiger climate maps under the SSP5-8.5 high-emission scenario for the 2061–2100 period. The results provide an overview of potential climate shifts, illustrated visually and based on different datasets (ERA5-Land, CRU, CMIP6) for historical and future periods. Beyond the climate analysis, the main contribution of this study lies in the development of a cartography-focused tool and workflow for processing and visualizing complex climate datasets. The objective was addressed through the development of a GEE-based tool that enables the automated generation of Köppen–Geiger climate maps. Overall, this work provides a cartographic framework supporting efficient climate zone geovisualization and can be improved for climate mapping and advanced spatial analysis. Future climate projections indicate that immediate actions are necessary to mitigate severe changes in climate zones across Kazakhstan. Under a high-emissions scenario, by the end of the century, the southern regions of the country are projected to experience significant desertification, while the majority of the land area may transform to arid and semi-arid conditions. Specifically, new climate zones such as BWh (Hot desert), BSh (Semi-arid), and Csa (Hot-summer Mediterranean) are projected to emerge in the west and south of the country. Additionally, the expansion of BSk (Cold Semi-Arid) and BWk (Cold Desert) zones is expected, indicating a trend toward aridification. Based on the results, the projected climate zones are primarily driven by temperature variations. While this study offers valuable insights, it is limited by the small number of meteorological stations used for model evaluation. Expanding the number of stations in future analyses could strengthen the reliability of the findings. It is crucial to highlight that the study did not perform bias correction to the climate models. This affects the accuracy of the output classified maps and misclassifies the future Köppen–Geiger climate zone as it is based on a raw model output. Furthermore, future research could focus on a more comprehensive analysis of the climate maps generated from individual models within the CMIP6 ensemble. Therefore, no decision should be based on only one model. Using as many models and stations as possible for evaluation is an important requirement for appropriate model selection to ensure accurate climate representation and correct classification. The analysis using inter-model datasets for climate maps demonstrates their benefit for understanding localized impacts; however, these maps are specifically tailored to individual regions, underscoring the need for regional assessments. In parallel, the development of scalable tools that can generate localized maps efficiently is essential. The GEE tool mitigates resource constraints by enabling on-demand climate zone mapping, empowering policymakers to visualize risks without specialized computational expertise. These findings highlight the critical role of geospatial technologies in advancing climate studies and supporting mitigation strategies.