Data Analysis of the Land Fire in California

Wenhan Zheng

Faculty of Science, University of Bristol, Bristol, U.K.

Keywords: Wildfire Prediction, Landfire Risk Assessment, Machine Deep Learning, Predictive Modeling, Climate Data

Analysis.

Abstract: On a typical afternoon in Los Angeles, California, on January 7, 2025, an unexpected natural disaster - the

"land fire" - broke out. This epidemic is slightly different from the past. The fire was not brought under timely

control, but continued to spread and decay, losing control. This article will attempt to focus on several types

of data collected so far. Firstly, it analyzes the linear relationship between firefighting costs and time, then

summarizes the different causes and percentages of fires in California and analyzes data on volcanic coverage

areas and total numbers before 2023. These data involve the analysis and prediction of wildfires that have

occurred in California over the past five years, including factors such as high-frequency periods, scale, and

causes, as well as attempts to use mathematical models to minimize and predict the possible consumption of

human and financial resources. Finally, some suggestions and prospects for fires were provided.

1 INTRODUCTION

To date, in the case of the previously mentioned

California mountain fires, strong winds continue to

accelerate the spread of the blazes, which have

affected more than 26,000 acres and forced the

evacuation of approximately 150,000 people

(Southern, Harding, Hurley, et al, 2025). So to better

prepare people to protect themselves in the face of

disaster, accurate fire prediction and the costs

involved have long-term implications for human

development. This research aims to optimise the

performance of the models by making more effective

predictions to provide more hilarious early warning

scenarios, allocate resources more efficiently and

reduce the likelihood of natural disasters threatening

people's lives (Hino & Field, 2023).

According to the annual statistics of natural

disasters in the United States, it is not difficult to find

that California, as a high incidence of hill fires, has

actually shown some problems in recent years, for

example, by the growing impact of greenhouse gas

emissions, the global warming has also made some

changes in the temperature of California, and with the

increase in human activities it is inevitable, because

people need to make a reasonable allocation of

https://orcid.org/0009-0004-7902-2954

limited resources. Rational deployment, so we should

prevent or reduce the damage caused by disasters

more efficiently (Abatzoglou & Williams, 2016).

According to research findings, it is not difficult to

see that as global temperatures rise, resulting in a

slowdown in the accumulation of water vapour in the

clouds, the most direct impact of reduced

precipitation is that the water content of the

vegetation will be reduced, which is more for the

occurrence of fires to create conditions conducive to

the occurrence of fires. Therefore, this has led to the

outbreak of mountain fires in high frequency.

Since the 1970s, the number of such disasters has

increased significantly, more than half of the increase

being due to man-made climate change. The area

covered by hill fires has also nearly doubled during

this period, and over the years, the occurrence of hill

fires has led to a very large economic expenditure,

resource loss, and even casualties.

The U.S. California Mountain fire is not the first

time. Still, in recent years, as the fire is more and more

difficult to control, the California mountain fire that

occurred in early 2025 brought unprecedented havoc

to the local people. Therefore, when more and more

people began to pay attention to this phenomenon, it

is not difficult to find that early many scholars have

noticed such a problem, hoping to have through the

292

Zheng, W.

Data Analysis of the Land Fire in California.

DOI: 10.5220/0013688200004670

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 292-298

ISBN: 978-989-758-765-8

study of mountain fires occurring in time or spatial

patterns. Among the many findings, Westerling et al.

(2006) used historical data to analyse seasonal

patterns, turning a vague abstract perception into a

more intuitive and rational mathematical model, he

argued that the first half of the year, in fact, that is,

spring and summer, California's mountainous and

forested areas are usually high-frequency occurrence

of mountain fires, summing up the occurrence of the

California Mountain Fire The reason for the

occurrence of the California mountain fires may not

be because of the season, but there must be some kind

of strong correlation between the two.

In subsequent analyses, some scholars have

suggested that a mixture of more frequent daily

human activities and insufficient management and

protection of forest resources have indirectly

contributed to the problem.

This paper focuses on the collected data on

wildfires that have occurred in California over the

past five years, including analysis of factors such as

high-frequency periods, scale, and causes of wildfires,

as well as attempts to use mathematical models to

minimize and predict the potential consumption of

human and financial resources.

2 RELATED WORKS

In recent years, with the continuous development of

data science and the unremitting efforts of scholars,

people have tried to believe that mathematical models

can help people solve many life problems. Predictive

modelling is the best solution to meet various natural

geological disasters and other emergencies. In the

paper of Abatzoglou & Kolden (2013), they took and

analyzed various collected data and constructed a

more perfect statistical model for the prediction of

mountain fires, because in the whole process, they

took into account as much as possible all kinds of

variables that may affect the results, topographic

factors, wind factors, climate conditions are all taken

into account, which makes the prediction based on

real data as a support to the model. This makes the

prediction models supported by real data more

accurate and more convincing.

Later on, with the continuous updating and

iteration of observation technology, remote sensing

technology can be more powerful through satellites

outside the sky, better use of spatial distribution, real-

time follow-up of various data, as well as UHF

transmission of real-time monitoring data through

satellites have become the best aid to assist in the

judgement process, but some human factors will

greatly affect the analysis process and results, which

is an insurmountable problem for the time being. This

is an insurmountable problem. But the emergence of

remote sensing technology, to a large extent, in

addition to human influence, completes the task of

data collection. And some scholars have borrowed

such remote sensing data combined with data models

to analyze the high-risk areas of the California

wildfires, and ultimately succeeded in predicting the

outbreak of several areas.

The deep learning model of the machine has made

the accuracy of the model in the prediction process

much better than before. Many researchers use

algorithms such as SVM (SVM: Support Vector

Machine), RF (Random Forest), etc. to perform

simulations and give the final results of the algorithm.

This process is also due to the rise of machine self-

learning technology, which helps research teams to

continuously improve and optimize, which is very

helpful to analyze and find out the potential factors

affecting the prediction results, which are different

from the normal logic of human thinking, and may be

overlooked.

The use of mathematical models is not really out

of reach for humans, on the contrary as early as in the

paper by Green, Kaiser & Shenton (2020) the

evolution of the regions around the mountain fires has

been predicted through the analysis of various

graphical data and modeled by deep convolutional

neural networks with very successful results in the

Sierra Nevada of the Western Sierras, so it is all the

more important to try to find commonalities of

successful cases in this process, which is very

meaningful for the development of new Machine

Deep Learning models. This process of finding

commonalities in successful cases is very relevant for

the development of new Machine Deep Learning

models.

Overall, as of the end of 2019, the researchers

have counted around 300 publications on the most

commonly used machine learning methods across a

wide range of problem domains, including Max Ent,

Artificial Neural Networks, Decision Trees, Genetic

Algorithms, and so on. This means that there are

differences in the ability of each model to learn and

produce results under different conditions and

environments, which leaves the human race to think

about progress and hopefully continue to develop or

derive a completely new algorithm (Jain et al., 2020).

More advanced deep learning methods such as

CNNs (Convolutional Neural Networks) and LSTMs

(Long Short-Term Memory Networks) are being

developed to optimize both temporal and spatial data,

Hosseini et al. (2020). In this process, through the real

Data Analysis of the Land Fire in California

293

occurrence of mountain fires after data collection, and

the prediction of comparison, it is not difficult to find

that the LSTM model for the California mountain fire

time series prediction, in the long time dependence of

the performance is pretty good. In recent years, with

the continuous development of data science and the

unremitting efforts of scholars, people have tried to

believe that mathematical models can help people

solve many life problems. Predictive modelling is the

best solution to meet various natural geological

disasters and other emergencies. In the paper of

Abatzoglou & Kolden (2013), they took and analyzed

various collected data and constructed a more perfect

statistical model for the prediction of mountain fires,

because in the whole process, they took into account

as much as possible all kinds of variables that may

affect the results, topographic factors, wind factors,

climate conditions are all taken into account, which

makes the prediction based on real data as a support

to the model. This makes the prediction models

supported by real data more accurate and more

convincing.

Later on, with the continuous updating and

iteration of observation technology, remote sensing

technology can be more powerful through satellites

outside the sky, better use of spatial distribution, real-

time follow-up of various data, as well as UHF

transmission of real-time monitoring data through

satellites have become the best aid to assist in the

judgement process, but some human factors will

greatly affect the analysis process and results, which

is an insurmountable problem for the time being. This

is an insurmountable problem. But the emergence of

remote sensing technology, to a large extent, in

addition to human influence, completes the task of

data collection. And some scholars have borrowed

such remote sensing data combined with data models

to analyze the high-risk areas of the California

wildfires, and ultimately succeeded in predicting the

outbreak of several areas.

The deep learning model of the machine has made

the accuracy of the model in the prediction process

much better than before. Many researchers use

algorithms such as SVM (SVM: Support Vector

Machine), RF (Random Forest), etc. to perform

simulations and give the final results of the algorithm.

This process is also due to the rise of machine self-

learning technology, which helps research teams to

continuously improve and optimize, which is very

helpful to analyze and find out the potential factors

affecting the prediction results, which are different

from the normal logic of human thinking, and may be

overlooked.

The use of mathematical models is not really out

of reach for humans, on the contrary as early as in the

paper by Green, Kaiser & Shenton (2020) the

evolution of the regions around the mountain fires has

been predicted through the analysis of various

graphical data and modeled by deep convolutional

neural networks with very successful results in the

Sierra Nevada of the Western Sierras, so it is all the

more important to try to find commonalities of

successful cases in this process, which is very

meaningful for the development of new Machine

Deep Learning models. This process of finding

commonalities in successful cases is very relevant for

the development of new Machine Deep Learning

models.

Overall, as of the end of 2019, the researchers

have counted around 300 publications on the most

commonly used machine learning methods across a

wide range of problem domains, including Max Ent,

Artificial Neural Networks, Decision Trees, Genetic

Algorithms, and so on. This means that there are

differences in the ability of each model to learn and

produce results under different conditions and

environments, which leaves the human race to think

about progress and hopefully continue to develop or

derive a completely new algorithm (Jain et al., 2020).

More advanced deep learning methods such as

CNNs (Convolutional Neural Networks) and LSTMs

(Long Short-Term Memory Networks) are being

developed to optimize both temporal and spatial data,

Hosseini et al. (2020). In this process, through the real

occurrence of mountain fires after data collection, and

the prediction of comparison, it is not difficult to find

that the LSTM model for the California mountain fire

time series prediction, in the long time dependence of

the performance is pretty good.

3 RESEARCH METHODOLOGY

3.1 Data Overview and Collection

The dataset used in this analysis contains information

such as the time of the hill fire, the exact location, the

cause of the fire, the duration, the estimated spread

area, the number of casualties, the expected damage

to property, casualty statistics, etc. Before analyzing

the acquired data, the analysis focused on the fires

that occurred in Los Angeles, California, USA,

because of the random nature of the fires.

1. The time of the occurrence

2. Duration of the disaster

3. The scene of the incident

4. Types of incentives

ICDSE 2025 - The International Conference on Data Science and Engineering

294

5. Casualty statistics

3.2 Statistic Analysis

In comparison with the 5-Year Average, the 2025

statistics show a much higher number of fires (164 vs.

76) and acres burned (40,695 vs. 30) compared to the

5-year average. This further confirms the severity of

the current fire season in comparison to previous

years.

Figure 1: Linear fit of costs (Picture credit: Original)

4 RESULTS OF THE

EXPERIMENT

4.1 Analysis of the data

With historical statistics to back up the analysis, it is

easy to see that the level of human activity is different

from a century ago, excessive behavioural activity

has already had a non-negligible impact on the results

of the model, and the anthropogenic variables, as

reported in (Mann et al.,2016), are shown to at least

make the significance of climatic factors misjudged

i.e. climatic conditions are misjudged to be the trigger

of the events such as those that lead to the mountain

fires probability would increase by 24% close to a

quarter. And that would lead the study in a direction

skewed away from the facts. So this emphasizes the

model and tests the ability of the model to combine

multiple factor variables and also puts a higher

demand on the accuracy of the correlation of the

variables.

Figure 1 shows a positive correlation, meaning

that as the years progress, the total firefighting costs

also exhibit an upward trend. The trend is represented

by the red linear fit line. Contrasting the data with

years, the number of wildfires in 2023 is below the

five-year and 10-year averages, and down from

previous years. The Correlation (R-value) shown in

the graph represents a linear fit, and visually, it shows

a linear relationship, indicating that the data points

generally follow an increasing trend. There are also

outliers that could not be ignored. During the

preprocessing part of the data before fitting the linear

relationship, it was calculated that the outliers occur

more frequently in the early 1980s and 2000s. They

are completely off the predicted straight line and

outside the IQR (Interquartile Range). The straight

line fitted by the data calculation, analysis, and most

of the cases are between the real data high and low

sides, the deviation is not big, which shows that the

linear model is still good for the prediction of the

expenditure model fitting effect.

4.2 Prediction and Conclusion

In the U.S. statistics show that most of the triggers of

life, in fact, are man-made, and common causes of

fires include high temperatures, low humidity in

forested vegetation, and thunderstorms in which sky

lightning strikes flammable materials or man-made,

but surprisingly the area burned and covered by

mountain fires caused by man-made causes is more

than twice the area burned by natural causes (Balch et

al.,2017).

The predictive results of the entire model appear

to be quite good and should continue to be successful

in predicting fire protection expenditures in the

coming years, but if more complex models or

variables can be considered, the predictions and

models will be more reliable.

Figure 2: Proportion of the causes (Picture credit:

Original)

Figure 2 gives a very clear picture of the different

types of causes of the California fires, and the

Data Analysis of the Land Fire in California

295

percentages. Of these, 71.9% are attributed to Arson,

Smoking, and Open Burning, which together make up

the most significant contributors, with the remainder

being mostly caused by aging machinery and

equipment, or damaged energy equipment.

The above statement does not mean that there is

no link between climate change and the frequency of

hill fires. On the contrary, in Turco et al (2023), it is

argued that for nearly a quarter of a century, between

1996 and 2020, the number of hill fires in the

California region of the United States was five times

higher than the number of fires that had occurred

between 1971 and 1995, and that these changes were

most directly attributable to the climate change that

was evident in the investigations and modelling

analyses.

Figure 3: Comparison of two kinds of wildwires

(Picture credit: Original)

Figure 3 shows the number of California fires

caused by both human-caused and natural factors

over the years, with the former portion being

significantly higher than the latter portion. Human-

caused factors (especially Arson, Open Burning, and

Smoking) are the primary drivers Human-caused

factors (especially Arson, Open Burning, and

Smoking) are the primary drivers behind wildfires in

California, accounting for over 75% of total fires.

In short, based on the data provided by CAL FIRE

each year, it is easy to see that fire protection

expenditures have been on the rise, and accordingly,

some policies or adjustments may be needed to help

better manage the situation, such as severe penalties

for arsonists, harsher fines, or more maintenance and

improvement of forests. For those outliers, although

they may not be used by the model, if they occur

consecutively, they indicate that the model is

inappropriate or that something very bad is going to

happen, which can serve as a good warning.

However, because the mountain fires occurring in

early 2025 are different from previous cases, they will

be ignored as outliers during data processing so as not

to affect the results of the analyses, and the paper will

concentrate on analyzing the complete data before

2023. As the figure 4 and figure 5 below shows.

Figure 4: Total number of the fires (Picture credit:

Original)

Figure 5: Area covered by hill fires

(Millions of Acres) (Picture credit: Original)

The horizontal dashed lines represent averages

over the past five and ten years, providing a clear

visual reference for 2023 comparison with historical

data.

5 SUGGESTIONS

5.1 Enhanced Fire Prevention and

Awareness Campaigns

Enhanced Fire Prevention and Awareness Campaigns

ICDSE 2025 - The International Conference on Data Science and Engineering

296

Strengthening the publicity and education on fire

prevention awareness from childhood, and reasonable

community-focused education are very targeted

preventive activities, which can effectively reduce the

number of hill fires caused by the human factor,

because the human factor is very often unconsciously

caused by human beings seemingly small behavior.

This is an approach that requires long-term persistent

supervision, but the effect will not be immediate.

5.2 Stronger Regulations and

Enforcement

Human-caused fires must be dealt with seriously, and

regulations must be strengthened, while utilities must

reinforce infrastructure such as fences, power grids

and walls to minimize the possibility of human

influence. Secondly, clearer regulations and penalties

should be added for the use of pyrotechnics and

automotive equipment.

5.3 Improve Firefighting and

Emergency Response

While maintaining current firefighting expenditures,

the use of various drone technologies, similar to the

class of drones, replaces humans in a variety of high-

risk rescue activities, data-monitoring activities,

emergency response, and other tasks. All of these

need more funding as support. Because robots can

detect fires earlier and more timely, at an earlier time

to contain the spread of fire.

Secondly, for accident prone areas, the use of

cameras should be more widespread and the

implementation of remote sensing technology is

imperative. Similarly, in the face of a lack of

materials, the optimal allocation of resources is also

necessary, as long as this can be avoided, the dilemma

of emergency rescue supplies. Emergency evacuation

drills, while enhancing people's self-protection, can

also make the face of disaster, human beings are

harmed can be reduced to a minimum.

6 CONCLUSIONS

Based on the statistical results, graphs, etc., it is clear

that the frequency of hill fires in California has been

escalating in recent decades. In particular, the area

covered by each fire has been expanding, and the

summer burned area in northern and central

California has increased about five-fold from 1996-

2021 compared to 1971-1995. Secondly, for a

seasonal problem such as hill fires, the extreme

seasonal anomalies that are generated illustrate how

the severity of hill fires has repeatedly challenged the

limits of human control over nature. These anomalies

need to be explained, and the dramatic changes in

climatic conditions leading to higher temperatures

and drying out of the ground as a direct result of the

fires are lengthening the duration of the high fire

season.

And while trying to find ways to control the fire,

humans should also face their problems, according to

the results of the data mediation shows that about

85% of the mountain fires are for ignition caused in,

human causes including accidents as well as

negligence but are not an excuse for humans to ignore

the problem.

For the existing mountain fire prediction models,

continuous optimisation and technological updates

are also important. First of all, AI algorithms should

be integrated into the process of predictive algorithms

to improve the ability to sense the state of affairs, so

that he can be the first time through the network to

obtain a variety of data at the same time, using their

arithmetic power to analyse satellite imagery,

mapping, providing rescue options route, and so on.

Similarly, it is important to continue to develop

hybrid modelling, as the models I have used in this

thesis are the most basic data analysis models, which

can vary greatly in accuracy when faced with

multivariate, dynamic data problems. In conclusion,

improving the entire algorithmic model is also a

multifaceted endeavour that needs to be taken in

tandem to better protect humans from themselves.

REFERENCES

Abatzoglou, J. T., & Kolden, C. A. 2013. Climate change

in western US deserts: Potential for increased wildfire.

International Journal of Wildland Fire, 22(5), 635-645.

Abatzoglou, J. T., & Williams, A. P. 2016. Impact of

anthropogenic climate change on wildfire across

western US forests. Proceedings of the National

Academy of Sciences, 113(42), 11770-11775.

Balch, J. K., Bradley, B. A., Abatzoglou, J. T., Nagy, R. C.,

Fusco, E. J., & Mahood, A. L. 2017. Human-started

wildfires expand the fire niche across the United States.

Proceedings of the National Academy of Sciences,

114(11), 2946–2951.

Faramarzi, H., Hosseini, S. M., Pourghasemi, H. R., &

Farnaghi, M. 2021. Forest fire spatial modelling using

ordered weighted averaging multi-criteria

evaluation. Journal of Forest Science, 67(2), 87-100.

Green, M. E., Kaiser, K., & Shenton, N. 2020. Modeling

wildfire perimeter evolution using deep neural

networks. arXiv. https://arxiv.org/abs/2009.03977

Data Analysis of the Land Fire in California

297

Hino, M., & Field, C. B. 2023. Fire frequency and

vulnerability in California. PLOS Climate, 2(2),

e0000087.

Jain, P., Coogan, S. C. P., Subramanian, S. G., Crowley, M.,

Taylor, S., & Flannigan, M. D. 2020. A review of

machine learning applications in wildfire science and

management. Environmental Reviews, 28(4), 478–505.

Mann, M. L., Batllori, E., Moritz, M. A., Waller, E. K.,

Berck, P., Flint, A. L., Flint, L. E., & Dolfi, E. 2016.

Incorporating anthropogenic influences into fire

probability models: Effects of human activity and

climate change on fire activity in California. PLOS

ONE, 11(4), e0153589.

Southern, K., Harding, D., Hurley, B., & Blanco, A. 2025.

California fire forces evacuation of Pacific Palisades in

Los Angeles. The Times.

Turco, M., Abatzoglou, J. T., Herrera, S., Zhuang, Y., Jerez,

S., Lucas, D. D., AghaKouchak, A., & Cvijanović, I.

2023. Anthropogenic climate change impacts

exacerbate summer forest fires in California.

Proceedings of the National Academy of Sciences,

120(25), e2213815120.

ICDSE 2025 - The International Conference on Data Science and Engineering

298