The Forest Fire Dynamic Change Influencing Factors and the Impacts on Gross Primary Productivity in China

Abstract

Forest fire as a common disturbance has an important role in the terrestrial ecosystem carbon cycling. However, the causes and impacts of longtime burned areas on carbon cycling need further exploration. In this study, we exploit Thematic Mapper (TM) and Moderate Resolution Imaging Spectroradiometer (MODIS) data to develop a quick and efficient method for large-scale forest fire dynamic monitoring in China. Band 2, band 4, band 6, and band 7 of MOD09A1 were selected as the most sensitive bands for calculating the Normalized Difference Fire Index (NDFI) to effectively estimate fire burned area. The Convergent Cross Mapping (CCM) algorithm was used to analyze the causes of the forest fire. A trend analysis was used to explore the impacts of forest fire on Gross Primary Productivity (GPP). The results show that the burned area has an increased tendency from 2009 to 2018. Forest fire is greatly influenced by natural factors compared with human factors in China. But only 30% of the forest fire causes GPP loss. The loss is mainly concentrated in the northeast forest region. The results of this study have important theoretical significance for vegetation restoration of the burned area.

1. Introduction

Forest fire occurs globally on various scales every year, causing economic, social, ecological, and environmental damage [1]. It is caused by natural phenomena as well as anthropogenic activities. It results in severe damage to wildlife, fertile forest floors, timber, human property, and certain rare plant species [2]. As a result of the pervasive harmful effects due to forest fires, they have received increasing recognition in past recent years. Therefore, precise mapping and monitoring of the location and temporal distribution of wildfires are important. In addition, forest fire as one of the most significant natural disturbance processes would also modify the structure and the composition of the vegetation [3,4]. It is closely related to the carbon cycling and greenhouse gas emissions [5,6]. The forest fire can burn away dead or decaying vegetation, facilitating the growth of new trees and burned trees or the soil surface, releasing large amounts of carbon. Forest fire disturbances have an important impact on ecosystem stability and renewal, and succession of forest ecosystems. [7,8]. The effects of forest fires on ecosystems are not clear for a long time. Long-time burned area mapping is a critical factor to investigate the causes of the forest fire and the impacts of forest fires on Gross Primary Productivity (GPP). This research provides theoretical support for intelligent management and prediction of forest fires to achieve carbon neutrality in China.

Traditionally, a field survey was used to monitor forest fires. But this method is time-consuming and laborious and lacks spatial information. To overcome this problem, remote sensing is one of the most efficient and cost-effective techniques for fire detection and mapping. At present, there are so many forest fire remote sensing monitoring products that provide fire data sources, but these data have their limitations in terms of both spatial and temporal resolution. Fornacca et al. (2017) compared four common fire products such as MCD45A1, MCD64A1, MCD14ML, and Fire_CCI [9]. Each product has its limitations in terms of accuracy in different fire ranges. Therefore, the timely and accurate mapping of burned areas is essential for fire management and climate change. The remote sensing technique provides a labor-efficient method to quickly locate the burned area distribution and detect the impacts and causes of forest fires [10,11]. In the past decades, extensive studies have been carried out on the detection of burned areas by remote sensing [3,10,12]. Medium-resolution satellites such as the Landsat series are widely utilized to identify the burned area. However, they cannot accurately depict the burned area change because of the limitations of the temporal resolution. Combining information from different sensors can fill the gaps between high temporal resolution and medium spatial resolution [13]. High temporal resolution satellite images such as MODIS and AVHRR greatly improve the capability to detect the long-time burned area [14,15,16]. Accurate forest fire distribution provides the basis for subsequent studies. It is significant to acquire the long-time fire distribution for studying its causes and effects.

Forest fire occurs every year in China, especially during the dry season. These fires are due to various factors such as dry weather, flammable materials, and human action [13]. The natural factor is the main determinant of natural forest fire [17]. Climate change affects the occurrences and dynamics of fire by changing meteorological factors such as air temperature, precipitation, and humidity [18]. The two possible main factors associated with climatic change are temperature and precipitation [19]. Pausas et al. (2004) found that the temperature and precipitation were significantly correlated with the burned areas [20]. Westerling et al. (2006) found that the forest fire was caused by reduced winter precipitation together with earlier spring snowmelt [21]. Lehmann et al. (2014) found that burned areas in Australia had a strong correlation with the average precipitation [22]. Wu et al. (2014) found that climate was the primary factor compared with human activity [23]. To sum up the above, the mechanisms and interactions leading to forest fires in China are poorly understood. The causes of forest fires need further exploration.

Forest ecosystem carbon cycling is an important component of the entire terrestrial ecosystem. The most significant impact of a forest fire can be seen in vegetation. Forest fire disturbances have a significant impact on ecosystem stability and sustainability [7,24]. On the one hand, plants usually die instantly due to considerably severe forest fires. Frequent and high-intensity fires will lead to permanent changes in ecosystems and their components [25]. On the other hand, forest fires have been an important mechanism for generating ecological succession by acting as an environmental filter, selecting species, and shaping ecosystem communities [26]. Forest fire contributes significantly to climate change, consuming and transferring carbon to the atmosphere [27]. However, carbon change is usually ignored after a forest fire. Observational data suggest that vegetation growth and soil carbon content gradually recover over the following years [28]. If ecosystems can be restored to their pre-disturbance state, the terrestrial ecosystem carbon cycling remains in long-term dynamic equilibrium [29]. The effects of forest fire disturbance on ecosystems are not clear for a long time.

The study proposes a new method to map burned areas using multisensor remote sensing data by taking advantage of the high temporal resolution of Moderate-Resolution Imaging Spectroradiometer (MODIS) and the medium spatial resolution of Thematic Mapper (TM). In this study, a new spectral index called Normalized Difference Fire Index (NDFI) was derived from MODIS surface reflectance data to timely and accurately acquire the forest fire distribution and its dynamic change. The objectives of this study are as follows: (1) to accurately obtain the spatiotemporal variations of fire distribution by NDFI; (2) to identify the causes of forest fires; and (3) to explore the impacts of forest fire on GPP.

2. Materials and Methods

2.1. Data and Data Processing

MOD09A1 data were downloaded from the National Aeronautics and Space Administration (NASA) (http://earthdata.nasa.gov/, accessed on 30 December 2020). It provides 8-days composite with 500 m spatial resolution data.

TM data with a spatial resolution of 30 m and a return cycle of 16 d were downloaded from the Institute of Remote Sensing and Digital Earth (http://ids.ceode.ac.cn/, accessed on 30 December 2020) and used to validate the forest fire distribution.

Meteorological data of temperature and precipitation datasets with 1 km spatial resolution were downloaded from the National Ecosystem Science Data Center (NESDC). The dataset is interpolated by the data from the National Meteorological Information Center (NMIC) of the China Meteorological Administration and the Daily Global Historical Climatology Network-Daily (GHCN-D).

Gross Primary Productivity (GPP) and nighttime-light data from 2009 to 2018 were downloaded from National Tibetan Plateau Data Center. The Advanced Very High-Resolution Radiometer (AVHRR) data of remote sensing and hundreds of flux stations around the world were used to generate the global high-resolution long-time series GPP. The unit of this data is gcm⁻² with a spatial resolution of 0.05 degrees. The nighttime-light data were produced by the convolutional Long Short-Term Memory network method.

2.2. Burned Area Monitoring Methods

Remote sensing is a more appropriate approach for large-scale and long-time studies. The monitoring method was developed by Feng. The methodology used to estimate the forest fire distribution from 2009 to 2018 was based on the approach proposed by Feng et al. (2016) [30]. This method is beneficial to the dynamic monitoring of long-time series over large areas. The most sensitive bands were used to construct the NDFI to effectively estimate the burned area distribution in this study [30]. The main processes include (1) Band 2, Band 4, Band 6, and Band 7 of MOD09A1 were selected as the most sensitive bands to forest fire; (2) the sensitive bands were selected to calculate NDFI for monitoring the burned area; (3) Convergent Cross Mapping (CCM) algorithm was used to analyze the causes of forest fires; and (4) trend analysis was used to explore impacts of forest fire on GPP.

In order to effectively monitor burned area distribution, TM2, TM4, TM5, and TM7 of TM or MODIS2, MODIS4, MODIS6, MODIS7 of MODIS were selected to calculate NDFI using the following equation:

$$NDFI=\frac{\left|TM7\right.-\left.TM5\right|}{TM4+TM2}$$

(1)

$$NDFI=\frac{\left|MODIS7\right.-\left.MODIS6\right|}{MODIS4+MODIS2}$$

(2)

where NDFI is the Normalized Difference Fire Index; TM7 is band 7 of TM; TM5 is band 5 of TM; TM4 is band 4 of TM; TM2 is band 2 of TM. MODIS7 is band 7 of MODIS; MODIS6 is band 6 of MODIS; MODIS4 is band 4 of MODIS; MODIS2 is band 2 of MODIS.

2.3. Causes and Impacts of Forest Fire Analysis Methods

CCM algorithm proposed by Sugihara et al. was initially applied to detect the causality of variables in complex ecosystems. It is a powerful new methodological approach that can help distinguish causality from spurious correlation in time series from dynamic systems. The technique is based on the idea that causation can be established if states of the causal variable can be recovered from the time series of the affected variable [31]. It uses Takens’ idea to detect if two variables belong to the same dynamical system. It is designed for causal discovery between coupled time series for which Granger’s method for detecting causality is shown to be unreliable. CCM is based on an algorithm that compares the ability of lagged components of one process to estimate the dynamics of another. In ecology, these processes might represent time series observations of environmental data, such as temperature, or of species data, such as population abundance [32]. In this study, we did not replace causality with correlation. CCM algorithm was used to analyze the causes of the forest fire. The forest fire, temperature, precipitation, and nighttime-light time series data were used to derive the causality.

A slope map was calculated at pixel scales to evaluate the spatial change of burned area and GPP using the following equation:

$$slope=\frac{n{\sum }_{j=1}^{n}jy-{\sum }_{j=1}^{n}j{\sum }_{j=1}^{n}y}{{\sum }_{j=1}^n{j}^2-{({\sum }_{j=1}^{n}j)}^2}$$

(3)

where n is the number of years, here n = 10; y is the burned area (GPP) in the jth year; the slope is the fitted slope of n years. The slope map indicates the variation in trend and range.

3. Results

3.1. Sensitive Bands Selection of Forest Fire

In this study, the huge forest fire that originated in China’s Daxinganling Mountains in 1987 was selected as the reference area which obviously reduces the influence of mixed pixel. Twenty random points were generated in the burned area (Figure 1). The DN values (data value range: 0~65,535) of these points in different bands of TM in different seasons are shown in

Table 1. Twenty selected random points values of TM in April.

Random Point	Band 1	Band 2	Band 3	Band 4	Band 5	Band 7
1	9523	10,286	10,777	13,353	15,720	12,999
2	9273	9832	10,101	13,024	14,135	11,495
3	9371	9797	9937	12,098	13,297	10,896
4	9315	9653	9812	12,420	13,729	11,055
5	9107	9449	9924	12,278	14,689	12,111
6	9530	9806	10,217	12,858	13,811	11,191
7	9506	9790	10,209	12,862	14,642	11,785
8	9139	9579	9758	12,133	13,045	10,932
9	9056	9626	9932	12,123	14,164	11,810
10	9542	10,130	10,634	14,231	15,157	12,229
11	9281	9948	10,490	14,294	14,701	11,821
12	9772	10,295	10,500	12,253	11,543	10,157
13	9462	9597	10,049	12,715	14,345	11,793
14	9134	9630	9794	12,574	13,009	10,909
15	9225	9751	10,464	13,177	16,109	12,983
16	9813	9998	10,236	11,943	11,844	10,152
17	9676	10,004	9966	12,698	12,870	10,739
18	10,016	10,465	11,052	13,461	15,553	12,366
19	9375	9845	10,106	12,861	12,574	10,748
20	9293	9682	9967	13,162	13,088	10,895

,

Table 2. Twenty selected random points values of TM in May.

Random Point	Band 1	Band 2	Band 3	Band 4	Band 5	Band 7
1	9120	9162	9267	9566	12,734	13,833
2	7500	7953	8252	8762	11,776	13,059
3	8710	8912	8853	9353	11,454	12,371
4	7546	7988	8120	8777	10,736	12,030
5	6882	7460	7844	8366	11,578	12,349
6	8623	8746	8845	9049	10,583	11,413
7	8615	8897	8840	9196	11,455	12,237
8	7082	7577	7943	8679	11,062	12,219
9	6759	7437	7826	8530	10,744	11,471
10	8594	8808	8917	9541	12,381	13,710
11	7890	8342	8555	9183	11,761	13,455
12	8413	8684	8668	9091	10,794	11,323
13	8293	8394	8422	8803	10,787	11,727
14	8612	8794	8615	9224	11,458	12,647
15	8172	8632	8630	9068	11,781	13,281
16	8707	8851	8917	9093	10,578	11,123
17	8717	8905	8840	9194	10,972	11,142
18	8440	8761	8849	9498	12,521	12,650
19	8541	8943	8734	9432	10,866	11,653
20	8440	8775	8978	9134	11,250	12,199

and

Table 3. Twenty selected random points values of TM in June.

Random Point	Band 1	Band 2	Band 3	Band 4	Band 5	Band 7
1	8297	8494	8370	9084	12,537	13,511
2	8275	8643	8776	9545	13,318	13,510
3	8284	8348	8243	8864	10,654	10,566
4	8366	8654	8785	9395	13,509	14,057
5	8464	8829	9068	10,166	14,288	14,329
6	8491	8648	8896	9606	12,488	13,162
7	8418	8800	8632	9008	10,751	11,524
8	8477	8512	8795	9707	13,315	13,919
9	8323	8676	8800	9557	13,604	14,465
10	8599	8765	8732	9437	12,791	13,455
11	8807	9059	9386	10,340	14,468	15,411
12	8643	8644	8759	9302	11,915	12,622
13	8850	9005	9196	10,216	15,731	16,260
14	8398	8578	8712	9279	12,027	12,510
15	8738	8876	9471	10,809	14,190	12,859
16	8493	8734	8839	9723	12,312	12,368
17	8416	8735	8707	9423	12,408	13,052
18	8550	8568	8429	9114	11,535	12,782
19	8729	8810	8901	10,207	14,624	14,952
20	8635	8797	9024	9904	13,662	14,272

. During these three months, forest changes can be very pronounced in the event of a fire. Therefore, these three months are the basic months for the occurrence of fires. The total correlation index(r) value of all types between band 5 and band 7 is the lowest; band 5 and band 7 exhibit a small disparity in their spectral responses to different land covers. The total correlation index(r) value of all types between band 2 and band 4 is the largest; band 2 and band 4 exhibit a large disparity in their spectral responses to different land covers (

Table 4. Correlation coefficient (r) value between different bands.

	Band 1	Band 2	Band 3	Band 4	Band 5	Band 7
Band 1	——	0.993081	0.991975	0.992561	0.908178	−0.63607
Band 2	0.993081	——	0.999959	0.999991	0.85274	−0.72228
Band 3	0.991975	0.999959	——	0.999989	0.847965	−0.72853
Band 4	0.992561	0.999991	0.999989	——	0.850461	−0.72528
Band 5	0.908178	0.85274	0.847965	0.850461	——	−0.25467
Band 7	−0.63607	−0.72228	−0.72853	−0.72528	−0.25467	——

). Therefore, these four bands are used to derive the Normalized Difference Fire Index (NDFI) in this study [33]. A lower NDFI value indicated a greater possibility of fire distribution.

3.2. NDFI Validity and Burned Area Recognition in China

Further, the huge forest fire that originated in China’s Daxinganling Mountains in 1987 was used to validate the validity of the NDFI. Burned area in China’s Daxinganling Mountains is shown in Figure 2. TM2, TM4, TM5, and TM7 of TM are used to calculate the NDFI image. A lower value indicated a greater possibility of burned area. This result is consistent with the actual situation. The MODIS data were used to calculate the NDFI for estimating the burned area at a large scale. The possible burned area distribution in China is shown in Figure 3. A lower value indicated a greater possibility of forest fires. The northeast forest fire in 2010 is used to validate the burned area result obtained from MODIS data. The validation result (Figure 4) shows that the result is consistent with the actual forest fire distribution. Meanwhile, monitoring results are consistent with existing studies [34].

3.3. Causes of the Forest Fire

CCM of the fire with temperature, precipitation and nighttime-light time series data was shown in Figure 5. Pearson’s correlation coefficient at the point of convergence for “B causes A” is greater than that for “A causes B”, where A represents forest fire changes and B represents the temperature, precipitation, and nighttime-light changes, respectively (Figure 5a–c). That means all these three factors can drive the forest fire. The temperature and precipitation represent natural factors. The nighttime-light represents the human factor [35]. Results for this CCM analysis suggest that forest fire is significantly affected by temperature and precipitation (Figure 5a,b). However, there is no obvious forcing for the burned area and nighttime-light (Figure 5c). This result indicates that forest fire is greatly influenced by natural factors compared with human factors.

3.4. Effects of the Forest Fire on Ecosystem Carbon Cycle

The slope map of burned area and GPP from 2009 to 2018 is shown in Figure 6. Forest fire has the decreased trend. The GPP loss areas caused by forest fires are the regions where forest fire increased and the GPP decreased (Figure 7). The analysis result shows that about 30% of the forest fire causes GPP loss. The loss is mainly concentrated in the northeast forest region. However, about 70% of the forest fire have no impact on GPP.

4. Discussion

4.1. Burned Area Distribution and Causes of Forest Fire

The burned area distribution is supported by previous studies. For example, Chen et al., (2017) have demonstrated that the burned area is mainly distributed in eastern China, especially in the forests of Heilongjiang Province in Northeast China [36]. Pang et al., denoted that the probability of forest fires in eastern China is higher than that in the western regions, and the probability of forest fires in the north and south is higher than that in Central China [37]. The results of these studies are consistent with our study. Meanwhile, forest fire is driven by various environmental factors [36]. Climatic change and anthropogenic factors are likely altering the fire distribution in most regions of China. The increasing forest fire in northeastern China may have resulted from an increased temperature and decreased precipitation and humidity [38]. In addition, longitude and latitude had the greatest influence on the occurrence of forest fires. This result is due to the uneven distribution of forest resources and regional differences in forest resources in China. The higher the vegetation cover the more likely they are to cause problems related to forest fires [37]. The Intergovernmental Panel on Climate Change states that “climate variability is often the dominant factor affecting large wildfires” [39]. Archibald et al., (2010) denoted that climate is a dominant control on fire activity, regulating vegetation productivity and fuel moisture [40]. Precipitation suppresses the forest fire activity and promotes flammable material accumulation. Additionally, the temperature provides burning conditions for the flammable materials. Human activity is now the primary source of ignition in tropical forests, savannas, and agricultural regions [41,42]. In our study, the forest fire primarily located in Northeast China is mainly caused by natural factors. This result is consistent with some previous studies. However, we only discuss the temperature and precipitation as the natural factor and the nighttime-light as the human action. More factors and quantitative analysis methods should be discussed. Meanwhile, different data sources will bring a lot of uncertainty to the results. Therefore, multi-source data on fire range identification and GPP impact factor identification process will be considered in future work.

4.2. Practical Implications of This Study

The forest fire as a major environmental and social issue has attracted widespread attention. It is significant to propose a simple and effective method to estimate the longtime burned area on a large scale. Many fire products can be obtained from the website. However, accurate and longtime burned area distributions are limited. It is necessary to acquire the longtime burned area distribution to analyze the causes of the forest fire and the impacts of forest fire on GPP. In fact, there are many factors affecting GPP, which can be divided into natural and human factors [43]. On the one hand, forest fire as an important disaster will bring a certain loss to GPP. On the other hand, fire as a common and natural disturbance in forest regions plays a critical role in determining the structure and composition of vegetation [44,45]. Many plants have acquired the adaptive ability to regenerate after a forest fire to keep dynamic disequilibrium [29]. Fires have been found to be important in maintaining vegetation diversity, structure, and functions in fire prone ecosystems such as savannas. Forest fires do not necessarily bring species diversity changes [46]. A recent study showed that savannahs store carbon despite frequent fires [47]. So not all forest fires have a negative impact on ecosystems. In our study, during the revegetation for many years, 70% of the forest fire has no impact on GPP. The result of this study is useful to improve the ecosystem model [48] and forest management [49]. It is significant for promoting carbon neutrality in China. However, we only analyzed the impacts of forest fires on GPP at a national scale ignoring spatial heterogeneity and influence factor differences in different regions. That is the limitation of this study. We will consider these differences in future work.

5. Conclusions

The proposed NDFI is able to obtain the forest fire distribution on a large scale and assess the variability simply and effectively. The forest fire change has an increased tendency from 2009 to 2018. These fires are significantly affected by temperature and precipitation compared with nighttime-light. It indicates that forest fire is greatly influenced by natural factors. But only 30% of the forest fire causes GPP loss. The loss is mainly concentrated in the northeast forest region. This study provides an effective way to understand the burned area dynamic changes, the causes of the forest fire and the impacts of forest fire on GPP. However, there are still many challenges and uncertainty to estimating forest fire distribution for long time series observation on large scale using remote sensing data. Multi-source data on fire range identification should be considered and more factors and quantitative analysis methods should be considered in further studies.