Understanding the Local Carbon Fluxes Variations and Their Relationship to Climate Conditions in a Sub-Humid Savannah-Ecosystem during 2008-2015: Case of Lamto in Cote d’Ivoire

The temporal variations of the Gross Primary Productivity (GPP), the Total Ecosystem Respiration (TER) and the Net Ecosystem Exchange (NEE), and their responses to meteorological conditions (e.g. temperature, radiative flux and precipitation) at Lamto, in wet savannah region across Côte d’Ivoire are analyzed using GFED-CASA and daily meteorological data recorded over the 2008-2015 period. The study shows the links between these carbon fluxes and climate variability at Lamto that is subject to high anthropogenic pressures and seasonal bushfires. The correlative statistics from multiple regression methods were used to assess the different relationships and show how they change in time. The results show important seasonal variability in the Gross Primary Productivity and the Total Ecosystem Respiration mainly associated with the changes in temperature and radiative flux. In addition, the statistical analysis suggests a high correlation between meteorological conditions and the GPP and TER. These climatic conditions may explain 83% and 79% of the variances of GPP and TER respectively. Moreover, the interannual variability of the Net Ecosystem Exchange indicates that around Lamto, in the subhumid savannah, the ecosystem behaves as a carbon sink similar to other West African ecosystems. On

Change (UNFCCC) by Côte d'Ivoire, whose objective is to stabilize greenhouse gas concentrations in the atmosphere at a level that prevents dangerous anthropogenic interference with the climate system [22], the Lamto region has become an area of interest for research on climate change due to GHG. This region offers enormous environmental and socio-economic potential for Côte d'Ivoire and climate impacts are not well understood. In this context, the current study aims to analyse and understand the responses of carbon fluxes (i.e. GPP, TER and NEE) to climate variations (air temperature, precipitation and radiative flux) at Lamto using data from the GFEDv4.1 inventory [23] and ground-based observations. This is mainly done through the assessment of the possible links between the carbon fluxes and climate variability at Lamto using correlative statistics and the evaluation of the different relationships and how they change in time. In addition, the long and short-term trends in these carbon fluxes are also analyzed by a diagnostic approach based on linear regression by least square fit to detect the period of the significant changes in their variations.
The work is organized as follows: Section 2 describes the study area, data and methodology used. Section 3 presents a statistical analysis of the response of carbon flux components (GPP, TER, NEE) to temporal variations in air temperature, precipitation and radiative flux recorded at Lamto station. A conclusion and perspectives are provided at the end.

Study Area
The ecosystem of Lamto (5˚02W and 6˚13N, Figure 1) is located in a tropical subhumid savannah across the Sudano-Guinean Transition Area [24]. The region is of about 160 km north of Abidjan and its climate is controlled by the West African Monsoon (WAM). The recorded annual mean rainfall is about 1200 mm, with important seasonal and interannual variability [24] while the mean annual temperature is ~27˚C with a seasonal temperature range of ±2˚C [25]. The main dry season extends from December to February and the wet season is from March to November, with a short dry season in August ( Figure 2). The region is characterized by tropical ferruginous soils under savannah, ferralitic soils under forest, hydromorphic soils at the bottom of the hill and black soils over amphibolites [26]. All these features offer to the Lamto region significant environmental and socio-economic potential.

Ground-Based Observation Data
The meteorological data used in this study are air temperature (˚C), rainfall (mm) and radiative flux (J/cm 2 ) recorded daily at the geophysical station of Lamto over the 2008-2015 period according to WMO standards. These data have been used to document the climate variability at Lamto [24] [32] and are very often stress factors for vegetation.

Methods
This work is based on a statistical analysis of carbon fluxes (GPP, TER and NEE) in relation to climatic conditions (temperature, rainfall and radiative flux). The responses of carbon fluxes to these climatic conditions are evaluated with linear models using the multiple regression method. The correlative statistics and their significance (p-value) according to Pearson [33] were also calculated. To homogenous the climatic data time series into line with that of the GFED-CASA data, the daily meteorological parameter was averaged monthly. In addition, the GPP, TER and NEE components of the carbon flux were calculated from the NPP and Rh variables of the GFED-CASA database following Equations (1) [34], (2) [21] and (3) [17] below. (1) where, CUE (value without unit) refers to the Carbon-Use Efficiency by the considered ecosystem [35].  Figure 3 shows the seasonal cycles of TER and GPP (in absolute values) observed at Lamto over the 2008-2015 period. The carbon flux TER representing the sum of autotrophic (i.e. CO 2 emission by the plant into the atmosphere) and heterotrophic (i.e. CO 2 emission from the decomposition of organic matter) respirations [36] shows a seasonal cycle with strong variations. These seasonal variations are characterized by a rapid decrease in flux values from April (182 gC·m −2 ·month −1 ) to August (118 gC·m −2 ·month −1 ) followed by an abrupt increase until November (185 gC·m −2 ·month −1 ) and a decrease until February (137 gC·m −2 ·month −1 ). In addition, the GPP (i.e. the total CO 2 captured by chlorophyll plants from the atmosphere through photosynthesis) [39] [29], shows a seasonal profile contrasted to that of the TER. As the TER, the GPP cycle is also characterized by a strong seasonal variation. However, there are two decrease phases in seasonal GPP values, from February (1883.35 ppb) to April (−204 gC·m −2 ·month −1 ) and from August (−94 gC·m −2 ·month −1 ) to November (−185 gC·m −2 ·month −1 ) and, two increase phases, from April (−204 gC·m −2 ·month −1 ) to August (−94 gC·m −2 ·month −1 ) and from November (−210 gC·m −2 ·month −1 ) to February (−120 gC·m −2 ·month −1 ). Globally, the seasonal cycle of GPP ranges from −213 gC·m −2 ·month −1 in November to −93 gC·m −2 ·month −1 in August. In addition, the breaks observed in seasonal trends of TER and GPP appear quite significantly during changes in rainfall regime (i.e. GSP/PSS/PSP/GSS/GSP). This behavior could probably be explained by the presence of favorable conditions (e.g. radiative flux, soil temperature, sensitive heat/latent heat ratio, biospheric mutation, etc.) to significant monthly variations in GPP and TER fluxes. However, a significant and important correlation (R 2 = 0.985 and p-value = 1.09·10 −9 ) is found between the monthly variation of TER and GPP (in absolute value). This result suggests that TER and GPP could be controlled by the same mechanisms such as drought. Indeed, Granier et al. [40] showed that drought decreases both GPP and TER fluxes, which is consistent with their seasonal trends associated to the recorded low values in the Lamto region over the study period ( Figure 3). In addition, the seasonal variation in the observed TER/GPP ratio is very significant, ranging from 0.86 in November to 1.26 in August with a mean value of 0.99. This mean value is higher than the TER/GPP ratio obtained in other regions. For example, Janssens et al. [41] found a mean ratio value of 0.80 in 18 European ecosystems while Law et al. [42] obtained 0.83 on the interval [0.55 -1.2] in various ecosystem types, including grasslands and crops.

Trends Analysis
An analysis of long and short-term trends in TER and in GPP is provided using a statistical diagnosis based on linear regression by least square fit [43]. This method objectively detects one or more trend breaks in the TER and GPP time series when they occur. The black contour provides significance at a 95% confidence level from the student t-test. variability (i.e. time segment ≤ 5 years) is related to an alternation of positive and negative changes that are associated to significant small-time scale fluctuations in TER and GPP.

Parametric Approach by Linear Models
The equations are an exception to the prescribed specifications of this template. You will need to determine whether or not your equation should be typed using either the Times New Roman or the Symbol font (please no other font). In order to know the local meteorological variable(s) (i.e. explanatory variable) that control the carbon fluxes (e.g. GPP and TER) variability, linear models were established with the corresponding determination coefficients (R 2 ) and their significance (p-value) ( Table 1). This approach is used to model carbon fluxes as a function of radiative flux and/or temperature and/or precipitation. The different linear functions (Equations (4)-(6), Equations (11)-(13)) obtained show that the monthly flux of GPP and TER are better correlated to local climatic conditions when the "pilot" variable is the radiative flux (Fr) (Equation (4) and Equation (11)) unlike temperature (Te) and rainfall (Pr). Indeed, the values of the correlation coefficient are important and very significant for GPP (R 2 = 0.50, p-value = 0.02) and TER (R 2 = 0.52, p-value = 0.008) with the radiative flux (Fr). However, there is no significant correlation between GPP and rainfall (R 2 = 0.13, p-value = 0.25) and temperature (R 2 = 0.25, p-value = 0.05), and also between TER and rainfall (R 2 = 0.17, p-value = 0.18) and temperature (R 2 = 0.28, p-value = 0.05). On the other hand, the results are improved when linear models with two (Equations (7)-(9)) and three (Equations (14)-(16)) variables are used. As in a humid tropical ecosystem such as that of the Lamto region, where climate variables are dependent and interact strongly with each other [24], the three-variable linear model would then be more representative of the impact of these meteorological conditions on the variability of GPP and TER. However, Equation (7), Equation (10), Equation (14) and Equation (17)) show that the effect of rainfall is marginal for GPP and TER seasonal variations. Considering rainfall in the linear model reduces the determination coefficient by 0.03 for the GPP and increases it by 0.02 for the TER while reducing their degree of significance by increasing p-value. In addition, the less significant effects due to rainfall can be integrated into the constant term used in each two-variable linear model (Equation (7) and Equation (14)) involving radiative flux (Fr) and temperature (Te). These linear models would then be more representative of the exchanges between the Lamto ecosystem and the atmosphere. Table 1 shows that GPP and TER are also better correlated under these conditions. Indeed, Equation (7) and Equation (14) explain 83% (p-value = 0.0004) and 79% (p-value = 0.000077) of the GPP and TER variances, respectively with high significance. In this case, the constant terms 1466.18 gC·m −2 ·month −1 in Equation (7) and 812.44 gC·m −2 ·month −1 in Equation (14) represent potential non-linear effects or coupled effects of radiative flux (Fr) or temperature (Te), such as those resulting from interactions with other variables. Equation (7) and Equation (14) obviously ignore other as-pects of local meteorological conditions, such as rainfall, relative humidity, wind speed and non-climatic factors, such as soil type, soil moisture and vegetation type. They also represent a useful first-order estimate of the seasonal variability of carbon fluxes for the Lamto ecosystem over the 2008-2015 period.

GPP and TER Responses to Radiative Flux and Temperature as Predictors in a Multiple Linear Regression
In this section, approach by multiple regressions previously made (section 3.2.1) shows that linear models with two parameters and using radiative flux and Temperature (i.e. Equation (7) and Equation (14)) as predictors better explain the seasonal variations in GPP and TER in the Lamto region. In addition, the Equation (7) and Equation (14) show that any increase in temperature (Te) systematically leads to decreases in GPP and TER fluxes with rates of −68.75 gC·m −2 ·month −1 ·˚C −1 and −35.29 gC·m −2 ·month −1 ·˚C −1 respectively. These results indicate that when temperature increases (resp. decreases), the ecosystem's carbon requirements (i.e. GPP flux) are reduced (resp. increased) to favour (resp. inhibit) the respiration mechanisms inducing a significant increase (resp. decrease) of TER exchanges between the ecosystem and the atmosphere. On the other hand, any increase (or decrease) in the radiative flux (Fr) also leads to an increase (or decrease) in the GPP and TER with rates of +0.45 gC·m −2 ·month −1 ·J −1 ·Cm 2 et +0.23 gC·m −2 ·month −1 ·J −1 ·Cm 2 respectively. These results clearly indicate that the radiative flux (Fr) favours the atmosphere-ecosystem exchanges and show the GPP and TER responses to these radiative flux effects.   [49] [50] in response to the seasons of increasing and decreasing temperature. In addition, many authors as Lafleur et al. [51], Zhang et al. [52] and Lee et al. [48] pointed out that low temperatures limit photosynthetic activity in ecosystems and reduce GPP flux. These seasonal variations of temperature also influence the TER flux characteristic of respiration during the year [11] [53]. They also show lagged peaks behind the peaks of GPP and TER fluxes during the December-January-February dry season (Figure 2 and Figure 3), probably due to a hysteresis effect whose underlying mechanisms and processes are not totally explained in many studies [46] [54] [55] [56]. This phenomenon of hysteresis has already been mentioned in the work of Zeppel et al. [57]. These authors underline that hysteresis occurs when an increase in a given independent variable α  does not lead to the same response scale in a dependent variable β, compared to a decrease in the variable α of the same magnitude.

Discussion
In addition, the role of temperature at the plant stage has been the topic of much investigation [53] [58] in a context of climate and agricultural development. These authors have shown the importance of temperature in the interactions between vegetation and the atmosphere. Indeed, temperature determines water needs and strategies to ensure its availability to fulfill demand. However, Niu et al. [46] point out the importance of taking into account both radiative transfer and temperature in the response of the GPP flux.
In addition, the models defined by Equation (7), Equation (10), Equation (14) and Equation (17) ( Table 1) indicate that the different responses of GPP and TER fluxes to environmental factors can be linear. These two-and three-variable equations involving on the one hand, radiative flux and temperature, and radiative flux, temperature and precipitation on the other hand explain most of the seasonal variation in carbon fluxes. Among these environmental factors, the radiative flux is the one that shows the most significant effect (i.e. pilot variable) on carbon fluxes, by explaining respectively 50% and 52% of the seasonal variances of GPP and TER (Equation (4) and Equation (11)). On the other hand, the results are improved when the radiative flux (Fr) and temperature (Te) are considered (Equation (7) and Equation (14)). In this case, these two variables explain 83% and 79% respectively of the seasonal variations in GPP and TER fluxes. These observations show the important and significant role of temperature and radiative flux in seasonal carbon fluxes responses. In contrary, the explicit consideration of rainfall (Pr) in the different linear models (Equation (10) and Equation (17)) does not show significant changes in the variance rates of these carbon fluxes. However, several studies [59] cited by [60]- [68] which analyzed the impact of climate variables on GPP and TER fluxes in several terrestrial ecosystems, showed the main role of precipitation in the seasonal and/or interannual dynamics of these carbon fluxes. Indeed, precipitation contributes to increasing 1) autotrophic respiration through increased vegetative growth [63] [65] [69], and  [75].
Moreover, in the case of this study, the results clearly indicate that only the combined effects of radiative fluxes and temperature explain most of the seasonal variability of carbon fluxes. These results also show the particular behaviour of the Lamto region compared to other ecosystems [59] [62] [65] by its geographical position (in the Sudano-guinean transition area at 6˚31N and 5˚02W) and its rainfall regime (very marked two rainy and dry seasons) and by the strong interannual variability of its climate, which can change from humid to subhumid [24]. The response of carbon fluxes to environmental factors (i.e. radiative flux, temperature and rainfall) can be estimated by linear models with two or three variables depending on local conditions; for example, in regions where rainfall is highly seasonal (especially in regions where rainfall is low or non-existent for several months), linear models with two variables involving radiative flux and temperature (i.e. Equation (7) and Equation (14)) could be considered. Otherwise, in regions where rainfall is very high, three-variable models involving radiative flux (Fr), temperature (Te) and rainfall (Pr) could be used to take into account the effects of this rainfall. In addition, the interannual variability of the Net Ecosystem Exchanges (NEE) shows negative values ( Figure 5) in the range [42.16; 97.95; −42.16] gC/m 2 /year (in absolute value), indicating that the ecosystem in the Lamto region behaves as a carbon sink. This behaviour is also observed in several modeling studies and direct and/or indirect in-situ measurement on various West African ecosystems, such as Niger/Wankama1 [59], Senegal/Dahra [76], Burkina-Faso/Nazinga [68] and Benin/Nalohou [66].
These studies have shown the singular nature of these mentionned ecosystems which behave as carbon sinks.In addition, the absence of a well-defined relationship between annual NEE and environmental factors (e.g. temperature, rainfall and radiative flux) could be due to the fact that: 1) there are several sources of carbon (CO 2 ) for ecosystem respiration [53], and each source has its own control factors, such as soil moisture, extreme pedological conditions [38] [53] [76]- [81]; 2) the observed interannual variations in temperature and radiative flux have small amplitudes less than 1.26˚C and less than 120 J·Cm −1 respectively [21] [82].

Conclusion
This study assessed the different responses of carbon fluxes (i.e. GPP, TER and NEE) to meteorological factors (i.e. Temperature, Radiative flux and Rainfall) over the 2008-2015 period around Lamto region. A statistical approach using multiple regressions highlighted the seasonal variations of GPP, TER and NEE.
In addition, the analysis based on linear models established with one climatic variable, show that the radiative flux, by explaining 50% and 52% of the variances of GPP and TER respectively, seems to be the most important environmental factor affecting the carbon fluxes. On the other hand, using linear model based on two climatic variables as predictors, the combined effects of the radiative flux and temperature are suggested to key role in the variation of carbon fluxes by explaining 83% and 79% of the variances of GPP and TER respectively. The seasonal changes in rainfall for its part, seems to have a negligible impact on the variation of the carbon fluxes. This is in agreement with previous studies which show that the interannual variability of carbon fluxes in West Africa is related more to the annual rainfall than the seasonal rainfall. This low impact could be linked to the fact that seasonal rainfall is low or sometimes non-existent over several months. In addition, the interannual variability of Net Ecosystem Exchange (NEE) has shown that the Lamto region behaves like a carbon sink similar to other West African ecosystems. However, there is no clear link found between NEE flux and temperature, radiative flux and rainfall. This absence of link suggests that the dynamics of NEE could either exhibit threshold responses to the effects of climatic factors or could not be determined mostly by climatic factors, but much more by exogenous parameters such as soil temperature, vegetation type, soil type, anthropogenic pressure and canopy structure. In order to better understand the main mechanisms responsible for the variability of carbon fluxes and to significantly reduce the uncertainties on the carbon balance in West Africa in a context of climate variability and change, it is necessary to take into account all these above-mentioned parameters which are beyond the scope of the current study.