Towards understanding the environmental and climatic changes and its contribution to the spread of wildfires in Ghana using remote sensing tools and machine learning (Google Earth Engine)

ABSTRACT

Data processing and climate characterisation to study its impact is becoming difficult due to insufficient and unavailable data, especially in developing countries. Understanding climate's impact on burnt areas in Ghana (Guinea-savannah (GSZ) and Forest-savannah Mosaic zones (FSZ)) leads us to opt for machine learning. Through Google Earth Engine (GEE), rainfall (PR), maximum temperature (Tmax), minimum temperature (Tmin), average temperature (Tmean), Palmer Drought Severity Index (PDSI), relative humidity (RH), wind speed (WS), soil moisture (SM), actual evapotranspiration (ETA) and reference evapotranspiration (ETR) have been acquired through CHIRPS (Climate Hazards group Infrared Precipitation with Stations), FLDAS dataset (Famine Early Warning Systems Network (FEWS NET) Land Data Assimilation System) and TerraClimate platform from 1991 to 2021. The objective is to analyse the link and the contribution of climatic and environmental parameters on wildfire spread in GSZ and FSZ in Ghana. Variables were analysed (area burnt and the number of active fires) through Spearman correlation and the cross-correlation function (CCF) (2001 to 2021). The tests (Mann-Kendall and Sens's slope trend test, Pettitt test and the Lee and Heghinian test) showed the overall decrease in rainfall and increase in temperature respectively (−0.1 mm; + 0.8°C) in GSZ and (−0.9 mm; + 0.3°C) in FSZ. In terms of impact, PR, ETR, FDI, Tmean, Tmax, Tmin, RH, ETA and SM contribute to fire spread. Through the codes developed, researchers and decision-makers could update them at different times easily to monitor climate variability and its impact on fires.

KEYWORDS:

• Climate change
• Google Earth Engine
• mitigation
• machine learning
• wildfire
• Ghana

1. Introduction

The ever-changing global climate has repercussions on many things such as the world's population, natural resources and their various ecosystems thus contributing to the increase in risks and disasters around the world (Trenberth, Fasullo, and Shepherd 2015; Thomas et al. 2018; WMO 2021; Kushawaha et al. 2021; Wentz 2015) for example between 2000 and 2019 over 475,000 people lost their lives as a direct result of more than 11,000 extreme weather events globally and losses amounted to around US$ 2.56 trillion (CIR 2021). This change is based on the assessment of multiple lines of evidence, global warming of 2°C (IPCC 2021). To this end, concerns have become great in recent decades about the increase in this global surface temperature (Ham 2018; IPCC 2021; CIR 2021; United Nations Environment Programme 2022), Many changes in the climate system become larger in direct relation to increasing global warming. They include increases in the frequency and intensity of hot extremes, marine heatwaves, heavy precipitation, and, in some regions, agricultural and ecological droughts; an increase in the proportion of intense tropical cyclones; and reductions in Arctic Sea ice, snow cover and permafrost (Cai et al. 2021; Donat, Angélil, and Ukkola 2019; IPCC 2021). For example, every additional 0.5°C of global warming causes discernible increases in the intensity and frequency of hot extremes, including heatwaves (very likely), and heavy precipitation (high confidence), as well as agricultural and ecological droughts in some regions (high confidence) (IPCC 2021). This situation makes countries and areas vulnerable depending on their specific climatic conditions. For example, the countries most affected in 2019 were Mozambique, Zimbabwe as well as the Bahamas and for the period from 2000 to 2019 Puerto Rico, Myanmar and Haiti rank highest (CIR 2021). To this end, the Global Climate Risk Index indicates a level of exposure and vulnerability to extreme weather events, which countries should understand as warnings to be prepared for more frequent and/or more severe events in the future (CIR 2021). This global phenomenon of climate change and its components hardly spares any component of life on Earth (Smith and Hitz 2003; Almer, Laurent-Lucchetti, and Oechslin 2017; Baez et al. 2017). Observed climate changes may have already influenced wildfire potential over parts of the globe (Stocks et al. 1998; Gillett et al. 2004; Westerling et al. 2006), and projected changes in climate over the next century are hypothesised to significantly alter global wildfire regimes (Flannigan et al. 2009). Weather conditions that are favourable for the start and spread of wildfires are brought on by climate change (Jones et al. 2022). Combining adverse land use and land cover shifting patterns with particular climate zones makes them more susceptible to forest fires (Angra and Sapountzaki 2022). The Australian bushfires of 2019/2020, for example, happened during fire weather conditions that were significantly more likely owing to climate change. Additionally, several large wildfires that have occurred in recent years have done so (Jones et al. 2022). Weather conditions, including ambient air temperature, relative humidity, and wind direction and velocity affect fire behaviour (Trollope 1984; Trollope and Trollope 2002). According to Bilgili and Saglam (2003), wind conditions have a dominant influence on fire behaviour; particularly, wind speed affects the rate of fire spread and flame height. Also, relative air humidity and air temperature affect fire behaviour indirectly through their effect on the moisture content of vegetation and litter (Bond and van Wilgen 1996; Bond and Keeley 2005).

Africa in general, and West Africa in particular, which is largely made up of developing countries, is increasingly exposed to climatic hazards (Wascal 2021; Tarif 2022; Trisos et al. 2022), making its efforts to combat them difficult (Maino and Emrullahu 2022). Local climate variability or climate change is causing severe disruption resulting in drought, and famine due to insufficient agricultural production (ECOWAS 2021). In tropical savanna ecosystems, the fuel load is mainly composed of herbaceous vegetation that makes up 75–90% of total annual biomass (Garnier and Dajoz 2001) with high amounts of standing dead fuel, resulting in high combustibility and fire risk (Hennenberg et al. 2006). In addition to this favourable predisposition of vegetation cover to wildfires, climatic constraints contribute greatly to their spread (Hanan et al. 2021; Jones et al. 2022; Kays and Ward 2021). Additionally, the various future climatic scenarios predict that flame length would rise by 4.6% to 15.69% (Molina, González-Cabán, and Rodríguez y Silva 2019). Therefore, as a result of the increased frequency and intensity of fire weather, climate change is placing constant upward pressure on fire around the globe. This upward pressure will get stronger with each degree of global warming (Jones et al. 2022).

At a time of great climatic upheavals that cause floods, droughts and disastrous fires in forest areas, for example, Ghana is hardly spared but is also confronted with the impacts of extreme climate events. Thus, three major physical impacts of climate change have been identified in Ghana: temperature change, changes in rainfall and sea level rise (EPA). Some impact studies carried out in Ghana showed increased temperatures and evapotranspiration, decreased and highly variable rainfall patterns and more frequent and pronounced dry spells in the selected representative river basins of the country (CSIRWRI 2000; Kasei, Barnabas, and Ampadu 2014). In the analysis of historical records of Ghana, the observed rate of change in minimum temperature for the period 1960–2010 was 37% for the northern part (Guinea Savannah Zones). The mean minimum temperature over the Savannah Zone is projected to increase by 1.10°C by 2040 and the mean monthly maximum temperature is expected to increase by 1.2°C and 2.1°C by 2040 and 2060, respectively (Issahaku, Campion, and Edziyie 2016; Incoom, Adjei, and Odai 2020). Also, according to Laux, Kunstmann, and Bárdossy (2008); Armah et al. (2011) and Laube, Schraven, and Awo (2012), Climate variability is the basis for the decrease in rainfall and the increase in temperature in the ecological zones of the Guinean savannah of northern Ghana. However, the research done by Baidu et al. 2017 revealed that the transition zone (forest-savannah mosaic zone), recorded almost quite stable rainfall amounts for all seasons. The majority of the research mentioned a proven variability of climatic parameters (precipitation and temperature). These parameters influence in one way or another the parameters like Palmer Drought Severity Index (PDSI), relative humidity (RH), wind speed (WS), soil moisture (SM), actual evapotranspiration (ETA) and reference evapotranspiration (ETR). Analysing their contribution to fire spread is essential given the importance of fire in the agricultural and non-agricultural activities of the population, to alert them to the job in real-time. However, access to climate data is difficult in tropical environments (Dinku 2019; Overpeck et al. 2011; Dinku et al. 2018), which leads us to adopt new programming and machine learning technologies (Google Earth Engine and R language) in this study. These increasingly used approaches (Yang et al. 2022; Mutanga and Kumar 2019; Sazib, Mladenova, and Bolten 2018; Sidhu, Pebesma, and Câmara 2018; Xing et al. 2022) allowed us to initially analyse and characterise the aforementioned variables in the Guinea-Savannah and Forest-Savannah mosaic zone in Ghana (1991–2021). After conducting Spearman correlation analyses and using the cross-correlation function (CCF) with fire data (area burnt and active fires) to highlight the parameters that contribute most to the spread of wildfires in the study areas. The emergence of cloud computing platforms such as Google Earth Engine (GEE) is an important advance that facilitates remote sensing applications (Gorelick et al. 2017). Computationally intensive methods are executed using parallel processing in the Google Cloud, allowing the processing of massive satellite datasets over large areas in settings with limited computational and internet resources. Because of these advantages, GEE has been widely used to develop remote sensing applications in a variety of fields (Tamiminia et al. 2020).

2. Materials and methods

2.1. Study sites

The districts considered in Guinea-savannah (West Gonja, West Mamprusi) for this research are located between 10° 27′0″ and 8°92′0″ N latitude and 0°28′0″ and 2°18′0″ W longitude with an area of 13,880.41 km² (Figure 1). Guinea-savannah is occupied by extensive wooded savannahs characteristic of the Guinean region, and the Open Guinean Savanna (OGS), characterised by natural wooded savannahs, is invaded by cultivated land CILSS (2016). According to Menczer and Quaye (2006), the Guinean-Savannah zone consists of tall grasses growing between widely spaced trees.

Figure 1. Geographic situation of the studies zones and districts.

About Forest-Savannah, the districts considered (Sene, Afram Plain), for this research are located between 6°35′0″ and 8°9′0″ N latitude and 1°40′W and 0°31′0″ E longitude with an area of 13,880.41 km². Structured into several sub-zones, the Forest-Savannah zone is located in central Ghana and composed of the Main, Eastern and Central Transitional Zones (MTZ, ETZ and CTZ) with an intermediate climate with two rainy seasons.

2.2. Material and data processing

To assess and analyse the link between climatic, and environmental parameters and wildfires (burned area and active fires), the present study has focused on certain parameters, namely for the study period (1991–2021) for climate characterisation and from 2001 to 2021 for correlation analysis between the latter and fire data; rainfall (PR), maximum temperature (T_max), minimum temperature (T_min), average temperature (T_mean), Palmer Drought Severity Index (PDSI), relative humidity (RH), wind speed (WS), soil moisture (SM), actual evapotranspiration (ETA) and reference evapotranspiration (ETR). Regarding the fire data, they are derived from Modis images (MCD64A1 for burned area and MCD14ML for active fires) treated and processed according to the methodology used by Oliveras et al. (2014), (see data availability statement). These parameters are generally considered to be the climate environment components that most influence the behaviour of wildfires (Guiguindibaye, Belem, and Boussim 2013). The availability of field observation data is a major challenge in less developed countries, and this was also revealed in this research, also by other authors as well (Dinku 2019; Overpeck et al. 2011; Dinku et al. 2018; Cervigni et al. 2015; Dinku et al. 2014; Kasei 2010). For this purpose, the data used is retrieved through platforms that host satellite data through machine learning approaches using Google Earth Engine (GEE) (Table 1). GEE (https://code.earthengine.google.com), provides a scalable, cloud-based, geospatial retrieval and processing platform (Yang et al. 2022; Mutanga and Kumar 2019; Kumar and Mutanga 2018) and is free for in-cloud data access, processing and management (Tamiminia et al. 2020).

Table 1. Characteristics of the data used in this research and their references

Parameters	Dataset availability	Dataset resolution	Dataset provider and retrieval platform	References/citations
PR	1981-01-01T00:00:00Z–2022-12-31T00:00:00	0.05° grid monthly and daily temporal resolution and a ∼4 km (1/24th degree) spatial resolution	CHIRPS Daily data University of California, Santa Barbara	Funk et al. (2015)
Tmax, Tmin, PDSI, WS, SM, ETA and ETR	1958-01-01T00:00:00Z–2021-12-01T00:00:00	0.05° grid: monthly and daily temporal resolution and a ∼4 km (1/24th degree) spatial resolution	TerraClimate University of California Merced	Abatzoglou et al. (2018); Wang-Erlandsson et al. (2016); Hegewisch and Abatzoglou (2020);
RH	1982-01-01T00:00:00Z–2022-12-01T00:00:00	0.1 × 0.1 degree (MERRA-2 and CHIRPS)	Famine Early Warning Systems Network (FEWS NET) Land Data Assimilation System; NASA GES DISC at NASA Goddard Space Flight Center	Funk et al. (2015) McNally et al. (2017)

Through this modern approach of remote sensing (RS), which plays an important role in data collection in many statistical areas (Yang et al. 2022), we were able to obtain the expected data. Also, GEE improves the accessibility and usability of the earth observation data and related tools by making them available to a wide range of researchers and the public (Sazib, Mladenova, and Bolten 2018). Its strength is based on the combination of petabyte-scale satellite imagery and geospatial data with worldwide analysis capabilities, providing easy global analysis and mapping, with easy access to data, and ‘unrestricted’ and stable processing power (Hu and Dong 2018; Chen et al. 2021). In addition to the above parameters, indices such as the Standardised Precipitation Index (SPI), Rainfall Anomaly Index (RAI), Standardised Anomaly Index (SAI), Ångström Indice (AI), Lowveld Fire Danger Index (FDI) have been applied to certain parameters to better characterise them. Statistical analysis was done using the R software.

2.3. Method of the link between fire and climatic parameter assessment

2.3.1. Data characteristics and assessment

• Precipitation, Standardised Precipitation Index (SPI) and Palmer Drought Severity Index (PDSI)

To emphasise the periods of water deficit in the study area, the dynamic of rainfall has been assessed from 1991 to 2021 using the Standardised Precipitation Index or Nicholson Index (SPI) (Table 1) and also PDSI has been assessed. The calculation of this index makes it possible to determine the degree of humidity or dryness of the environment (Bergaoui and Alouini 2001). The Standardised Precipitation Index (SPI) was developed by McKee, Doesken, and Kleist (1993). It is a statistical indicator used for the characterisation of local or regional droughts. Based on a long-term precipitation history (minimum 30 years), it quantifies the deviation of the precipitation of a period, deficit or surplus, from the historical average precipitation of the period. It can be normal, wet or dry. It also makes it possible to interpret the dynamics of the vegetation cover with the evolution of rainfall (McKee, Doesken, and Kleist 1993). Other tests (Pettitt and Mann-Kendall stationarity tests) were carried out to better assess trends. The SPI has been evaluated through the following equation: (1) $\mathrm{S}\mathrm{P}\mathrm{I}={\displaystyle \frac{Pi-P_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}}{\sigma }}$(1) SPI = Rainfall index for the year i; P_i= is the total rainfall for a year i; P_mean = annual average of rainfall observed over the whole series; σ = Standard deviation of annual rainfall observed for a given series.

The interpretation of the results of the SPI calculation was made according to the SPI classes and their degree of dryness or humidity (Table 2). Negative values of the SPI correspond to a dry year, whereas positive values indicate wet years.

• Palmer Drought Severity Index (PDSI)

Drought has different meanings to different people, depending on how a water deficiency affects them (Alley 1985). As a result, droughts have been classified into many different types. Meteorologic, agricultural, and hydrologic drought who is more commonly used. Meteorological drought is generally defined in terms of below-average precipitation over some time. Agricultural drought refers to a shortage of water in the root zone of crops such that the yield of plants is reduced considerably. Hydrologic drought is generally defined in terms of low levels of streamflow, reservoir storage, groundwater, or some combination (Alley 1985). The PDSI is a drought index that characterises the cumulative departure of local mean soil moisture conditions based on a simplified water balance calculation. Thus, it's categorised as a meteorological drought index, and it quantifies the water departure from the soil surface (Svoboda and Fuchs 2016). Although the PDSI has several limitations (Alley 1984; Karl and Knight 1985; Werick et al. 1994), it is commonly used throughout the world to quantify observed drought and drought projections (Ficklin et al. 2016). PDSI introduced by Palmer (1965) is calculated based on the parameters of the water balance of Thornthwaite and Mather (1955). The standardised measure of PDSI ranges from (Table 3) −4 (dry) to +4 (wet), with values below −3 representing severe to extreme drought (Palmer 1965). the descriptive and trend analysis were similarly assessed monthly over the period (1991–2021) through the time series data. The PDSI can be formulated as the following equation: (2) $X_{\left(i\right)}={\displaystyle \frac{z\left(i\right)}{\alpha }+\text{β}{X}_{\left(i-1\right)}}$(2) With, $X_{\left(i\right)}$ is the PDSI amount in ith month, $z\left(i\right)$ is the humidity anomaly index in ith month, Xð Þ i1 is the PDSI amount of the previous month, $\alpha$ and $\text{β}$ are the climatic coefficients of PDSI.

Table 2. Drought classification according to the SPI values (McKee, Doesken, and Kleist (1993))

Class of SPI	−1 < SPI < 0	1 < SPI < 2	SPI > 2	0 < SPI < 1	−2 < SPI < −1	SPI < −2
Degree of drought or wetness	Wetness moderate (WM)	Wetness high (WH)	Wetness extreme (HE)	Drought moderate (DM)	Drought high (DH)	Drought extreme (DE)

Table 3. PDSI categorisation of drought severity (Palmer 1965).

PDSI value	Drought category
4.00 or more	Extremely wet
3.00–3.99	Very wet
2.00–2.99	Moderately wet
1.00–1.99	Slightly wet
0.50–0.99	Incipient wet spell
0.49 to −0.49	Near normal
−0.50 to −0.99	Incipient dry spell
−1.00 to −1.99	Mild drought
−2.00 to −2.99	Moderate drought
−3.00 to −3.99	Severe drought
−4.00 or less	Extreme drought

2.3.2. Time series change detection and trend analysis

• Pettit test

The Mann–Whitney-Pettitt test is derived from the Mann–Whitney test, modified by Pettitt (Pettitt 1979). This test is non-parametric and is based on the fact that the absence of a break in a series means that the hypothesis is null (H_o) and the presence of a break constitutes the alternative hypothesis (H₁). It thus, makes it possible to identify a change in the behaviour of the data series by comparing two means. A break is defined as a change in the probability law of the random variables whose successive realisations define the time series studied (Fristch, Masson, and Marieu 1998). The implementation of the test assumes that for any time t between 1 and N, the time series (Xi), from i = 1 to t and from t + 1 to N belong to the same population. The test variable is the maximum absolute value of the variable Ut, N. In case the null hypothesis is rejected, an estimate of the break date is given by the time t defining the maximum in the absolute value of the variable Ut, N: (3) $U_{t,N}=\sum _{i=1}^t\sum _{i=t+1}^N{D}_{ij}$(3) with: (4) $D_{ij}=\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(x_i-x_j\right)\{\begin{array}{c}1if\left(x_i-x_j\right)>0\\ 0if\left(x_i-x_j\right)=0\\ -1if\left(x_i-x_j\right)<0\end{array}$(4)

We use the variable $K_N$ to test Ho such as $K_N$= $\underset{1}{max}\le t<T\left|U_{t,N}\right|$

In addition to the Pettitt test, which was applied to three of the parameters considered (PR, SM and RH), the Lee and Heghinian (Hubert and Carbonnel 1993) test was also applied because of its strength in analysing the change in a time series. The test is based on the assumption that the series is normally distributed. The tests are Bayesian procedures, which assume an evolution of the series as follows: (5) $x_i=\{\begin{array}{c}\mu +{ϵ}_i,i=1\dots ,\tau \\ \mu +\sigma +{ϵ}_i,i=\tau +1,\dots ..,N\end{array}$(5) With ${ϵ}_i$ are independent and normally distributed with a mean equal to zero and a variance equal to ${\sigma }^2.\tau$ is the position in time and is the scope of the possible change in the mean (Bougara et al. 2020; Hubert and Carbonnel 1993; Traore et al. 2014).

• Mann-Kendall test

The Mann–Kendall test statistic (S), is a nonparametric method for trend analysis that has been widely applied to regional climate studies (Mann 1945; Kendall 1975). The Mann-Kendall test was used to detect the existence of a single overall trend within a series. The Mann-Kendall test, which is based on Kendall's t-rank correlation statistic, is used to show the degree of significance of the trend and to determine the direction of the trend in the time series. it simply gives information on the course and a measure of the significance of observed trends of climatic parameters (Siraj et al. 2013). Thus, the trend analyses were performed for all the parameters using this test. The S-statistic is defined by: (6) $S=\sum _{j-1}^{n-1}\sum _{i+1}^n\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(x_i-x_j\right)$(6) where $x_i$ was the data value at time, $i$, $n$ is the length of the dataset, and $\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(x_i-x_j\right)$ is the sign function, which is computed as shown follows: (7) $\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(x_i-x_j\right)\{\begin{array}{c}1if\left(x_i-x_j\right)>0\\ 0if\left(x_i-x_j\right)=0\\ -1if\left(x_i-x_j\right)<0\end{array}$(7) for n > 10, the test statistic Z (Kendall tau) approximately follows a standard normal

distribution as in the following equation: (8) $Z=\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}\left(S\right)}}\mathrm{i}\mathrm{f}S>0}\\ 0\mathrm{i}\mathrm{f}S=0\\ {\displaystyle \frac{S+1}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}\left(S\right)}}\mathrm{i}\mathrm{f}S<0}\end{array}$(8) With $\mathrm{V}\mathrm{a}\mathrm{r}\left(S\right)$ is the variance of statistic S. Thus, a positive Z value indicates an increasing trend, whereas a negative value indicates that the trend is decreasing.

• Temperature and Wind speed

According to Hatfield and Prueger (2015), temperature is one of the main factors influencing the rate of plant development. The higher temperatures predicted by climate change and the risk of more extreme thermal events will have an impact on plant productivity (Hatfield and Prueger 2015). The latter dries them out and makes them fragile in the face of water stress. Thus, maximum, minimum and average temperatures were considered (monthly) in our assessment through a descriptive approach and the Mann-Kendall test as well. For Wind speed, the same approach was used. In the context of burnt area dynamics, the wind is a key parameter because, according to Guiguindibaye, Belem, and Boussim (2013), fire spread is closely linked to wind speed and vegetation height

• Relative humidity

Relative humidity (RH) is expressed in percentage (%) and is the ratio of the amount of water in the air (absolute humidity) to the maximum amount it can hold at a given temperature before condensing. It is influenced by temperature variation. In the present work, the data was computed through the GEE platform using the Clausius–Clapeyron approach due to the lack of relative humidity (Table 4) data for the whole period under consideration. Thus, the specific humidity was used for its calculation. It was then analysed through a descriptive approach (monthly analysis over the study period) and the Mann-Kendall test as well. According to Clapeyron’s (1834) and Clausius’s (1850) approach, Relative humidity is just $e/e_s$, the ratio of vapour pressure to saturation vapour pressure or $w/w_s$ the ratio of the mass mixing ratios of water vapour to actual and saturation values. Then, the specific humidity can be expressed as the mass mixing ratio of water vapour in the air, defined as follows: (9) $q\equiv {\displaystyle \frac{m_v}{m_v+m_d}={\displaystyle \frac{w}{w+1}\approx w}}$(9) Relative humidity can be expressed as the ratio of water vapour mixing ratio to saturation water vapour mixing ratio, $w/ws$, with: (10) $w_s\equiv {\displaystyle \frac{m_{vs}}{m_d}={\displaystyle \frac{e_s{R}_d}{R_v(p-e_s}\approx 0.622{\displaystyle \frac{e_s}{p}}}}$(10)

Thus, from Clausius–Clapeyron: (11) $e_s\left(T\right)=e_{s0}\exp \left[\left({\displaystyle \frac{L_v\left(T\right)}{R_v}}\right)\left({\displaystyle \frac{1}{T_0}-{\displaystyle \frac{1}{T}}}\right)\right]\approx 611\exp \left({\displaystyle \frac{17.67(T-T_0}{T-29.65}}\right)$(11) Knowing $w$ and $w_s$ Relative Humidity (RH) is: (12) $\mathrm{R}\mathrm{H}=100{\displaystyle \frac{w}{w_s}\approx 0.263pq\left[\exp \left({\displaystyle \frac{17.67(T-T_0}{T-29.65}}\right)\right]}$(12)

• Evapotranspiration (ET)

Evapotranspiration is a reference value (Table 4). It is the quantity of water evaporated by the growing plant cover, mowed regularly with a supply of water and nutritional elements to cover the plant's needs (Legras 2014). The ET therefore only depends on climatic conditions such as solar radiation, wind speed, atmospheric humidity etc., as all other factors are fixed (type of plant cover, phenological stage, water supply). In this study, Actual (ETA) and reference (ETR) evapotranspiration have been used. They were retrieved from the TerraClimate platform, using a one-dimensional soil water balance model (Finkelstein et al. 2022) and ASCE (American Society of Civil Engineers) Penman-Montieth. Actual Evapotranspiration (ETA) represents the maximum water demand for a reference grass crop calculated according to the Penman-Monteith method (Penman 1948) whereas reference evapotranspiration (ETR) percentiles is similar to the Evaporation Demand Drought Index (EDDI), (Hegewisch and Abatzoglou 2020). Thus, Daily grass-reference ET (ETR) was computed using the standardised ASCE form of the Penman-Monteith equation (ASCE-EWRI 2005). The data obtained allowed us to evaluate and analyse the evolution of the two types of evapotranspiration (ETA, ETR). The Penman-Monteith reference evapotranspiration equation with fixed stomatal resistance values for the area under the grass is as follows: (13) $\mathrm{E}\mathrm{T}\mathrm{R}={\displaystyle \frac{0.408\mathrm{\Delta }\left(\mathrm{R}\mathrm{n}-G\right)+{\displaystyle \frac{C_n}{T+273}{u}_2\left(e^o-e\right)}}{\mathrm{\Delta }+\gamma \left(1+C_d{u}_2\right)}}$(13)

• Soil Moisture

Soil Moisture (SM) is expressed as a percentage (%) and is the ratio of the amount of water in the air (absolute humidity) to the maximum amount it can hold at a given temperature before condensing. It is influenced by temperature variation. The soil moisture data derived is from a one-dimensional soil water balance model (Finkelstein et al. 2022). The Soil-Water-Balance (SWB, see Table 5) model has been developed to allow estimates of potential recharge to be made quickly and easily (Finkelstein et al. 2022). The code calculates components of the water balance at a daily timestep through a modified version (Table 6) of the Thornthwaite-Mather soil-moisture-balance approach (Westenbroek et al. 2010). Thus, the data obtained allowed us to characterise the study areas in terms of soil water availability.

2.3.3. Anomalies assessment in the time series data (RAI and SAI)

In addition to SPI and PDSI, the anomalies were calculated in the rainfall time series to better identify the periods of pejoration and majoration. For this purpose, the Rooy (Rooy 1965) Rainfall Anomaly Index (RAI) was calculated monthly and classified to analyse the frequency and intensity of the rainfall (Table 7). RAI, developed and first used by Rooy (1965) and adapted by Freitas (2005), constitutes the following equations: (14) $\mathrm{R}\mathrm{A}\mathrm{I}=3\left[{\displaystyle \frac{N-\tilde{N}}{\acute{M}-\tilde{N}}}\right],\mathrm{F}\mathrm{o}\mathrm{r}\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{e}\mathrm{s}$(14) (15) $\mathrm{R}\mathrm{A}\mathrm{I}=-3\left[{\displaystyle \frac{N-\tilde{N}}{\overline{U}-\tilde{N}}}\right],\mathrm{F}\mathrm{o}\mathrm{r}\mathrm{n}\mathrm{e}\mathrm{g}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{e}\mathrm{s}$(15)

Table 4. Designation of variables used in for computing of RH.

Variables used	Designations
$q$	Specific humidity or the mass mixing ratio of water vapour to total air (dimensionless)
$m_v$	Specific mass of water vapour (kg)
$m_{vs}$	Specific mass of water vapour at equilibrium (kg)
$m_d$	Specific mass of dry air (kg)
$w$	Mass mixing ratio of water vapour to dry air (dimensionless)
$w_s$	Mass mixing ratio of water vapour to dry air at equilibrium (dimensionless)
$e_s\left(T\right)$	Saturation vapour pressure (Pa)
$e_{s0}$	Saturation vapour pressure at T0 (Pa)
$R_d$	Specific gas constant for dry air (J kg^−1 K^−1)
$R_v$	Specific gas constant for water vapour (J kg^−1 K^−1)
$p$	Pressure (Pa)
$L_v\left(T\right)$	Specific enthalpy of vaporisation (J kg^−1)
$T$	Temperature (K)
$T_0$	Reference temperature (typically 273.16 K) (K)

Table 5. Data requirements for application of the SWB model (Finkelstein et al. 2022).

Data type

Gridded (ARC ASCII or Surfer grid)

Land use/land cover

Surface-water flow direction (D8)

Hydrologic soil group

Available water capacity

Tabular

Soil and land use properties lookup table

The climate at a single station

Matrix of soil-water retention for given accumulated potential

water loss

Table 6. Designation of variables used in for computing of ETR (ASCE-EWRI 2005).

Variables used	Designations
$\mathrm{E}\mathrm{T}\mathrm{R}$	Reference evapotranspiration (mm.d^−1 for daily time steps or mm h^-1 for hourly time steps),
$\mathrm{R}\mathrm{n}$	Calculated net radiation at the crop surface (MJ.m^−2.d^−1 for daily time steps or MJ.m^−2.h^−1 for hourly time steps),
$G$	Soil heat flux density at the soil surface (MJ.m^−2.d^−1 for daily time steps or MJ.m^−2.h^−1 for hourly time steps),
$T$	Mean daily or hourly air temperature at 1.5–2.5-m height (°C),
$u_2$	Mean daily or hourly wind speed at 2-m height (m s^−1),
$e^o$	Saturation vapour pressure at 1.5–2.5-m height (kPa), calculated for daily time steps as the average saturation vapour pressure at maximum and minimum air temperature,
$e$	Mean actual vapour pressure at 1.5–2.5-m height (kPa)
$\mathrm{\Delta }$	Slope of the saturation vapour pressure-temperature curve (kPa °C^−1),
$\gamma$	Psychrometric constant (kPa °C^−1),
$C_n$	Numerator constant that changes with reference type and calculation time step (K mm.s^3 Mg^−1.d^−1 or K mm.s^3 Mg^−1 h^−1)
$C_d$	Denominator constant that changes with reference type and calculation time step (s.m^−1).
Units for the 0.408 coefficient are m^2 mm MJ^−1.

Table 7. Rainfall Anomalies Index classification (Freitas 2005).

	RAI range	Classification
Rainfall Anomaly Index (RAI)	Above 4	Extremely humid
	2–4	Very humid
	0–2	Humid
	−2 to 0	Dry
	−4 to −2	Very dry
	Below −4	Extremely dry

With: $N$ = current monthly/yearly rainfall (mm); $\tilde{N}$ =  average monthly/yearly rainfall (mm) of the historical series (mm); $\acute{M}$ = average of the 10 highest monthly/yearly precipitations of the historical series (mm);  =  average of the 10 lowest monthly/yearly precipitations of the historical series (mm); and positive anomalies have their values above average and negative anomalies have their values below average.

Therefore, for mean temperature, relative humidity and soil moisture, the Standardised Anomaly Index (SAI) (Koudahe et al. 2017) was performed to evaluate the changes and fluctuations that occurred and monthly calculated.

SAI is expressed as follows: (16) $\mathrm{S}\mathrm{A}\mathrm{I}={\displaystyle \frac{X_t-X_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}}{\delta }}$(16) With: $X_t$: annual mean observation values during time t; $X_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}$: long-term mean of the series considered throughout the observation $\delta$: standard deviation of the time series considered throughout the observation.

2.3.4. Estimating of fire risk indices

To evaluate the risk of fire as a function of specific climatic parameters over the study period (2001–2021), the following will be used: Ångström and Lowveld Indices.

• Ångström Indice (AI)

The Angström index (AI) is a very simple fire index developed in Sweden and designed to be calculated mentally. It only requires relative humidity and temperature as input variables (Chandler et al. 1983). It is frequently used in Sweden and South Africa (Willis et al. 2001; Ångström 1949).

It has been used in some parts of Scandinavia for indicating expected fire starts on a given day (Chandler et al. 1983). The Angström index is estimated by the equation: (17) $\mathrm{A}\mathrm{I}=\left(\mathrm{R}\mathrm{H}/20\right)+(27-T_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}})/10$(17) with T_mean (°C), the average temperature and RH (%) the relative humidity. If:

• AI > 4, fire occurrence is unlikely

• 2.5 < AI < 4.0, fire conditions are unfavourable

• 2.0 < AI < 2.5, fire conditions are favourable

• AI < 2, fire occurrence very likely

• • Lowveld Fire Danger Index (FDI)

In this study, the Lowveld Fire Danger Index (FDI) was utilised, which is commonly used in South Africa and gives an acceptable assessment of short-term fire risk. The Lowveld model has been certified as South Africa's official wildfire danger rating system (Lall and Mathibela 2016). This is a modified version of the Zimbabwe fire hazard index (Meikle and Heine 1987). The model employs the same inputs as the McArthur models, which have been scaled to generate a simple model that can readily calculate index values without the need for sophisticated calculations. Then, as indicated in the following equation: (18) $\mathrm{F}\mathrm{D}\mathrm{I}=(T_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}-35)-\left(\right(35-T_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}\left)/30\right)+\left(\right(100-\mathrm{R}\mathrm{H})\times 0.37)+30$(18) With T_mean (°C): the average temperature and RH (%): the relative humidity.

The FDI categories are as follows:

• 0 < FDI < 20, very low risk;

• 21 < FDI < 45, moderate risk;

• 46 < FDI < 60, high risk;

• 61 < FDI < 75, very high risk;

• 76 < FDI < 100, extremely high risk.

2.3.5. Statistics analysis on the assessment of the link between fires, climatic and environmental parameters

2.3.5.1. Assessment of the link between fires, climatic and environmental parameters

Correlation analysis between variables was done to assess the degree of association between burnt areas, active fires and the considered climatic and environmental parameters. Thus, non-parametric Spearman correlation (Ramsey 1989) and cross-correlation analysis (equation) were carried out in R software respectively to evaluate the correlation between the variables and those who influence the fires (burnt areas and the active fires) over the period studied (2001–2021) in the studies zones. The cross-correlation function (CCF) allows the statistical analysis of the relationship between two variables in a system, provided that both variables are time series data and are assumed to be stationary in terms of mean and variance (Shumway and Stoffer 2011). It is expressed as follows:

CCF between $X_i$is called the $Y_{i+k}$order cross-correlation of X and Y. The sample estimate of this cross-correlation, called $r_k$ is calculated using the formula: (19) $r_k={\displaystyle \frac{{\sum }_{i=1}^{n-k}\left(X_i-\overline{X}\right)\left(Y_{i+k}-\overline{Y}\right)}{\sqrt{{\sum }_{i=1}^n{\left(X_i-\overline{X}\right)}^2{\sum }_{i-1}^n{\left(Y_i-\overline{Y}\right)}^2}}\mathrm{w}\mathrm{i}\mathrm{t}\mathrm{h}\overline{X}={\displaystyle \frac{1}{n}\sum _{i=1}^n{X}_i\mathrm{a}\mathrm{n}\mathrm{d}\overline{Y}={\displaystyle \frac{1}{n}\sum _{i=1}^n{Y}_i}}}$(19) $k$ called the time index is allowed to be either positive or negative. The large sample standard error of the sample cross-correlation is $1/\sqrt{n}$ so that large sample confidence limits are $\pm 1/\sqrt{n}$.

Concerning Spearman correlation, it is expressed by the following equation: (20) $\rho =1-{\displaystyle \frac{6\sum d_i^2}{n\left(n^2-1\right)}}$(20) $\rho$ =  Spearman’s rank correlation coefficient; $d_i$ = difference between the two ranks of each observation; $n$ = number of observations.

3. Results

3.1. Variability of parameters and trends

This section characterises the climate and environmental parameters from 1991 to 2021. This characterisation has been done through the analysis of their variability as well as the evaluation of trends in certain parameters, without forgetting the probable breaks and their spatial presentation (Figure 6).

3.2. Climate and environmental variability

3.2.1. Rainfall variability and evapotranspiration and soil moisture

The variation of precipitation, evapotranspiration (ETA, ETR) and soil moisture data show an oscillating pattern (Figure 2) over the study period considered (1991–2021).

Figure 2. Monthly rainfall variation in Guinea-savannah (a) and Forest-savannah mosaic (b) zones.

Monthly rainfall amounts range from 0 mm to 332 mm per month in the Guinea savannah zone, and 0 mm to 377 mm per month in the forest-savannah mosaic zone. The ATE, REE and SM vary respectively from 4 mm to 150 mm, 91–204, and 20 mm to 223 mm per month in the Guinea savannah zone while in the forest-savannah mosaic zone, they are respectively, 6–148 mm/month (ATE), 85–165 mm/month (REE) and 15–149 (SM) mm/month. The annual average is 1257.8 mm/year in the forest-savannah mosaic zone, with June, July and September being the wettest months, while in the Guinean savannah zone, it is 1017.9 mm/year, with June, August and September being the wettest months.

3.2.2. Variability of climatic and hydrological drought periods

The analysis of drought periods was assessed by considering climatic drought (SPI) and hydrological drought (PSDI). This assessment allowed us to highlight wet and dry periods in general.

An examination of the climatic drought periods (DPSs) over the period 1991–2021 in the Guinea Savannah area generally shows two main periods: the dry period and the wet period. However, the period is characterised by a situation dominated by moderate humidity followed by a moderate drought after which comes the severe drought period (Table 8 and Figure 3). Thus, over the whole study period, the years of severe drought and moderate humidity have the same proportions 19.4%, i.e. 6 (six) years in the series studied. Concerning drought periods, they are dominated by moderate and high drought periods with respectively 25.8% and 19.4% of the time series. There are also periods of extreme drought in recent years such as in 2018 and 2019. The hydrological drought (Table 8 and Figure 3) is more pronounced in this ecological zone. It is marked by periods of extreme drought (3%) without a period of extreme moisture (0). The period is however dominated by periods of Moderate Drought (16.1%); Mild Drought (29%); Incipient Dry Spell (19.4%).

Figure 3. Annual variation of SPI and PDSI in Guinea-Savannah (a) and Forest-savannah mosaic (b) zones.

Table 8. Variation in climatic (SPI) (a) and hydrological (PDSI) (b) drought in Guinea-Savannah and Forest-savannah mosaic zones.

According to SPI values		EW	HD	HW	ED	MD	MW
Guinea-savannah zone	Years	1991, 1999, 2018, 2019	1992, 1998, 2002, 2005, 2014, 2015	2003, 2009, 2010, 2021	2001, 2013, 2017	1993, 1995, 1996, 2006, 2007, 2008, 2011, 2012	1994, 1997, 2000, 2004, 2016, 2020
	Number of years	4	6	4	3	8	6
	Percentage (%)	12.9	19.4	12.9	9.7	25.8	19.4
Forest-savannah mosaic Zone	Years	1991, 1995, 1999, 2007, 2010, 2011, 2019, 2021	1994, 1998, 2004, 2005, 2014	2008	1992, 2001, 2013, 2015, 2020	1996, 2002, 2006, 2009	1993, 1997, 2000, 2003, 2012, 2016, 2017, 2018
	Number of years	8	5	1	5	4	8
	Percentage (%)	25.8	16.1	3.2	16.1	12.9	25.8
Extreme Wetness (EW); High Drought (HD); High Wet (HW); Extreme Drought (ED); Moderate Drought (MD); moderate Wetness (MW)
According to PDSI values		ED	SD	MD	MiD	IDS	NN	IWS	SW	VW
Guinea-savannah zone	Year	2017, 2016, 2002	2020, 2021	1993, 2019, 2015, 2001, 2006	2018, 2007, 2005, 2003, 2013, 1998, 2014, 1994, 2012	2008, 1999, 1995, 2011, 1992, 2010	2004	2000, 2009	1996, 1997, 1991	0
	Number of years	3	2	5	9	6	1	2	3	0
	Percentage (%)	9.7	6.5	16.1	29.0	19.4	3.2	6.5	9.7	0.0
Forest-savannah mosaic zone	Year	2017, 2016, 2021, 2001, 1993	2002, 2018, 1994	1992, 2015, 2006, 1998	2000, 2020, 1997, 2012, 2003, 2010, 2007, 2013	2004, 2011, 1995, 2005, 2019, 2014, 1999	2009, 1996	0	2008	1991
	Number of years	5	3	4	8	7	2	0	1	1
	Percentage (%)	16.1	9.7	12.9	25.8	22.6	6.5	0.0	3.2	3.2
Extreme Drought (ED); Severe Drought (SD); Moderate Drought (MD); Mild Drought (MiD); Incipient Dry Spell (IDS); Near Normal (NN); Incipient Wet Spell (IWS); Slightly Wet (SW); Very Wet (VW)

As for the Forest-savannah mosaic zone, over the entire study period, extreme humidity and moderate humidity years are dominant (25.8%, i.e. 8 (eight) years each in the series studied. Regarding drought periods, the area also experienced periods of extreme climatic drought (16.1%) and moderate drought (16.1%). There was also a period of high humidity during the 200. The hydrological drought is no less in this ecological zone. It is marked by periods of extreme drought (16.1% or 5 years) with one Very Wet year (3.2%) (1991). The period is however dominated by periods of Mild Drought (25.8%); Incipient Dry Spell (22.6%).

To this end, on a temporal scale, the SPI and PDSI (Figure 8), indicate the irregular character of rainfall and soil moisture, and thus highlight climatic and hydrological (dry and wet) episodes.

3.2.3. Thermometric variability

The monthly temperature trend according to the ecoregion (Figure 4) shows that average temperatures (T_mean) vary from 25°C to 31.9°C in the Guinea-savannah zone. These temperatures vary from 24.8°C to 30.9°C in the forest-savanna mosaic zone. Observations of thermal extremes show that:

• The maxima (T_max) varied from 28.5°C to 37.7°C in the two zones (Guinea-savannah and forest-savannah mosaic).

• Minima (T_min) increased from 18°C to 24.7°C in the Guinea Savannah zone, and from 19.2°C to 25°C in the forest-savannah mosaic zone. The year 2019 was the hottest in both ecological zones with average temperatures above 30°C during the dry season with March and April as the hottest months.

Figure 4. Monthly temperature variation in Guinea-Savannah (a) and Forest-savannah mosaic (b) zones.

3.2.4. Relative humidity variability and wind speed

The variation of relative humidity (RH) and wind speed (WS) show an oscillating pattern (Figure 5) over the whole study period (1991–2021).

Monthly relative humidity varies between 12.6% and 82.6% per month in the Guinea-savannah zone, 39.1% to 84.4% per month in the forest-savannah mosaic zone, while wind speed varies between 0.8–2.6 m/s and 0.7–1.8 m/s respectively. The annual average is 56% in the Guinean savannah zone for RH with August and September being the wettest months, 2018 being the wettest year and 2001 the least wet. In the forest-savanna mosaic zone, it is 70.1% (annual average) with July and August as the wettest months, 1995 as the wettest year and 1993 as the least wet. The velocity peaks at 2.6 m/s in the Guinean savannah and 1.8 m/s in the forest-savannah mosaic zone (Figures 5 and 6).

Figure 5. Monthly variation of relative humidity and wind speed in Guinea-Savannah (a) and Forest-savannah mosaic (b) zones.

Figure 6. Monthly spatial presentation of climatic and environmental parameters in Guinea-Savannah (a) and Forest-Savannah mosaic (b) zones.

3.3. Trends in climate and environmental time series data

3.3.1. Anomalies and fire risk indicators

3.3.1.1. Rainfall anomaly

The monthly rainfall anomalies oscillate on either side of the normal (1991–2021) with increasing and decreasing trends. They ranged from −6.2 to +4.1 mm in the Guinean savannah zone, while the forest-savannah mosaic zone recorded anomalies ranging from −3.04 to +2.06 mm. In the forest-savanna mosaic zone, a significant decrease in annual rainfall was identified during the period 2012–2020, preceded by a slight improvement in rainfall from 2007 to 2011. From 2013 to 2017, a significant deterioration in rainfall conditions was recorded, followed by a significant recovery in rainfall from 2018 in the Guinean savannah (Figure 7).

Figure 7. Annual rainfall anomaly in Guinea-Savannah (a) and Forest-savannah Mosaic (b) zones.

3.3.1.2. Thermometric anomaly

The graphs (Figure 8) show the magnitude of the thermometric anomalies from 1991 to 2021.

Figure 8. Annual thermometric anomaly in Guinea-Savannah (a) and Forest-Savannah mosaic (b) zones.

In general, the average monthly temperature is higher than normal (1991–2021) showing an upward trend. The magnitude of monthly temperatures varies from −1.8 to +2.6°C in the Guinean savannah area, while the forest-savannah mosaic area recorded anomalies varying from −2.2 to +2.4°C. These values show a slight increase in temperature in these areas (positive anomaly). This increase is most noticeable during the period 2014–2021 in the Guinean savannah and from 2016 to 2021 in the forest-savannah mosaic zone.

3.3.1.3. Soil moisture anomaly

The graphs (Figure 9) show the anomalies within the soil moisture data series from 1991 to 2021. In general, soil moisture is lower than normal (1991–2021) showing a downward trend. The average monthly moisture varies from −0.96 to +2.3 mm in the Guinean savannah zone, while the forest-savannah mosaic zone recorded anomalies varying from −0.92 to +1.2 mm. These values show an overall decrease in humidity at the level. This is reflected in the mean annual variation with deficits (2013–2017) in both ecological zones with a slight increase in the forest-savanna mosaic zone from 2019 onwards. These phases were preceded by non-uniform periods where oscillations were noticeable in both zones with periods of decline (1992, 1998, 2001, 2002, 2005 and 2006) in the Guinean savannah zone while in the forest-savannah mosaic zone, few deficits were observed (1998, 2000, 2001).

Figure 9. Annual soil moisture anomaly in Guinea-Savannah (a) and Forest-Savannah mosaic (b) zones.

3.3.1.4. Humidity anomaly

Relative humidity, expressed as a percentage, is the ratio of the amount of water in the air to the absorption capacity at a given temperature (Figure 10).

Figure 10. Annual moisture anomaly in Guinea-Savannah (a) and Forest-Savannah mosaic (b) areas.

The hygrometric anomalies for this parameter varied from −2.2% to +1.3% in the Guinea-Savannah zone, and from −2.8% to +1.3% in the forest-savannah mosaic zone. The average monthly anomalies are +0.337% and +0.97% in the Guinea-Savannah and Forest-Savannah mosaic zones respectively. From 1991 to 2021, the monthly mean relative humidity varied between 51.2% (January) and 80.95% (July) in the forest-savannah mosaic zone while in the Guinea savannah, it varied from 25.6% (January) to 77.8% (August). Seasonal variations in hygrometry are similar to those of rainfall with maximum rainfall observed in the same months. These hygrometric variations over time, in duration and length, are an indicator of climate change.

3.3.1.5. Fire risk indices (AI and FDI)

The calculated indices were used to assess the fire risk in the two ecological zones considered. Thus different approaches taking into account the same components (relative humidity and precipitation) were evaluated. Thus, in both zones, the indices are opposite in terms of risk. The FDI in both zones shows a higher fire risk in the last 8 years (2013–2021) as well as from 1998 to 2004 interspersed with periods of low fire risk (Figure 11)

Figure 11. Annual variation of fire risk indices in Guinea-Savannah (a) and Forest-Savannah mosaic (b) areas.

These indices are then correlated with the parameters used as well as the fire parameters (burnt area, active fires) to highlight their capacity to warn about fire danger.

3.3.2. Trends in variables and breaks in rainfall and relative humidity time series data

3.3.2.1. Break in rainfall and humidity series

The detection of breaks in stationarity of rainfall and hygrometry by the Pettitt and Lee-Heighinian tests made it possible to observe a change in the evolution of relative humidity and rainfall from 1982 and 1998 (Table 9). The application of the Pettitt test (Table 9) on the rainfall and relative humidity series did not, however, detect any break in the two ecological zones (statically insignificant test at 95%). On the other hand, the Lee-Heighinian test identified a significant break in 2017 and 2003 respectively for rainfall (PR) and relative humidity in the Guinean savannah zone. In the forest-savanna mosaic zone, 1991 (RP) and 2002 (RH) were detected as break years.

Table 9. Period of HR and SM breakdown in Guinea-Savannah (a) and Forest-Savannah mosaic (b) areas.

		Pettitt test			Lee-Heghinian test
		Kt (Break year)	Test U	Significant level	Break year	Break Probability	Significant level
Guinea-savannah zone	PR	No break	–	–	2017	0.1484	**
	RH	No break	–	–	2003	0.1523	**
Forest-savannah mosaic zone	PR	No break	–	–	1991	0.387	**
	RH	No break	–	–	2002	0.1227	**

** p < 0,01; Precipitation (PR); Relative Humidity (RH); Soil Moisture (SM).

3.3.3. Trends of the variables

3.3.3.1. Mann kendall test

The table below shows the results of the Mann-Kendall test applied to the data set.

The Mann-Kendall test (Table 10) shows a significant increasing trend at the 100% threshold (Z-test > 0) within the minimum (T_max), maximum (T_mean) and mean (T_min) temperature data in both ecoregions, except for ETA, PR, RH, WS (Guinea-Savannah zone) and ETA, PR, PDSI, RH, SM (Forest-Savannah mosaic zone), where both non-significant decreasing and increasing trends were identified. The WS shows a significant downward trend in both zones but with a non-significant downward trend detected in the Guinean savannah zone. Across the two ecoregions, ETR showed a significant upward trend, while PDSI showed a significant downward trend in the Guinean zone but not in the Savannah Forest Mosaic zone. ETA, however, shows a significant downward trend (Guinea-Savannah zone) and an upward trend (Forest-Savannah zone). Only SM in the Guinea-Savannah zone and WS in the Forest-Savannah zone are significant at the 10% and 5% levels respectively.

Table 10. Trend of different variables in Guinea-Savannah (a) and Forest-Savannah mosaic (b) areas.

	Variable	Tau	S	Z	p-value	Sen's lope S
Guinea Savannah zone	ETA	−0.166	−77	−1.292	0.1965^ns	−0.255
	ETR	0.389	181	3.059	2.218 × 10^−3**	0.267
	PR	−0.019	−9	−0.136	0.8918^ns	−0.024
	PDSI	−0.402	−187	−3.161	1.57 × 10^−3***	−0.103
	RH	0.148	69	1.156	0.2478^ns	0.068
	SM	−0.209	−97	−1.632	0.1028**	−0.496
	Tmax	0.557	259	4.385	1.159 × 10^−5***	0.042
	Tmean	0.665	309	5.235	1.651 × 10^−7***	0.036
	Tmin	0.570	265	4.487	7.221 × 10^−6***	0.035
	WS	−0.196	−91	−1.530	0.1261^ns	−0.003
Forest-Savannah mosaic zone	ETA	0.058	27	0.442	0.6586^ns	0.077
	ETR	0.557	259	4.385	1.159 × 10^−5***	0.352
	PR	−0.041	−19	−0.306	0.7597^ns	−0.076
	PDSI	−0.101	−47	−0.782	0.4343^ns	−0.040
	RH	0.075	35	0.578	0.5633^ns	0.025
	SM	−0.196	−91	−1.530	0.1261^ns	−0.389
	Tmax	0.583	271	4.589	4.453 × 10^−6***	0.045
	Tmean	0.695	323	5.473	4.429 × 10^−8***	0.043
	Tmin	0.665	309	5.235	1.651 × 10^−7***	0.041
	WS	−0.286	−133	−2.244	0.02486*	−0.003

ns = non significative; * p < 0,05; ** p < 0,01; *** p < 0,001; Trend (At 95% level); Mann Kendall Statistic (S); Normalised Test Statistic (Z); Sen's slope (S) At 95% level.

3.3.4. Correlation between fire variables, climatic and environmental variables

3.3.4.1. Spearman correlation between variables, climatic and environmental variables

The correlation matrix (Figure 12) gives a first idea of the associations between variables and climatic variables according to the phytogeographical zone. The correlation between climatic indices, hazard indices and variables was calculated using monthly data. Both negative and positive correlations were significant at p < 0.0.5, p < 0.01 and p < 0.001.

Figure 12. Spearman's correlation matrix in Guinea-Savannah (a) and Forest-savannah mosaic (b) areas (Significant at *p < 0.05; ** p < 0.01; and *** p < 0.001).

The correlation matrix gives a first idea of the existing relationship between the different climatic, environmental and variables according to the ecoregions. These parameters are relatively well correlated with each other (Figure 12) according to the matrix. In the Guinea-Savannah zone, the statistically insignificant variables (p > 0.05) are WS, T_mean, PDSI, and ETR for the burnt area (BA) and WS, SM, PDSI, ETR for active fires (HS), while for the forest-savannah zone, we have WS, T_mean, PDSI for the two variables (BA and HS). The other variables are highly significant (p < 0.001) except for ETR (p < 0.01) in the forest-savannah zone.

The most correlated variables for BA and HS are T_min (negative correlation; rho = −0.44 and −0.62), and the fire risk index FDI (positive correlation; rho = 0.34 and 0.41), AI (negative correlation; rho = −0.44 and −0.56) for the Guinean savannah while in the forest-savannah zone, we have: T_min (negative correlation; rho = −0.63 and −0.61), and the fire risk indices FDI (positive correlation; rho = 0.58 and 0.62), AI (negative correlation; rho = −0.77 and −0.80), RH (negative correlation; rho = −0.74 and −0.74).

In addition, all ecoregions show a proportional negative correlation with precipitation (PR) and a positive correlation with maximum temperature (T_max).

3.3.4.2. Cross-correlation of fire variables, climatic and environmental variables

Exploratory analyses revealed a strong correlation between certain climatic variables and fire variables. However, correlation does not necessarily lead to causality. To investigate which factors, impact fire dynamics, an intercorrelation was used for the time series. Therefore, cross-correlation functions (CCFs) were calculated between the data series with different lags (−21 to +21 months).

3.3.4.3. CCF between the variables and the burned area

In the Guniea-Savannah area, it is noted that the CCF (Table 11) reached its maximum at lag 12, −12, and 13 for the variables RH, AI and FDI. The results show that the variables decrease and increase simultaneously, while ETA, ETR, PDSI, SM, and WS evolve oppositely to the burnt areas. CCF peaked at time h = 10 for T_max, T_mean and T_min while PR peaked at time h = 0, which shows that temperature and rainfall recorded respectively at month t + 10 and t + 0 are associated with area burnt at month t + 10 and t, thus showing that temperature takes effect from month 10 while the impact of rainfall is noticeable at time t or it is in deficit. On the other hand, the variables SM, ETR, ETA and PDSI reached a peak at lag −3, −1 and −7. These results indicate that these factors influence the area burned about 3 (SM, ETR), 1 (ETA) and 7 (PDSI) months before their occurrence.

Table 11. Cross-correlation function between climatic, environmental and burned area variables in Guinea-Savannah (a) and Forest-Savannah mosaic (b) zones.

	Cross-correlation function (Burned area and parameters)
	Guinea-savannah zone		Forest-savannah mosaic zone
	CCFmax*	Lag (month)	CCFmax*	Lag (month)
ETA	−0.59	−1	0.37	19
ETR	0.514	−3	0.514	−3
PR	−0.523	0	−0.523	0
PDSI	−0.094	−7	−0.099	−19
RH	−0.644	12	0.394	6
SM	−0.428	−3	−0.428	−3
Tmax	0.511	10	0.539	−2
Tmean	0.58	10	0.601	−2
Tmin	0.459	9	0.465	−15
WS	−0.452	−9	−0.452	−9
AI	−0.684	−12	−0.702	12
FDI	0.586	−13	0.595	11

Actual Evapotranspiration (ETA); Reference Evapotranspiration (ETR); Precipitation (PR); Palmer Drought Severity Index (PDSI); Relative Humidity (RH); Soil Moisture (SM); Maximum Temperature (T_max); Average Temperature (T_mean); Minimum Temperature (T_min); Wind Speed (WS); Ångström Indice (AI); Lowveld Fire Danger Index (FDI); CCFmax*: Maximum significant value (p < 0.05) of the cross-correlation function.

For the Forest-Savannah mosaic area, the CCF peaked at lag 12, −2, and 11 for the variables AI, T_mean and FDI. The same variables decrease and increase simultaneously, while PR, PDSI, SM, WS, and AI evolve oppositely to the burnt areas. The CCF in this area peaked at time h = −2 for T_max and PR peaked at time h = 0, which shows that temperature and rainfall recorded respectively in month t – 2 and t + 0 are associated with the number of areas burnt in month t – 2 and t, thus showing that the impact of temperature takes effect from the second month (vegetation drying) while the impact of rainfall is perceptible at time t where it is deficient. In contrast, the variables SM, ETR, ETA and PDSI peaked at lag −3, −19. These results indicate that these factors influence the area burned about 3 (SM, ETR), and 19 (ETA, PDSI) months before their occurrence (favourable condition).

In general, PDSI does not significantly impact the area burned in the two ecological zones. The rest of the variables contribute to the spread of wildfires.

However, ETR, T_max, T_mean, T_min and FDI are the positively correlated variables that influence the area burnt in the Guinea Savannah zone. The other negatively correlated variables show an opposite relationship, while in the Forest-Savannah mosaic zone, ETA, ETR, RH, T_max, T_mean, T_min and FDI are the variables that are positively correlated with fire damage.

3.3.4.4. CCF between the variables and the active fires

Concerning the contribution of the variables on the persistence of active fires, it appears that in the Guniea-Savannah area, the CCF (Table 12) reached its maximum at lag 11, 10 and −1 for the variables ETA, FDI and T_mean. ETA, PR, and WS evolve oppositely to the area burnt. The CCF reached a peak at lag for T_mean (h = 10) and PR (h = −12), which shows that temperature and rainfall recorded respectively at month t + 10 and t-12 are associated with the number of active fires detected at month t + 10, and t-12, thus showing that the impact of temperature takes effect from the tenth month onwards while the impact of rainfall is perceptible at time t – 12 (the twelfth month before the possible passage of fire action). In contrast, the variables SM, ETR, ETA and PDSI peaked at lag –21, −3, 11 and 17. These results indicate that these factors influence active fires about −3 (ETR), −21 (SM), and 11 (ETA) months before and 17 (PDSI) months after their occurrence.

Table 12. Cross-correlation function between climatic, environmental and active fires variables in Guinea-Savannah (a) and Forest-Savannah mosaic (b) zone.

	Cross-correlation function (Active fires and parameters)
	Guinea-savannah zone		Forest-savannah mosaic zone
	CCFmax*	Lag (month)	CCFmax*	Lag (month)
ETA	−0.602	11	0.408	8
ETR	0.548	−3	0.548	−3
PR	−0.538	−12	−0.553	0
PDSI	−0.122	17	−0.122	17
RH	0.424	18	0.427	6
SM	0.496	−21	−0.431	9
Tmax	0.565	−2	0.565	−2
Tmean	0.63	10	0.649	−2
Tmin	0.519	9	0.519	9
WS	−0.48	15	−0.496	3
AI	0.356	18	0.36	7
FDI	0.639	−1	0.639	−1

Actual Evapotranspiration (ETA); Reference Evapotranspiration (ETR); Precipitation (PR); Palmer Drought Severty Index (PDSI); Ralative Humidity (RH); Soil Moisture (SM); Maximum Temperature (T_max);Average Temperature (T_mean); Minimum Temperature (T_min); Wind Speed (WS); Ångström Indice (AI); Lowveld Fire Danger Index (FDI); CCFmax*: Maximum significant value (p < 0.05) of the cross-correlation function.

For the Forest-savannah mosaic zone, the CCF peaked at lag −2 and −1 for the variables T_mean and FDI. The correlation is not uniform, as PR, PDSI, SM, WS, and AI evolve in opposite ways to active fires. The CCF in this area peaked at time h = −2 for T_mean, which shows that the temperature recorded in month t–2 is associated with the number of fires detected in month t – 2, thus showing that the impact of temperature takes effect from the second (drying of vegetation) month before it worsens. The variables PR, SM, ETR, ETA and PDSI peaked at lag 0, 9, −3, 8 and 17 respectively. These results indicate that these factors influence or favour active fires about, 9 (SM), 8 (ETA) and 17 (PDSI) before and −3 (ETR) their occurrence, on the other hand, PR does not show a septic month.

In general, PDSI does not significantly impact active fires in the two ecological zones. However, ETR, RH, SM, T_max, T_mean, T_min, and FDI are the positively correlated variables influencing area burned in the Guinea savannah zone and the other negatively correlated variables show an opposite relationship, while in the Forest- Savannah mosaic zone, ETA, ETR, RH, T_max, T_mean, T_min, and FDI are the variables that positively contribute to fire starts.

4. Discussion

4.1. Variability and trends of climate variables in ecological zones

The study of climatic parameters showed that the study areas face climatic variability during the study period (1991–2021). However, the intensity of the variation was not uniform across the ecological zones. Temperatures have increased over the study period. Average temperatures increased by about +0.8°C (Guinea-Savannah) and +0.2°C (Forest-savannah mosaic zone) over the period (i.e. 0.02° and 0.01°C/year respectively). This increase could be explained by ongoing global changes (IPCC 2021). The results of the identified mean temperature trends remain lower in the transition zone but higher in the Guinean savannah to the north than those observed over the whole of Ghana by Abbam et al. 2018. They observed that mean decadal temperature has increased by (0.6°C from 26.8°C (1900–1910) to 27.4°C (2006–2015) which according to their research pointed to a national mean temperature increase of 1.6°C over the period 1961–2010 (i.e. 0.03°C/year). However, this value remains in line with the average increase of 0.85°C over the period 1885–2012 (or 0.01°C/year) observed by the IPCC at the global level (IPCC 2014) or 0.18°C/decade over the period 1979–2010 (or 0.02°C/year) recorded at the West African level (MoFa 2015; Sultan, Defrance, and Iizumi 2019). These temperature increases are particularly pronounced during the period 2016–2021 according to the results and will be seen to increase by 0.5°C under RCP 2.6 and 2.5°C under RCP 8.5. Maximum air temperatures are likely to increase by 1°C under RCP 2.6 and 2°C under RCP 8.5 by the year 2080 (Klutse, Owusu, and Boafo 2020). The period 2016–2021 is thus recorded nationally as a hot period, corroborating the results found at the local level (Guinea-Savannah and Forest-Savannah mosaic zone) (Issahaku, Campion, and Edziyie 2016; Asante and Amuakwa-Mensah 2014; Bessah et al. 2022). These thermometric increases negatively impact the immediate environment, thus contributing to the spread of wildfires. Thus, the variability of rainfall and temperature recorded in the two ecological zones may have an impact on some parameters that have recorded a downward trend such as SM PDSI. However, the overall HR showed an upward but not significant trend. This result corroborates the results of Dwamena, Tawiah, and Akuoko Kodua 2022 which show an increase in RH in the Forest-Savannah zone. However, there was a decrease in overall rainfall in the guinea-Savannah and Forest-Savannah mosaic zones of −0.6 mm and −0.70 mm respectively, which was also linked to global warming (IPCC 2021). Similar results were found in West Africa by Nicholson, Funk, and Fink (2018). Similarly, several studies on rainfall and temperature variability at the Ghanaian level (Lacombe, McCartney, and Forkuor 2012; Nkrumah et al. 2022; Oduro-Afriyie & Adukpo 2009; Owusu and Waylen 2009; Paeth and Hense 2004) and in the different ecological zones (Issahaku, Campion, and Edziyie 2016) are consistent with the previous subnational analysis of decreasing rainfall and increasing temperature. The year 1998 was identified as the warmest year within the ecological zones during the first decade of the series studied (1991–2000). Previous observations have shown that 1998 was the warmest year of the twentieth century and had the strongest El Niño Southern Oscillation, resulting in an increase in mean air temperature of 0.5–1°C during the 1997–1998 dry season in Africa (Lean and Rind 2009). Extreme climatic conditions would thus favour the occurrence of vegetation fires, which are becoming increasingly frequent on a national scale.

4.2. Climate variability and fire dynamics

Analysis of the relationships between climatic, environmental parameters and indices using Spearman and CCF correlations showed that the best predictors of the number of fires and area burned were ETR, FDI, T_mean, T_min, T_max in the Guinea-Savannah zone and ETA, ETR, RH, T_mean, T_min, T_max and FDI in the Forest-Savannah mosaic zone. The other covariates were inversely correlated with fire activity. These findings confirm several previous works assessing the impact of climate on wildfires (Kays and Ward 2021; Barbero et al. 2015; Hanan et al. 2021). The increase in temperature and vapour pressure deficit combined with the lack of precipitation has significantly increased the load and aridity of the fuel (vegetation) (Turco et al. 2016; Abatzoglou et al. 2016). Hanan et al. (2021) have shown that climate change was the key driver of increasing burn probability and the frequency of large fires. Thus, the climate components especially those considered in this research influence the behaviour of wildfires differently. The drier Guinea-Savannah area, for example, does not have the same response to fire events as the wetter Forest-Savannah area because of its proximity to the lake. These different realities have also been highlighted by other authors when considering ecological conditions, such as fire regimes vary over space and time across the globe, and while climate change is a major factor increasing the frequency of large wildfires (Mouillot et al. 2003; Abatzoglou and Williams 2016). However, it must be emphasised that the increase in humidity is not significant over the study period. Similar results were demonstrated by Guiguindibaye, Belem, and Boussim (2013) who showed that flame heights were inversely correlated with relative humidity in Tanzania. Relative humidity had a strong impact on fire occurrence. Some covariates such as SM and PR, also provide information on their contribution to fire spread. Their negative correlation shows that their decrease favours the start of fires and that their increase limits fires, but the results show a decreasing trend. Thus, we can be sure in this case that they are the second key factors after T_mean, T_max, T_min and RH in the study areas (Jones et al. 2022).

Long days of low humidity in a month tend to dry out all forest fuels. Even a few days of low humidity can increase fire risk (Xiao, Zhang, and Ji 2015). Although AI is not an indicator of fire activity in this research, they indicate the vulnerability of areas to fire as it is well correlated according to Spearman's correlation, which could give an idea of the influence of fire on microclimate (Strydom and Savage 2016). Archibald et al. (2009), showed that rainfall and dry season length are determinants of burnt areas in southern Africa. However, in the correlation analyses, FDSI was found to be weakly correlated, despite being the variable that informs on hydrological drought. In addition, the DM shows a negative correlation, which indicates that its decrease would set the stage for the passage of fires. It is correlated with the same CCFmax in both ecological zones. Therefore, it favours the spread of fires in both ecological zones.

It can be difficult to disentangle the impacts of individual drivers on fire activity because fire is the result of the simultaneous occurrence of three factors: a stock of fuel; fire weather conditions that are sufficiently dry to desiccate the fuel; and a human or natural ignition source (Abram et al. 2021; Bistinas et al. 2014; Forkel et al. 2019; Kelley et al. 2019; Pausas and Ribeiro 2013; Teckentrup et al. 2019) Therefore, the influences of climatic variables on the number of fires should not be generalised to other regions of the country or another area of fire activity. Krawchuk et al. (2009) examined global fire occurrence with climatic variables and showed that the average temperature of the warmest month, annual precipitation and the average temperature of the wettest month are the most important explanatory variables. However, this nexus of drivers and constraints on fire leads to debate about the causes of major wildfire events and the contributions of bioclimatic and human factors to those events (Jones et al. 2022). Thus, the potential benefits to mitigating avoided fire risk of meeting the 1.5°C ambitious targets of the Paris Agreement rather than the 2.0°C commitment, our results further indicate that weather-related fire risks could also rise substantially if the 2.0°C commitment is not achieved (Jones et al. 2022).

5. Conclusion

In the present study, the analysis of climatic and environmental parameters crossed with fire variables allowed us to confirm the effective existence of a relationship between climate variability and fire in the ecological zones considered (Guinea-Savannah and Forest-Savannah mosaic zone) by this research. Indeed, the objective is to find probable correlations and this was done by Spearman's correlation. This correlation was then evaluated and analysed through cross-correlation function (CCF) to detect the most relevant parameters in terms of impact on the fire. The result of this research is that ETR, FDI, T_mean, T_max, T_min, RH, ETA and SM are the major influences on fire spread in both ecological zones but with some differences in terms. Mostly temperature is increasing with a decrease in rainfall in both zones as a result of the variability analysis. This study used the Google Earth Engine platform for climate data acquisition. Currently very promising as it can process a very large amount of data in a short time. It is also essential for data storage and is accessible at any time. This should be adopted especially in developing countries as it will allow the data to be updated as the written codes will be preserved and can be updated easily. The innovative aspect of this research is the use of data from artificial intelligence platforms like GEE. This option is one of the few options used in Ghana and will make monitoring less complex in the near or distant future. Thus, in this study, the code can be updated at different times easily to update climate data in the study areas for long-term monitoring of climate variability and its impact not only on fires but on many aspects such as agriculture, flood monitoring and other extreme weather events. This research could lead to the prediction of wildfires using machine learning techniques.

Acknowledgements

The authors are particularly grateful to the University for Development Studies, West African Center for Water, Irrigation and Sustainable Agriculture (WACWISA) for providing logistics to facilitate the completion of this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data (climatic and environmental data) used in this paper are respectively available at https://code.earthengine.google.com/fa9846d34c532e5f6d043ad4f7376640; https://code.earthengine.google.com/0eba04b7ded613d5a05e811ea2b5ee36 for Guinea-Savannah zone and Forest-savannah zone and fire data on https://doi.org/10.5281/zenodo.7749206.

Additional information

Funding

This research was made possible through funding support from the Islamic Development Bank (IsDB).