The study was conducted in the agroforestry parkland of Sob village (Niakhar municipality, Fatick region), located in the groundnut basin of Senegal, within the Sahelo-Sudanian climatic zone of West Africa (Fig. 1). The climate is characterized by a long dry season (8–9 months) with high temperatures and strong diurnal variations, and a short rainy season from late June to mid-October (Delaunay et al., 2019).

Soils are locally known as “Dior” and classified as Arenosols (IUSS Working Group WRB, 2022). The topsoil has low organic matter (< 1 %) and phosphorus (< 3 mg kg⁻¹), a sandy texture (> 85 % sand), and an acidic pH (Malou et al., 2021; Siegwart et al., 2022). Rainfed agriculture predominates. The main cropping system includes pearl millet (Pennisetum glaucum L.) and groundnut (Arachis hypogaea L.) in biennial rotation, with occasional intercropping of cowpea (Vigna unguiculata L.).

The site hosts the “Faidherbia Flux” station (14°29^′44.916^′′ N; 16°27^′12.851^′′ W; FLUXNET ID: SN-Nkr), a long-term research platform for monitoring ecosystem services in agroforestry systems. It is dominated by F. albida, a nitrogen-fixing, reverse-phenology tree with deep roots accessing groundwater (Roupsard et al., 1999). The tree density is ∼ 13 trees ha⁻¹, with canopies covering ∼ 10 % of the soil surface (Roupsard et al., 2020). The EC tower is installed at 20 m height, approximately 12.5 m above the canopy. The study field is a typical 'bush field', characterized by low soil fertility, no mineral fertilization, and off-site export of crop residues and manure (Malou et al., 2021).

Statistical analyses were performed using the R software (R Core Team, 2023). To compare the mean values of climatic parameters between the FS and Sh situations, a non-parametric Mann-Whitney test was used when both the normality (shapiro.test) and the homogeneity of the variance (Levene Test, R package “Car”; Fox et al., 2023) were not confirmed. This approach was similarly applied to compare the seasonal dynamics of CO₂ fluxes between FS and Sh, as well as between the chamber-based and Eddy Covariance (EC) methods. Means and standard deviations were computed using the “skim” function from the R package “skimr” (Waring et al., 2022).

Respiration (Rch) (Eq. 4) and GPP (GPPch) models (Eq. 7) were fitted using non-linear least squares regression, implemented in the library in R “nls.multstart” (Padfield et al., 2025). For the GPPch model, parameters α and β with non-significant p-values were removed, and then the remaining values were interpolated and smoothed using a “spline” function from the “zoo” library in R (Zeileis et al., 2024). Ordinary least-square linear regressions were fitted between the measured and the modeled values derived from. Model performance of Eqs. (4) and (7) was evaluated by fitting ordinary least-square linear regressions between the measured and the modeled values using R2, root mean square error (RMSE), and the bias metrics. Given that the primary objective of these equations was to accurately reproduce the seasonal dynamics of the CO₂ fluxes to fill gaps in data, particular emphasis was placed on R2, with a higher value reflecting a better fit of the model to the measurements.

Correlation analysis was conducted between chamber CO₂ fluxes (FCO₂ch, Rch, GPPch) and soil temperature (Tsoil, °C), air temperature (Tair, °C), VWC, the leaf area index of groundnut plants in the chambers (LAIch), and the fitted parameters for respiration – Rref – and photosynthesis – α and β. This analysis was performed using the “cor.test” function from the “stats” package in R (Lüdecke et al., 2021), applying the Spearman method.

The threshold of the daily mean soil temperature (Tsoil, °C) at which the cumulative daily respiration (Rch, g C-CO₂ m⁻² d⁻¹) began to decline was determined using segmented regression from the R package “segmented” (Muggeo, 2003). The associated uncertainty (standard error) of this estimate was evaluated through a bootstrap procedure.

The standard error of the total annual flux was estimated using the error propagation method. This calculation considered the mean standard deviation of daily fluxes (g C–CO₂ d⁻¹) and the effective number of measurement days (365). For each FS and Sh condition, the mean daily standard deviation was multiplied by the square root of 365 to obtain the annual standard error. The resulting values were then weighted by 90 % for FS and 10 % for Sh to derive the overall standard error of the annual flux sum, which was subsequently converted to Mg C–CO₂ ha⁻¹.

During the experiment, the cumulative rainfall was 550 mm, which was representative of the interannual average. Precipitations were lowest in July and highest between August and September, a period that typically corresponds to the peak of the rainy season (Fig. 2a). Global radiation ranged between 5.8 and 32.4 MJ m⁻² d⁻¹ (data not shown). The daily mean VWC in the chambers showed significant variation, ranging from 1 % at the end of the dry season to a maximum of 30 % during the rainy season (Fig. 2a). While VWC was similar during the rainy season, it remained consistently higher in FS than in Sh throughout the dry season (p < 0.05), which was unexpected. However, it should be noted that the last rain of October 2021 recharged the FS chambers more effectively, likely due to foliage rainfall interception by F. albida which had just put on leaves at that time, potentially explaining this discrepancy in VWC.

Within the chamber, the daily mean Tsoil ranged from 26 °C in April to 37.5 °C at the end of the dry season (Fig. 2b), while Tair varied between 23.7 °C and 35.5 °C (Fig. 2c). However, during instantaneous daily peaks, Tsoil could exceed 45 °C in May (data not shown). As expected, both daily mean Tsoil and Tair were significantly higher in FS compared to Sh situations (p < 0.05), with Tsoil and Tair averaging respectively 1 and 0.5 °C lower under the tree canopy.

The seasonality of daily cumulative of GPPch.stand showed similar dynamics between FS and Sh, with higher variability during the rainy season than during the dry season (Fig. 5). Daily total Rch peaked during the rainy season at 5.1 g C-CO₂ m⁻² d⁻¹ for FS and 5.4 g C-CO₂ m⁻² d⁻¹ for Sh, while the maximum GPPch.stand values were comparable at around -15.0 g C-CO₂ m⁻² d⁻¹ for both FS and Sh (Table 1; Fig. S7a, b, c, and d). In the dry season, Rch decreased (Fig. 5), averaging 0.5 g C-CO₂ m⁻² d⁻¹ for FS and 1.0 g C-CO₂ m⁻² d⁻¹ for Sh. GPPch declined well before harvest (senescence) and remained nil during the dry season (Fig. 5). During the rainy season FCO₂ch peaked in absolute value at around 11.0 g C-CO₂ m⁻² d⁻¹ for FS and Sh (Fig. 5) (Table 1; Fig. S7e and f), while FCO₂ch values were the same as Rch during the dry season. In absolute terms, the mean Rch and GPPch were significantly higher under Sh as compared to FS, by factors of 1.3 and 1.2, respectively. Conversely, the mean FCO₂ch was significantly higher in absolute value under FS (0.4 g C-CO₂ m⁻² d⁻¹) than under Sh (0.2 g C-CO₂ m⁻² d⁻¹) (Table 1).

The annual cumulative Rch values were 392.8 g C-CO₂ m⁻² for FS and 574.5 g C-CO₂ m⁻² for Sh. The GPPch fluxes reached -539.5 g C-CO₂ m⁻² for FS and -632.6 g C-CO₂ m⁻² for Sh. The net annual cumulative C exchange (FCO₂ch) was -146.7 g C-CO₂ m⁻² in FS and -58.1 g C-CO₂ m⁻² in Sh.

The chamber-based daily cumulative respiration (Rch) and GPPch showed significant and positive correlations with the leaf area index (LAIch), both at a distance from the trees (FS) and under the trees (Sh) (Table 2). The influence of LAIch on GPPch was stronger (r = 0.86 for FS and Sh) than its influence on Rch (r = 0.60 for FS; r = 0.69 for Sh). Soil VWC was also positively correlated with Rch and GPPch, both in FS and Sh. However, the influence of soil VWC on Rch was stronger under Sh compared to FS, while its influence on GPPch was similar in both situations (FS and Sh). Soil temperature showed weak negative correlations with Rch (in FS and Sh) and with GPPch (only in Sh). Finally, no significant correlations were found between Tair, and any of the CO₂ fluxes (Table 2).

The chamber-based daily total CO₂ fluxes, gap-filled and weighted according tree cover were compared with the fluxes obtained using the EC method (Fig. 6).

During the rainy season, both total respiration and GPP showed comparable dynamics between the two methods, with synchronised peaks and higher variability compared to the dry season (Fig. 6). The maximum value of Reco.EC, peaked at 13.5 g C-CO₂ m⁻² d⁻¹ (Table 3). The initial value of Rch.stand was comparable to Reco.EC but peaked only at 5.1 g C-CO₂ m⁻² d⁻¹ (Table 3), meaning

a third of the peak of Reco.EC. The maximum GPP, was -14.3 g C-CO₂ and -14.6 g C-CO₂ m⁻² d⁻¹ for GPP.EC and GPPch.stand, respectively (Table 3). This indicates that the LAI-based standardisation and upscaling approach were realistic, at least up to the peak of groundnut growth.

On average, Reco.EC was significantly higher than Rch.stand, by a factor of 2.3. GPP.EC was also significantly higher than GPPch.stand, but only by a factor of 1.2 (Table 3).

During the dry season, Reco.EC and Rch.stand gradually decreased. The values for Reco.EC remained higher than for Rch.stand, which was fairly consistent with the contribution of the Ra tree above-ground compartment, even if this difference seemed to disappear at the end of the dry season (Fig. 6). The measured “Birch effect” was highest for Rch.stand in 2021, but was the opposite in 2022 due to a system failure at the beginning of the rainy season. The maximum value of GPP.EC reached -2.4 g C-CO₂ m⁻² d⁻¹ when the trees were at their maximum of foliage, after harvest and while weeds were still present in the field. However, after the harvest, chamber photosynthesis (GPPch.stand) was nil (Table 3).

During the dry season, the cumulative contribution of F. albida to ecosystem respiration (Ra tree) was 139.6 g C-CO₂ m⁻². This represent 14 % of the total annual cumulative Reco, which was estimated at 1000.0 g C-CO₂ m⁻² (Table S4). The contribution of trees (GPP tree) to total annual GPP in absolute term was 270.2 g C-CO₂ m⁻², equivalent to ∼ 23 % of the total annual cumulative GPP of the ecosystem measured by EC (1180.0 g C-CO₂ m⁻²) (Table S4).

The ratio between these two components (Ra tree/GPP tree) in absolute terms was 0.52, reflecting a carbon use efficiency (CUE) of 0.48 (Table S4).

The upscaled chamber-based annual cumulative total respiration flux (Rch.stand) was estimated to be 4.1 ± 0.18 Mg C-CO₂ ha⁻¹ (Table 4). In comparison, the annual Reco.EC was 10.0±0.49 Mg C-CO₂ ha⁻¹ (Table 4), more than two times larger than Rch.stand.

The upscaled GPPch.stand reached an annual cumulative value of -5.5 ± 0.83 Mg C-CO₂ ha⁻¹, whereas the annual cumulative GPP.EC was -11.8 ± 0.53 Mg C-CO₂ ha⁻¹ (Table 4).

The net annual vertical C balance, based on both methods, was estimated at -1.4 ± 0.46 Mg C-CO₂ ha⁻¹ for chambers (FCO₂ch.stand) and -1.8 ± 0.17 Mg C-CO₂ ha⁻¹ for Eddy Covariance (NEE.EC) (Table 4).

In this study the Lloyd and Taylor (1994) model, based on a modified Arrhenius-type formulation, was used to model nocturnal soil respiration fluxes for estimating daytime respiration. Unlike the classical Arrhenius equation, this model includes the (Tsoil-T0) term in the denominator of the exponential expression (Eq. 4), which inherently limits the effects of high temperatures by progressively reducing the temperature sensitivity of soil respiration as temperatures rise above a given threshold. This structural feature produces a flattening of the respiration–temperature relationship at elevated temperatures, thereby preventing the overestimations (Lloyd and Taylor, 1994).

The Lloyd and Taylor model has successfully been widely applied, primarily in boreal and temperate ecosystems (Lasslop et al., 2010; Reichstein et al., 2003), and relies on the assumption of comparable thermal conditions between daytime and nighttime periods (Juszczak et al., 2012). In our study, instantaneous soil temperatures ranged from 20.7 to 45.8 °C during the day and from 22.1 to 45.0 °C at night, indicating largely overlapping thermal ranges between the two periods. Model parameters were recalibrated using five-day fixed windows, which provided sufficient temporal resolution while capturing seasonal dynamics of soil respiration.

This study represents one of the first applications of the Lloyd and Taylor model in a Sahelian semi-arid context. While Arrhenius based models are known to potentially overestimate fluxes under extreme temperatures due to physiological limitations, over the range of temperatures observed in this study, the modeled soil respiration was not overestimated (Fig. S6a and b). Thus, the model used in this study appears to provide a realistic representation of soil respiration under local conditions. However, this conclusion is site-specific and should not be interpreted as a general validation of temperature-based models across all semi-arid environments. Such models should be systematically validated with respect to temperature to ensure their reliability.

In our agroforestry context, seasonal variability in CO₂ fluxes closely followed rainfall dynamics, peaking during the wet season and declining sharply in the dry season, consistent with soil moisture depletion and crop senescence. This pattern is typical of semi-arid ecosystems (Ago et al., 2016a; Brümmer et al., 2008; Guillen-Cruz et al., 2023; Macharia et al., 2020; Mosongo et al., 2022; Wieckowski et al., 2024).

Respiration and photosynthesis were primarily driven by soil moisture and LAI, reflecting the system's sensitivity to water availability and crop dynamics. Soil moisture enhanced both processes by stimulating microbial activity and supporting plant growth (Borken et al., 2002; Conant et al., 2004; Merbold et al., 2009; Yu et al., 2020; Zhao et al., 2016). The stronger correlation between soil moisture and respiration under F. albida canopy (Sh: r = 0.78) compared to full sun (FS: r = 0.51) suggests greater microbial sensitivity to moisture beneath trees. This likely reflects enhanced substrate availability, resulting in stronger post-rainfall respiration pulses (Meisner et al., 2015) and supporting the “fertile island” effect, where trees improve local soil conditions (Eldridge et al., 2024). Photosynthetic capacity also responded to soil moisture, as shown by positive correlations with LAI and key physiological traits such as light use efficiency (α) and maximum CO₂ uptake rate (β) (Gonsamo et al., 2019; Qiu et al., 2023; Zhang et al., 2024b).

In contrast, the influence of soil temperature (Tsoil) on respiration was weakly negative in both FS and Sh, indicating a thermal threshold beyond which respiration declined – estimated at 32 ± 1.5 °C in FS and 29.5 ± 1.9 °C in Sh (Fig. S9a and b), similar to findings in Eastern Ghana (Owusu-Prempeh et al., 2024). This inhibition likely results from decreased enzymatic and microbial activity under combined heat and water stress (Liu et al., 2018; Richardson et al., 2012). In semi-arid regions, soil respiration often becomes decoupled from temperature due to seasonal moisture constraints (Jia et al., 2020; Tucker and Reed, 2016; Warren, 2014), with microbial activity limited during dry periods despite favourable temperatures. This decoupling helps explain the weak or absent correlation between Tsoil and soil moisture (Fig. S8b), particularly under Sh (r = –0.28). Management practices such as organic inputs can also modulate these dynamics, adding further variability to soil respiration responses (Meena et al., 2020; Oyonarte et al., 2012; Rong et al., 2015; Xue and Tang, 2018).

Mean total soil respiration values were consistent with those reported in other low-input agricultural systems across sub-Saharan Africa (Mapanda et al., 2010; Pelster et al., 2017; Rosenstock et al., 2016). In full sun (FS), the mean respiration (1.0 ± 0.9 g C-CO₂ m⁻² d⁻¹) closely matched values measured by Wachiye et al. (2020) in a semi-arid Kenyan field at 1158 m altitude (1.1 ± 0.1 g C-CO₂ m⁻² d⁻¹). This similarity likely reflects comparable environmental conditions, including moderate rainfall (∼ 550 mm yr⁻¹) and low soil organic carbon and nitrogen contents (< 1 %) in the 0–20 cm layer of sandy soil. In contrast, respiration under F. albida canopy (Sh: 1.6 ± 1.1 g C-CO₂ m⁻² d⁻¹) was higher, likely due to additional autotrophic respiration from tree roots and greater organic inputs beneath the canopy. Nonetheless, this flux remains close to values observed in low-input sorghum fields on sandy loam soils in eastern Ghana (1.7 ± 1.1 g C-CO₂ m⁻² d⁻¹), despite higher rainfall (950–1000 mm yr⁻¹) in that region (Owusu-Prempeh et al., 2024).

Cumulative annual respiration fluxes fell within the range reported for Sahelian croplands (250–450 g C-CO₂ m⁻²) (Brümmer et al., 2009) and other sub-Saharan African agricultural systems (Kim et al., 2016). The cumulative flux under tree cover is similar to that measured in cassava fields in eastern Tanzania (440 g C-CO₂ m⁻² yr⁻¹), despite the latter receiving higher rainfall (∼ 1115 mm yr⁻¹) (Rosenstock et al., 2016). This convergence may stem from comparable soil fertility constraints, with low soil organic carbon (1 %–1.7 %) and nitrogen contents (< 0.5 %). In contrast, the slightly lower cumulative flux in FS may reflect less favourable microclimatic conditions – such as elevated soil temperatures and increased aridity away from tree cover – limiting microbial activity (see Sect. 4.1).

Across sub-Saharan Africa, soil respiration fluxes based on static chamber measurements show high spatial variability, largely shaped by climate and land use. For example, Owusu-Prempeh et al. (2024) found higher respiration in woodlands (3.8 ± 0.8 g C-CO₂ m⁻² d⁻¹) and grazed areas (2.7 ± 1.7) than in croplands (1.7 ± 1.1) in humid eastern Ghana. This gradient was linked to differences in soil moisture and organic matter. Similarly, Rosenstock et al. (2016) reported much higher fluxes in highland pastures in Kenya (3.8–4.4 g C-CO₂ m⁻² d⁻¹) compared to cultivated fields in eastern Tanzania (1.2 ± 0.2), highlighting the role of vegetation cover and soil fertility.

A notable increase in respiration and photosynthesis fluxes was observed under F. albida trees (Sh) compared at a distance from trees (FS). This increase may indicate the potential role of F. albida in modulating CO₂ exchange dynamics (Rch and GPPch) within this agro-silvo-pastoral system. These results are consistent with preliminary findings from similar environments (Duthoit et al., 2020).

Numerous studies have investigated the effect of tree species on greenhouse gas fluxes, particularly CO₂, revealing significant variations across different ecological contexts (Bréchet et al., 2021, 2025; Klaus et al., 2024; Mazza et al., 2021; Ramesh et al., 2013; Rheault et al., 2024). However, the underlying mechanisms by which trees influence these dynamics are not yet fully understood.

In general, agroforestry systems have been well-documented for their ability to provide a range of ecosystem services (e.g., Assefa et al., 2024; Bado et al., 2021; Kuyah et al., 2019; Rolo et al., 2023). Specifically, Faidherbia-based agroforestry systems may play a crucial role in regulating CO₂ exchanges between the soil and atmosphere. F. albida-based agroforestry systems are recognized for enhancing both soil organic and mineral fertility (Bayala et al., 2020; Dilla et al., 2019; Sileshi, 2016; Sileshi et al., 2020; Stephen et al., 2020), mainly through litter accumulation and direct inputs from livestock excreta under their canopies. Additionally, the extensive root system of F. albida trees helps concentrate mineral nutrients, contributing to the formation of a “fertile island” effect under the trees (Siegwart et al., 2022; Eldridge et al., 2024). Moreover, F. albida improve water infiltration (Diongue et al., 2023; Faye et al., 2020; Sarr et al., 2023), enhance soil moisture retention (Clermont-Dauphin et al., 2023) and contribute to reduced soil temperatures (de Carvalho et al., 2021; Lopes et al., 2024; Sida et al., 2018). These changes foster a more favourable environment for soil microbial activity and crop development (Diack et al., 2024; Diene et al., 2024; Leroux et al., 2020; Roupsard et al., 2020) under the trees compared to open areas. Consequently, this likely explains the stronger effect of soil moisture and the leaf area index of groundnuts on Rch under the trees, resulting in higher total respiration (Table 2). For photosynthesis, the effect of these parameters was similar in both FS and Sh (Table 2). However, the significantly higher intensity of GPPch under Sh can be explained by greater light use efficiency (α) and a higher maximum CO₂ uptake rate at light saturation (β) in this shaded environment. In agroforestry systems, light use efficiency can at least partially mitigate the reduction in photosynthetically active radiation under tree canopies (Charbonnier et al., 2017).

Similar results have been observed in different climatic conditions and ecosystems. Gomes et al. (2016) investigated soil respiration using mobile chambers (LI-8100-102 model) under trees in coffee-based agroforestry systems and in the open areas in Minas Gerais, Brazil. These studies were conducted with agroecological management practices, such as weeding, intercropping maize between coffee rows, and mulching. The agroforestry systems exhibited lower air and soil temperatures (at 5 and 10 cm depth) and higher air and soil humidity compared to the open areas (Gomes et al., 2016). These authors observed greater spatial variability in soil respiration in agroforestry system (34.1 %) compared to the open areas (24.2 %). This variability was mainly linked with fluctuations in labile carbon and total nitrogen, reflecting more favourable soil microclimate for microbial activity in agroforestry system. In contrast, soil temperature (10 cm depth) accounted for most of the variability observed in the open areas, where the absence of tree canopy resulted in high soil temperatures and low soil moisture (Gomes et al., 2016). Likewise, Van Haren et al. (2010) reported 38 % higher soil respiration near large trees (DBH > 35 cm) in clay-rich Amazonian forests compared to open sites. Interestingly, the magnitude of CO₂ fluxes was independent of tree species, indicating that canopy effects may outweigh species-specific traits in some contexts. In our study, F. albida's influence on CO₂ fluxes aligns with this general pattern observed in tropical agroforestry. However, the mechanisms linking individual tree species to microbial and physicochemical drivers of CO₂ dynamics remain insufficiently understood and warrant further investigation (Jevon et al., 2023).

A rapid increase in soil respiration was observed following the first rainfall events, particularly under F. albida. This phenomenon can be attributed to the lower bulk density of the soil under the trees (Clermont-Dauphin et al., 2023; Siegwart et al., 2023), which potentially lead to CO₂ accumulation during the dry season due to higher soil organic matter (SOM) (Siegwart et al., 2023). Additionally, the sensitivity of microbial communities to subtle variations in soil moisture, compounded by the tree effect, may further explain this phenomenon, as outlined in Sections 4.1 and 4.3. This phenomenon, known as the “Birch effect” (Birch, 1958), has been reported across various semi-arid ecosystems in sub-Saharan Africa (Ago et al., 2016b; Fan et al., 2015; Wieckowski et al., 2024), as well as other semi-arid ecosystems globally (Roby et al., 2022; Yan et al., 2014; Yu et al., 2020). In these contexts, the “Birch effect” may result from the displacement of soil gas phases by the piston effect generated during water infiltration (Singh et al., 2023). Furthermore, microbial communities in semi-arid environments adopt osmoregulatory mechanisms to withstand water deficit (Warren, 2014), which is particularly pronounced during the dry season. This phenomenon reduces soil microbial metabolism (Schimel et al., 2007). Upon rapid soil rewetting, especially after prolonged dry periods, soil microbial metabolism process is swiftly reactivated, leading to a transient pulse in respiration and a CO₂ release (Barnard et al., 2020; Kim et al., 2012; Manzoni et al., 2020; Vargas et al., 2018). Isotopic signatures of soil respiration provide evidence supporting the hypothesis that these pulses result from the rapid mineralisation of necromass or osmolytes excreted by microorganisms under drought stress (Schimel et al., 2007; Unger et al., 2010). Additional factors may amplify the “Birch effect”. For instance, drying-rewetting cycles can induce physical disruption of soil aggregates, enhance oxygen penetration and thereby expose previously protected organic matter to microbial decomposition (Rabbi et al., 2024). This increases substrate availability and subsequently boosts soil respiration fluxes.

The magnitude of the “Birch effect” is modulated by the severity and duration of drought. Thus, at our study site, given the 8- to 9-month-long dry season, the “Birch effect” is particularly intense. Indeed, extended drought periods promote greater accumulation of microbial necromass and intensify hypo-osmotic stress responses upon rewetting (Singh et al., 2023).

Results revealed high seasonal variability, with higher values during the rainy season compared to the dry season. This seasonal pattern aligns with findings from studies in the Sahel using the EC method for CO₂ flux measurements (Brümmer et al., 2008; Tagesson et al., 2015; Agbohessou et al., 2023, Wieckowski et al., 2024). Comparable patterns have also been documented at the ecosystem scale in other semi-arid environments (Ago et al., 2014; Archibald et al., 2009; Ardö et al., 2008; Jia et al., 2020; Quansah et al., 2015; Williams et al., 2009; Zhang et al., 2024a).

Several comparative studies between chamber and EC methods have reported both congruent and divergent CO₂ flux estimates (Bastviken et al., 2022; Poyda et al., 2017; Riederer et al., 2014; Tang et al., 2008; Wang et al., 2010). In the present study, ecosystem respiration fluxes during the rainy season exhibited notable discrepancies measurements between EC (Reco.EC) and upscaled chamber-based (Rch.stand). This is attributable to differences in the flux components captured by each method. Specifically, Reco.EC included respiration from below- and above-ground tree parts, crops (groundnuts and cowpeas), weeds, and soil, whereas Rch.stand accounted only respiration from below-ground tree, groundnut crop, and soil. Therefore, as expected, Reco.EC (4.6 ± 3.2 g C-CO₂ m⁻² d⁻¹) were significantly higher than Rch.stand (2.0 ± 1.1 g C-CO₂ m⁻² d⁻¹).

For chamber-based GPP measurements, values were standardised (GPP-stand) by the field leaf area index (LAI.field). This allowed it to improve comparability with GPP.EC when trees were leafless in the rainy season. In both cases, GPP accounted only for crops (groundnut and cowpea) and weeds, as trees were non-photosynthetic in the rainy season. Despite this standardisation, GPP.EC values (-5.1 ± 3.6 g C-CO₂ m⁻² d⁻¹) were significantly higher than GPPch.stand values (-4.2 ± 4.3 g C-CO₂ m⁻² d⁻¹). However, no divergence was observed in August, and the intensity of the peak of GPP in September was similar in both methods, but from the onset of groundnut senescence, when weeds became the dominant photosynthetic contributors. Thus, during the groundnut growth season, with leafless F. albida trees and almost no weeds, GPP measurements from EC and chambers generate closely comparable results. Therefore, this provides an initial form of cross-validation between the two methods. It is important to note that the EC method integrates CO₂ fluxes over a larger spatial scale, encompassing all ecosystem components (Baldocchi, 2003), while the chamber method captures fluxes on a smaller scale (i.e., at the 0.25 m² scale). This scale disparity can introduce uncertainties when upscaling chamber-based fluxes to the field, as vegetation composition within chambers does not represent the EC footprint average vegetation. This makes upscaling chamber-based measurements challenging. Nevertheless, the standardisation we applied on chamber photosynthesis by LAI has been relatively successful.

During the dry season, Reco.EC included respiration from below- and above-ground tree parts (with leaves) and bare soil, while Rch.stand measured only below-ground tree and bare soil respiration. Consequently, the difference between Reco.EC and Rch.stand was solely attributable to above-ground tree respiration (Ra tree above-ground). In terms of GPP, chamber measurements were nil, whereas GPP.EC reflected only GPP trees.

The transition period, characterised by groundnut senescence, tree leaf regrowth, and weed proliferation, introduced further complexity, amplifying method-specific discrepancies. Rch.stand measurements facilitated the estimation of tree contribution to Reco.EC (Ra tree) and the verification of the consistency for EC results in terms of carbon use efficiency (CUE), estimated here at 0.48. This value indicates that nearly 50 % of the carbon captured by trees is allocated to biomass. The CUE estimate here is well comparable to the global average across diverse ecosystems, climates, and management practices (0.49 ± 0.14) (Tang et al., 2019). Similar CUE values have been reported for semi-arid grasslands (0.46 ± 0.10), but our value is notably lower than those documented for wetlands (0.61 ± 0.13) (Tang et al., 2019). Overall, these findings reinforce the plausibility of our assumptions regarding the compartment contributions to Reco.EC and Rch.stand, thereby providing a second cross-validation of the EC-Ch comparison. However, despite a frequently assumed CUE of 0.5 in models, global estimates span a broad range (0.20 to 0.82), depending on ecosystem type and management practices (DeLucia et al., 2007; Tang et al., 2019). This underscores the importance of refining carbon flux models to better represent the biophysical processes governing CO₂ exchange in semi-arid agroforestry systems. The combined use of EC and chamber methodologies offers a comprehensive perspective on ecosystem-scale CO₂ flux dynamics, advancing the understanding of carbon cycling in these environments.

The net annual carbon balance was quantified at -1.4 ± 0.46 Mg C-CO₂ ha⁻¹ with the chamber method and -1.8 ± 0.17 Mg C-CO₂ ha⁻¹ by the Eddy Covariance (EC), indicating that the studied agro-silvo-pastoral system functions as a net carbon sink. These findings corroborate the system potential role in mitigating greenhouse gas emissions, consistent with previous studies reporting vertical CO₂ flux balances in semi-arid ecosystems (Rahimi et al., 2021; Tagesson et al., 2015; Agbohessou et al., 2023, Wieckowski et al., 2024).

The estimated net C exchange balance is close to the reported mean for Sahelian ecosystems (-1.6 ± 0.5 Mg C-CO₂ ha⁻¹; Tagesson et al., 2016b). The EC-based net C exchange balance (-1.8 ± 0.17 Mg C-CO₂ ha⁻¹) is also similar to the value of -1.9 ± 0.4 Mg C-CO₂ ha⁻¹ reported for semi-arid savannas of northeastern Benin, despite higher annual rainfall (1495 mm; Ago et al., 2016b). Furthermore, our EC estimate is close to the average net C exchange reported for West African terrestrial ecosystems (-2.0 ± 1.5 Mg C-CO₂ ha⁻¹; Ago et al., 2016a).

However, estimates from Tagesson et al. (2015) (-2.7 ± 0.07 Mg C-CO₂ ha⁻¹) for a semi-arid savannah in Dahra, Senegal, located between the 300 and 400 mm isohyets, were comparatively higher. This is potentially attributable to specific characteristics of that specific savannah site, such as herbaceous vegetation cover during the rainy season, the presence of evergreen trees, and land management practices linked to pastoral livestock activities (Tagesson et al., 2016a).

The net annual C balance estimates presented in this study are, in fact, representing vertical fluxes only, given that they exclude organic matter (OM) imports and, more critically, exports, introducing uncertainties. Notably, the export of crop residues and direct inputs from animal excreta – particularly significant in “bush fields” during the dry season – were not accounted for. In our case of “bush field”, crop residues are exported to feed livestock, while livestock faeces are collected for use as fuel or manure in “home fields”. Such practices may lead to a significant soil organic carbon stocks depletion (Malou et al, 2021), potentially diminishing the net C budget (-1.4 ± 0.46 Mg C-CO₂ ha⁻¹) over time and shifting the system closer to carbon neutrality (Assouma et al., 2019).

These results should be contextualized within the broader framework of climate change and semi-arid ecosystem management. Although agro-silvo-pastoral systems can function as apparent annual carbon sinks, they remain highly sensitive to interannual rainfall variability and escalating anthropogenic pressures. Sustainable management practices, particularly regarding C inputs/outputs from the system regarding crop harvest, residues exportation, and cattle free manuring, must be taken into account to confirm the capacity of the system to act as effective a carbon sink.

This study benefited from the inverse phenology of F. albida, allowing for direct comparison between chamber-based GPP (GPPch.stand) and ecosystem-level GPP (GPP.EC) during the leafless period of the trees. However, the system spatial heterogeneity – common in agroforestry – posed challenges for accurately partitioning CO₂ fluxes among trees, crops, and soil. A key limitation was the development of weeds during the late rainy season, which complicated the attribution of fluxes, particularly during the transitional period. Additionally, while GPPch was successfully standardised by LAI for upscaling, this was not feasible for respiration. Respiration integrates both autotrophic and heterotrophic components, which respond to different drivers and are not directly linked to LAI, limiting the precision of upscaled Rch.

Future improvements should aim to separately quantify respiration sources – tree roots, crops, and microbial (heterotrophic) respiration – and account explicitly for the weed layer, to refine flux partitioning in such complex agroforestry systems.

Furthermore, the present study constitutes only an intermediate step delivering a first integrated estimate of the main vertical CO₂ exchanges (photosynthesis, respiration, and net ecosystem exchange) as a base for a forthcoming paper that will present a more comprehensive carbon budget of the ecosystem. Establishing such a carbon budget would require substantial additional data acquisition and poses considerable methodological challenges. In particular, quantifying carbon inputs/outputs associated with free-ranging livestock grazing would be difficult to achieve with acceptable accuracy. It must also be recognised that the system is in a dynamic, non-steady state, characterised by marked inter-annual variability as well as periods of carbon storage and release, which are difficult to constrain empirically except through modeling.