Research on a Random Forest Regression Model for Climate

Prediction in the Context of Wildfires

Junru Lou

Reading Academy, Nanjing University of Information Science and Technology, Ningliu street, Nanjing, China

Keywords: Wildfire, Random Forest Regression Algorithm, Climate Change, Global Warming.

Abstract: Until now, research has sometimes used the survivorship curves which is generated by statistics on tree age to

estimate the fire frequency. However, due to the infrequency of fires, it is hard to infer the existing woodland

studies about he relationship between fire occurrence and extent and short-term climate change. This paper

has an in-depth analysis of the quantitative relationship between fire size and greenhouse gas emissions with

the integrating global fire scale data, greenhouse gas emissions data and global temperature change data.

Random forest regression algorithm is used on this research and supports the analysis of the quantitative

relationship between fire size and greenhouse gas emissions. At last, the emission levels of greenhouse gases

will be analyzed, and the impact of greenhouse gas emissions on global warming will be discussed. This paper

has a goal of building a prediction model based on the wildfire burning scale. It will be used to predict the

level of impact on global warming. It is expected here to provide new insights into the mechanisms of global

climate change and provide a scientific basis for formulating effective environmental protection and fire

management policies.

1 INTRODUCTION

Recently, the wildfires raging through Los Angeles in

the United States have caught great attention of the

entire world. There are many extremely serious

consequences for local society, economic property

and many other areas with the widespread and

uncontrollable of wildfire. With those great threats,

the personal and property safety of local residents are

badly damaged. Besides that, Wildfire gives unique

challenges to conservation because it various greatly

in time and space indicating the randomness in nature

(McKenzzie et al.,2004). These fires cause massive

loss of vegetation and also destruction of wildlife

habitat showing the Fires destroy the stability of

ecosystem. Global environment also shows great

threat because of fires. What else, the spread of smoke

and hazardous substances from burning areas causes

much side effects on air quality. Human health and

life have been affected because of all those

shortcomings.

Establishing a national wildfire database has

become an important goal for research and disaster

https://orcid.org/0009-0005-7370-6551

prevention in various countries. America is a typical

example which faces many needs and challenges. The

core data elements of these databases include

locations, serial numbers, final locations and recorded

control date of the fire. This information is widely

used for geospatial fire analysis and risk assessment

(Short et al., 2014). To support this goal, many

models has been developed, which meant to capture

the multiple factors that influence the behavior as

completely as possible. The behaviors of individual

fires that can be described by existing data and

standard models. This may better weighting the

relationship between wildfire spread and ecosystem

response(Moritz et al., 2004). Besides of that, one

research shows that ability to accurately estimate the

occurrence of catastrophic events is particularly

critical when the distribution of extreme events is

clearly correlated and the frequency of abnormal

events exceeds expectations (Holmes et al., 2008).

One organization based on Google Earth Engine

(GEE) frame builds a large publicly available datasets

which is covering a wide range of observational

variables. These datasets not only contain many fire

events and their related variables, but also get the

392

Lou, J.

Research on a Random Forest Regression Model for Climate Prediction in the Context of Wildﬁres.

DOI: 10.5220/0013698100004670

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Data Science and Engineering (ICDSE 2025), pages 392-399

ISBN: 978-989-758-765-8

combination of the impact of human activities. It

provides a foundation for wildfire spread prediction,

which is widely used in machine learning and

computer vision analysis(Huot et al.,2022).

While wildfire raging in forests and other

vegetated areas, large range of greenhouse gases will

be release into the atmosphere, such as CO2.

Carbon

dioxide is one of greenhouse gases that causes

global warming. The rising concentrations of

carbon dioxide in the atmosphere pose enormous

environmental challenges to the earth(

Nunes et al.,

2023). Expect those gases, burning process will also

generate a lot of dust pollution, such as PM2.5.

Previous studies have shown that worsening air

quality due to climate warming will have a huge

impact on humans in terms of PM2.5

concentrations (

Liu et al.,2021). These two aspects

making an impact together to exacerbate global

warming. This phenomenon has long term influence

on earth's climate system. However, although great

effort are used to prevent the occurrence of wildfires.

Some wildfires may still occur because of some

natural or man-made causes. While these Inevitable

wildfires spreading, It is crucial to evaluate the

specific impact on global warming. These actions is

meant to release the side impact on environment.

What to do next is to minimize its impact on climate

change. With these steps， the adverse impact of

wildfires on the global climate can be reduced as

much as possible. Some targeted prevention efforts

could also be developed based on the burning level.

In this way, the balance of global ecosystem could be

maintained better.

Despite the progress made in fire management,

the key to addressing the challenge is to use mature

data processing methods such as machine learning to

understand complex wildfire phenomena and

mitigate their impacts (Bot et al.,2021). This paper

choose the random forest regression (RF) algorithm

to analysis the updated data. This model is meant to

fit existing training set data and evaluated the model's

fitting effect through the test set data. Through

building multiple decision trees and combining their

prediction results, this model can fit features and

trends to training set data. The model will evaluate on

the test set data in order to measure its prediction

accuracy and fitting effect. It shows strong anti-noise

and nonlinear fitting capabilities that is really suitable

to analysis the multidimensional features in complex

datasets. Through analyzing the model performance,

some references will be provided for the subsequent

optimization of pollution classification rules. The

burning scale is mainly represented by the global

burning area. At the same time, the main factors

affecting global warming include carbon emissions

and air pollution, which is represented by CO2 and

PM2.5 respectively. By fitting global land burnt area

and material emissions into a neural network model,

the corresponding models and fitting results are

supplied. The next step is to do classification on the

impact of wildfires on global warming. After the

above analysis, professionals could take appropriate

action according to the research impacts.

At the first part of this paper, the data sources, pre-

processing steps, and some detail descriptions of data

are elaborated carefully. Because of these data pre-

processing analysis, the following predictive models

and classification models could be selected targeting

fire emissions and pollution scale. The following part

shows the pollutants emissions prediction results

based on fire size. At the same time the classification

of the wildfire pollution scale is introduced. After all

the basic data and model descriptions, the research

results were analyzed in depth. There are several

contents that get detailed explanation, which are main

findings, study limitations, and directions for future

research. Through this structured research framework,

this paper gets a comprehensive discussion of fire

emissions and their classification.

2 METHOD

2.1 Data Source and Preprocessing

This research uses data resource from the Our World

in Data (OWID) platform. This platform is

established Oxford University research team and

provided with continuously updated data resources.

This team is an authority data providing organization

which is meant to do research on the global

development problems. As a world-known open data

resource library, OWID uses multidimensional data

visualization technology. It provides standardized

datasets which is covering a wide range of fields for

researchers. These data has a relatively high academic

value after strict quality control and verification.

Spatial autocovariates, derived from neighboring

estimates of the response variable, have improved the

accuracy of burned land maps using satellite data and

have improved the classification accuracy of large-

scale land cover maps (Koutsias et al., 2010).

In the research process, four key data-set are

chosen: annual-area-burnt-by-wildfires, annual-area-

burnt-by-wildfires-gwis, annual-carbon-dioxide-

emissions and annual-pm25-emissions-from-

wildfires. Through combining the first two databases,

Research on a Random Forest Regression Model for Climate Prediction in the Context of Wildﬁres

393

a complete interannual wildfire combustion area time

series was constructed from 2003 to 2024. At the

same time, the next two databases are used as the

independent observational indicators of the emission

of CO₂ and PM₂.₅.

In the data preprocessing link, the multi-source

datasets are performed with system integration. It is

used to ensure the unity of the temporal dimension

and integrity of the data structure. What's more, we

do restart sorting to three key variables, which are

burning area, CO₂ emission and PM₂.₅ emission, in

the time order. It lays foundation for the subsequent

timing analysis. Thanks to the strict data quality

control system of OWID platform, the original data

used in this research has the high quality in the

integrity and reliability. Therefore, there is no need to

conduct a routine data cleaning or do missing values

filling. These profits do strong basic guarantee to the

accuracy and credibility of this research.

2.2 Study Area and Timescale

In the study, six continents were selected as the

research objects: Asia, Africa, Europe, North

America, South America, Oceania and the global

level was included in the spatial coverage. There are

three criteria to select the area used in the study: the

diversity of geographical distribution, the differences

in climatic conditions, and the heterogeneity of

wildfire characteristics. With these three criteria, the

result will be possessed with full representativeness

and universality.

In terms of the temporal dimension, this research

adopts an annual time scale. Data from 2003 to 2024

are selected to analyze. There are some

considerations to decide the timeframe. First of all,the

year 2003, marking the widespread global attention

of wildfire, is a Key milestone of the significantly

rocketing climate change problem. Secondly, the data

of 2024 is the newest data that is available. This can

reflect the updated trends of the worldwide wildfire.

Through the Systematic analysis of this 22-year time

series of data. It can not only effectively capture the

spatiotemporal evolution of global wildfires, but also

improve the fitting accuracy and prediction ability of

the model. Besides of that, the annual timescale can

reflect both the Long-term trends in wildfire activity

and the periodicity characteristics. It can make

significant reduce the short-term fluctuations

interfere with research results. Because of these steps,

the scientific and reliability of data analysis can be

ensured successfully.

2.3 Model Building

2.3.1 A Prediction Model of Wildfire Size on

Emissions

In this study, the random forest regression algorithm

was used to construct the prediction model. The

random forest (RF) regression method is particularly

popular because of its broad applicability, tolerance

for nonlinearities in the data, and adaptability to high-

dimensional feature spaces (many predictors). It

bootstraps parts of the data, grows a decision tree on

each part, and then aggregates the predictions (Borup

et al., 2023). It is meant to analysis the relations

hip of

wildfire scale and pollutant emissions

quantitatively.

The model predicts the emissions of two major

pollutants, CO2 and PM2.5, respectively. Random

forest regression is an

ensemble learning algorithms,

which is chosen because of its advantage on

dealing with nonlinear relationships and high-

dimensional data aspects. This algorithm could not

only have the high forecast accuracy, but can also

exhibits good model stability.

In order to validate model performance, this

research takes the evaluate the predictions using a

visual approach. These Scatter distribution plot are

made by plotting the scatter distribution of the

actual observed values of pollutant emissions with

the predicted values of the model. Among which

the blue spots represent the actual observations. At

the same time the red spots represent the predicted

observations. Through comparing the analysis of

Spatial distribution characteristics of these two sets

of data points, it can be intuitively estimated the

accuracy of the model's predictions and how well

they agree with actual observations.

2.3.2 Wildfire Pollution Scale Classification

Model

In order to get further exploration of the

relationship between wildfire scale and pollutant

emissions, this research established a wildfire scale

classification mode based on the distribution of

CO2 and PM2.5 emissions. The model classifies

wildfire events into three pollution levels: Light,

Moderate, and Heavy, in order to quantitatively

assess the difference in environmental pollution

caused by wildfires of different scales.

While doing the design of the classification

criteria, the res

earch takes the quantile method to

ICDSE 2025 - The International Conference on Data Science and Engineering

394

dynamically determine the partition threshold.

Specifically speaking, the categorical boundaries

for CO2 and PM2.5 emissions were determined

based on the first quartile (33%) and second

quantile (66%) of their distribution.

Thereinto, the classification threshold for CO2

emissions is:

𝐶𝑂



𝑏𝑜𝑢𝑛𝑑𝑎𝑟𝑦1 = 2 × 10



(1)

𝐶𝑂



𝑏𝑜𝑢𝑛𝑑𝑎𝑟𝑦2 = 4 × 10



(2)

The classification thresholds for PM2.5

emissions are:

𝑃𝑀2.5𝑏𝑜𝑢𝑛𝑑𝑎𝑟𝑦1 = 7.8 × 10



(3)

𝑃𝑀2.5𝑏𝑜𝑢𝑛𝑑𝑎𝑟𝑦2 = 1.5 × 10



(4)

1. Light pollution: CO2 and PM2.5 emissions are

lower than the first quantile;

2. Moderate pollution: CO2 and PM2.5 emissions

are both below the second quartile;

3. Heavy pollution: CO2 or PM2.5 emissions are

higher than the second quartile.

By implementing a classification function,

classify_pollution_level, this research

labelled each

record in the data set with a pollution level. In order

to Presents the classification results directly, the

research uses scatter plot of CO2 and PM2.5

emissions based on pollution level colouring.

Through the visualized result, the classification

model can effectively distinguish the

environmental pollution degree of wildfires of

different scales and provide scientific support for

relevant decision-making in order to make

scientific support for the decisions. For example,

for severe pollution incidents, prevention and

control measures can be prioritised to reduce their

potential harm to ecosystems and public health.

2.4 Model evaluation and optimisation

In this study, the dataset was divided by stratified

random sampling. This can ensure the

representativeness of the sample and the reliability of

the experiment. Here, it is divided the original

datasets into a training set and a test set at a 7:3 ratio.

70 percent of data are used as the training model and

30 percent of data are used as testing model. To

prevent model bias that may result from uneven data

distribution or sequential effects, the random shuffle

function (shuffle = True) is enabled in the data

partitioning process. Random forest regression

prediction models for CO₂ and PM₂.₅ emissions were

constructed, respectively in this study. In terms of

model parameter settings, the main configurations are

as follows:

100=− estimatorsn

(the number of the tree

in the forest is 100)

42=− staterandom

(Random seeds are

fixed to ensure reproducibility of results.)

2.4.1 Model parameter settings

The parameters of the random forest regression

model are shown in table 1.

Table 1: Model Parameters of CO2 and PM2.5 Emissions

Paramete

Value

ootstrap True

1 ccp_alpha 0.0

2 criterion squa

erro

3 Max_depth

one

4 max_features 1.0

5max

leaf

nodes

one

6max

samples

one

7min

impurit

decrease 0.0

8min

samples

leaf 1

9 min_samples_spli

10 min_weight_fraction_le

0.0

11 monotonic_cs

one

12 n_estimators 100

13 n

obs

one

14 oob

score False

15 random

state 42

16 verbose 0

17 warm star

False

2.4.2 Model evaluation metrics

In order to comprehensively evaluate the prediction

performance of the model, a multi-dimensional

evaluation index system was constructed in this

study. Five representative evaluation indicators

were selected:

Mean Square Error (MSE), Root

Mean Square Error (RMSE), Mean Absolute Error

(MAE), Mean Absolute Percentage Error (MAPE),

Coefficient of Determination (R²).

These indicators

quantitatively evaluate the model performance

from different dimensions such as error level,

relative error, and goodness-of-fit. Table 2 shows

the specific application results of the above

evaluation indicators in CO₂ and PM₂.₅ emission

prediction models:

Research on a Random Forest Regression Model for Climate Prediction in the Context of Wildﬁres

395

Table 2: Evaluation results of the emission model

Metric CO2 Emissions PM2.5 Emissions

0 MSE 80418906238585136.00 5111890742130.03

1 RMSE 283582274.20 2260949.08

2 MAE 205941023.96 1467302.85

3 MAPE 19.82% 25.40%

4 R² 0.9841 0.9433

3 RESULTS

3.1 Projections of Wildfire Size on

Emissions

In this research, the random forest algorithm is used

to predict the relationship between the wildfire scale,

which is represented by the burning area, and CO2

and PM2.5 emission. Through the analys

is of

comparison of actual observations with model

predictions, we evaluate the performance of the

model.

3.1.1 Prediction Results of CO₂ Emissions

Figure 1 represents the comparison of the actual and

predicted emission of CO2. As this figure shows, the

random forest algorithm can better capture the

relationship between CO2 emissions and combustion

area. Although the predictions of the model are

slightly off in some high burning areas, the forecast

result is overall close to the actual value. All these

things above shows this model excels when dealing

with nonlinear relationships that can be effectively

used to predict

Figure 1: The comparison of the actual and predicted

emission of CO2 (Picture credit: Original)

3.1.2 Forecast Results for PM2.5

Emissions

Figure 2 shows the comparison of the actual and

predicted emission of PM2.5. Similar to the predicted

result of the CO2 emission, the random forest festival

does well in the forecast of PM2.5. The model can

accurately reflect the trend of PM2.5 emission with

the burning area, especially in the middle level

burning areas. However, in the extremely high

burning areas, the prediction accuracy of the model

will decrease slightly, which may be related to the

sparsity of the data distribution.

Figure 2: The comparison of the actual and predicted

emission of PM2.5 (Picture credit: Original)

3.1.3 Model Performance Evaluation

In general, the random forest model shows high

accuracy while predicting the CO2 and PM2.5

emission. There is a strong correlation between the

prediction results of the model and the actual

observations which shows the burned area is and

important factor affecting emissions. However, there

is still development area in the prediction accuracy of

the model in extreme classes. More complex model

structure or more

characteristic variables could be

used to improve prediction performance in the

future.

ICDSE 2025 - The International Conference on Data Science and Engineering

396

3.2 Wildfire Scale Classification

Results

In this chapter, the global wildfire scale is made a

grade classification which is based on the ratio of

burnt area to pollutant emissions. At the same time

the spatiotemporal distribution characteristics of

different level of wildfire are analyzed. It is clear that

the impact of wildfire on environment is accurately

assess through comparing the grade of the resources.

Pollutant ratios and wildfire scale classifications

Figure 3 shows the result of classification of

wildfire size classification based on PM2.5 and CO₂

emissions and burned area ratio. It can be seen that

PM2.5 ratio fluctuates between 0.1 and 0.6. At the

same time CO2 ratio remains relatively stable at a low

level. Through analyzing all those ratios, wildfire

scale is divided into the following levels:

1. Low pollution level: PM2.5 ratio less than 0.2,

CO₂ ratio less than 0.2. These types of wildfires

typically burn a small area and have a limited impact

on air quality.

2. Medium pollution level: PM2.5 ratio between

0.2 and 0.4, CO₂ ratio between 0.2 and 0.4. These

wildfires burn a moderate area and have a significant

impact on local air quality.

3. High pollution level: PM2.5 ratio is higher than

0.4, and CO₂ ratio is higher than 0.4. These wildfires

burn large areas and have a serious impact on regional

and even global air quality.

Figure 3: the result of classification of wildfire size

classification based on PM2.5 and CO₂ emissions and

burned area ratio (Picture credit: Original)

The impact of wildfires on the environment can be

assessed more accurately with the help of the

classification of wildfire scale based on pollution

ratio. The high-level wildfires which occur in a

specific time and area have such severe impact on

global climate change and air quality. The future

result can get further combination with

meteorological data and human activity factors. With

these developments, accuracy of wildfire prediction

and management will be improved.

4 DISCUSSION

4.1 Result Discussion

This paper is based on the predicted result of the CO2

and PM2.5 emissions. While estimating the impact of

wildfires on global greenhouse gas concentrations, it

is clear that there is a significant positive correlation

between wildfire burned area and CO2 and PM2.5

emissions. Especially during severe pollution events,

the impact of wildfires on greenhouse gas

concentrations is more prominent. Combined with the

spatial and temporal distribution of wildfire-prone

areas, there are many potential drivers on climate

change, including increased extreme weather events,

increased dryness of vegetation, and human activities

interfering with natural ecosystems. Differences in

wildfire characteristics between regions are the main

reason in emission contributions, for example,

different vegetation types can affect combustion

efficiency and the types of emissions; climate

conditions can intensify the scale and frequency of

wildfires; and human activities can also have a

significant impact on wildfire emissions. The

interaction of these factors results in significant

differences in the contribution of wildfires to climate

warming in different regions. It shows great

prediction while using the random forest model on the

prediction aspect. The nonlinear relationships

between burning area and CO2 and PM2.5 emission

could be easily caught. However, there are also some

limitation of the random forest model, for example

the high reliance on data volume and relatively weak

interpret ability. In the future, there may be more

machine learning models used to deal with this

project which may make some comparison to find a

best fitted one. With the future effort, the predictive

performance will be improved

4.2 Study Limitations

4.2.1 Data Quality

The data used in this study may have issues with

uneven time spans or insufficient spatial coverage,

particularly in areas where wildfires are frequent but

monitoring capabilities are weak. Furthermore,

measurement errors in CO2 and PM2.5 emission data

may impact the model results, such as insufficient

sensor accuracy or biases introduced by data

interpolation methods.

Research on a Random Forest Regression Model for Climate Prediction in the Context of Wildﬁres

397

4.2.2 Model Selection

Although the random forest model performed well in

this study, it did not fully consider other potential

features (such as meteorological conditions and

vegetation types) and their interactions. Additionally,

median-based classification methods, while simple

and intuitive, may not adequately reflect the dynamic

changes in pollution levels. Future research could

explore the introduction of dynamic thresholds or

machine learning-based classification methods to

enhance classification accuracy.

4.3 Research

Future research could have further understanding on

several aspects of further research into wildfire

emissions and their impact on climate change. First of

all, more influencing factors could be added in order

to build multivariate regression or deep learning

models to more fully capture the complex

mechanisms of wildfire emissions. For example,

changes in climate could significantly influence the

wildfire combustion efficiency and emissions spread.

The differences in vegetation types affect the types of

fuels and emission intensity. Secondly, with the

combination of the climate model, the long term

influence could be studied on the impact of wildfire

emissions of greenhouse gas concentrations and

climate warming. Through simulating these, It can

assess the contribution of wildfire emissions to global

temperature increases and reveal their potential

changes under different climate conditions. After all,

while predicting the influence of wildfire emissions

to global temperature rise, evaluation of potential

effects of wildfire emissions should be made in order

to analyze the economic and environmental benefits

of international cooperation and localized control

measures. After combing the help of understanding

the reflection of wildfire on climate change. These

could provide stronger scientific support for

addressing climate change.

5 CONCLUSIONS

Deal to the dynamic regulation of ecosystems, the

existing information on species types and climate

conditions around the world. As a result of this, the

current rules of wildfire area burned and greenhouse

gas emissions still remains unchanged. As climate

change intensifies its impact on ecosystems, the way

of prediction of the impact of climate change on

greenhouse gas emissions, especially impact of

wildfires on emissions such as CO2 and PM2.5, will

make challenges facing by the climate protection

manager. Although the current research has shown

that there are Potential associations between wildfire

emissions and climate change. Considering the

uncertainty about climate change, this paper uses the

simplest representative, which are CO2 and PM2.5,

to predict the predicting the specific impact of

wildfires on greenhouse gas emissions as a

preliminary attempt.

This paper uses the simplest representative, which

are CO2 and PM2.5, to predict the predicting the

specific impact of wildfires on greenhouse gas

emissions as a preliminary attempt in this area. The

model evaluation results show that the mean square

error (MSE), mean absolute error (MAE) and

coefficient of determination (R²) all have high

prediction accuracy. The pollution classification

method based on Emission ratio intuitively reveals

the distribution characteristics of pollution levels in

time. New research ideas are provided to quantify the

impact of wildfire emissions on climate change,

which may better face the challenges of climate

change

On the scientific aspect, the result provides

database support and methodological references for

the prediction of global climate change, especially on

the aspect of discussing wildfire emissions and global

warming. It fills the gap in related research. At the

same time, on the policy level, this result provides a

basis for developing wildfire emission prevention and

control policies. Prioritize responding to severe

pollution incidents and reducing their harm to

ecosystems and human health. Scientific support is

also provided for the regional and global climate

policies optimization.

In the future research, Comprehensive analysis of

multiple factors, such as meteorological conditions,

vegetation types and human activities, should be

taken into consideration to reduce the impact of

confounding variables on experimental results. At the

same time, the prediction model could be constructed

as a more accurate one, which is specially designed

for this problem, to uncover the complex mechanisms

of wildfire emissions. What’s more, Regional

differences can be conducted based on wildfire

characteristics in different regions. In order to

improve these shortcomings, It is important to

promoting data sharing and interdisciplinary

collaboration. Combining the power of climatology,

ecology and data science to jointly address climate

change. These hard works will further develop the

understanding of understanding wildfire emissions

and their impact on climate change and provide

strong support for science-based responses to climate

change

ICDSE 2025 - The International Conference on Data Science and Engineering

398

REFERENCES

Borup, D., Christensen, B. J., Mühlbach, N. S., & Nielsen,

M. S. 2023. Targeting predictors in random forest

regression. International Journal of Forecasting, 39(2),

841-868.

Bot, K., & Borges, J. G. 2022. A systematic review of

applications of machine learning techniques for

wildfire management decision

support. Inventions, 7(1), 15.

Holmes, T. P., Huggett Jr, R. J., & Westerling, A. L. 2008.

Statistical analysis of large wildfires. The economics of

forest disturbances, 79, 59-77.

Huot, F., Hu, R. L., Goyal, N., Sankar, T., Ihme, M., &

Chen, Y. F. 2022. Next day wildfire spread: A machine

learning dataset to predict wildfire spreading from

remote-sensing data. IEEE Transactions on Geoscience

and Remote Sensing, 60, 1-13.

Koutsias, N., Martínez-Fernández, J., & Allgöwer, B. 2010.

Do factors causing wildfires vary in space? Evidence

from geographically weighted regression. GIScience &

Remote Sensing, 47(2), 221-240.

Liu, S., Xing, J., Westervelt, D. M., Liu, S., Ding, D., Fiore,

A. M., ... & Wang, S. 2021. Role of emission controls

in reducing the 2050 climate change penalty for PM2.

5 in China. Science of the Total Environment, 765,

144338.

McKenzie, D., Gedalof, Z. E., Peterson, D. L., & Mote, P.

2004. Climatic change, wildfire, and

conservation. Conservation biology, 18(4), 890-902.

Moritz, M. A., Morais, M. E., Summerell, L. A., Carlson, J.

M., & Doyle, J. 2005. Wildfires, complexity, and highly

optimized tolerance. Proceedings of the National

Academy of Sciences, 102(50), 17912-17917.

Nunes, L. J. 2023. The rising threat of atmospheric CO2: a

review on the causes, impacts, and mitigation

strategies. Environments, 10(4), 66.

Short, K. C. 2014. A spatial database of wildfires in the

United States, 1992-2011. Earth System Science

Data, 6(1), 1-27.

Research on a Random Forest Regression Model for Climate Prediction in the Context of Wildﬁres

399