1. Introduction
The formulation (including physical parameterization and tuning) of a climate model plays a paramount role in the representation of regional processes, especially those underpinning the rainfall system [
1,
2]. For instance, incorporating additional components of the Earth system, such as coupling with an oceanic model, can result in a significantly enhanced credibility of climate projections in regions where air-sea interaction is strong [
3,
4,
5,
6]; introducing aerosol interactions substantially influences regional patterns' change of both temperature and precipitation [
7,
8]; neglecting the physiological response of plants to the increase in greenhouse gases in the atmosphere results in a substantial underestimation of extreme temperature increases across Europe [
9]; mixing in stable boundary layers, in particular, the momentum mixing which has a strong impact on the surface heat significantly modifies surface-atmosphere interactions [
2,
10].
To properly assess how formulations designed and validated over the historical climatology can impact the future climate, it is essential to evaluate how the physical processes most relevant for a region's climate are simulated, and how they are modified under future warming [
11]. Such process-based evaluations and analyses are becoming more and more common as they can provide insight into understanding the relevant processes driving e.g. the projected changes and, hence, provide more confidence in the projections [
12,
13,
14,
15,
16,
17]. For instance, [
14,
15] investigated processes driving changes in the Congo basin in December-February (one of the dry seasons) and September-November (one of the wet seasons) respectively, using a set of coarse resolution global climate models (GCMs) participating in the phase 5 of the Coupled Model Intercomparison Project (CMIP5, [
18]). They found that in DJF, the region is projected to become wetter due to increased convection and changes in atmospheric patterns. In SON, the west experiences wetter conditions in the north and drier conditions in the south due to factors like Atlantic sea surface temperatures and moisture transport. In the east, there's an increase in precipitation over the northeast region, linked to changes in the African easterly jet and moisture convergence. However, such analysis of projections in the Congo basin is still not widely used as a common approach, especially for RCMs.
Another important aspect of process-based evaluation is the assessment of the plausibility of expected changes. The model projected change (e.g projected drying) is not necessarily linked to the model's performances on the historical climate (e.g. dry bias)
[12,19]. The sometimes-non-respect of the 'wet-gets-wetter' mechanism, which consists of an enhanced moisture convergence, in response to projected climate change, in regions which already feature convergence is illustrative to this end. For instance, using a set of Coordinated Regional Climate Downscaling EXperiment (CORDEX) regional climate models (RCMs) over West Africa,
Ref [20] found a decrease in the dynamic component of moisture budget over areas of intense rainfall, matching with a weaker moisture convergence.
The seasonality of the Congo basin’s (10°-35°E; 10°S-10°N) precipitation is strongly correlated with that of the moisture convergence/divergence [
21,
22,
23]. CMIP models simulating dry (wet) biases also feature strong divergent moisture from (convergent moisture into) the region [
16,
24]. This region is characterised by two wet seasons from March to May (MAM) and from September to November (SON), and two dry seasons from December to February (DJF) and from June to August (JJA) (see
Figure 1a), matching with the time of peaks of moisture convergence and divergence respectively. Some work [
23,
25,
26,
27,
28] demonstrated this moisture originating from both Atlantic and Indian Oceans, and from other continental sources (dynamical contribution), and from the local evaporation through the recycling process (thermodynamic contribution). Ref [
29] showed that the increased atmospheric column moisture due to warming responds to the thermodynamic contribution. This corresponds to the situation of decreasing evaporation minus precipitation; the dynamic contribution involves changes in atmospheric circulation on regional and local changes in precipitation and its extremes.
Ref. [
2] evaluated a modified version of the Rossby Centre Atmospheric RCM RCA4 over the CORDEX-Africa domain with a reduced turbulent mixing in the stable boundary layer. Such change reduced turbulent mixing in the convective planetary boundary layer (PBL) and improved simulated local wind shear [
10]. They found that this modified version (hereafter RCA4-v4) improves the annual cycle of precipitation over Central Africa, and reduces the rate of dry biases compared to the standard version used in CORDEX (hereafter RCA4-v1; see [
16]): in fact, RCA4-v1 generally features dry biases over the Congo basin, related to a stronger mid-tropospheric moisture divergence, in turn, associated with a misrepresentation of African Easterly Jets (AEJs; [
16]). In contrast, RCA4-v4 considerably attenuates the dry biases shown by RCA4-v1 and improves the annual cycle of precipitation, in response to a sequence of offsetting effects in a chain of processes [
2,
17]: in fact, RCA4-v4 simulates a stronger low-level moisture convergence, associated with a strengthening of the Congo basin low-level cell [
30] and thus Low-Level westerlies (LLWs). In turn, the moisture surplus from LLWs overfeeds mesoscale convective systems via AEJs, thereby increasing the rainfall amount. Both versions of RCA4 are found plausible to model the Congo basin’s climate system, where the word ‘plausible’ is understood as the ability of an RCM to model interlinkages between precipitation and related physical processes and drivers, although biases may still persist.
In the present study, we focus on the process-based evaluation of the simulated future (2071-2100) Congo basin climate system projections from the two configurations of RCA4 (following
[2,17]). The objectives of this study are dual: first, examine changes in the basin’s precipitation associated with some drivers over the late 21
st century. Second, understand the effect of the RCM formulation on projected precipitation changes. The paper is structured as follows: details of the model are briefly described in the next section (
Section 2);
Section 3 summarises changes in precipitation and examines mechanisms underlying precipitation changes; the effects of the model formulation are highlighted in section 4, and the paper closes with a summary and result discussions in section 5.
2. Model data and Method
Data used in this work are derived from two different configurations (see details in
Table 1) of the latest version of the Rossby Centre Atmospheric (RCA) model RCA4 (namely, RCA4-v1 and RCA4-v4), which are used to dynamically downscale two GCMs (see details in
Table 1) participating in CMIP5. Details of RCA4 configurations are extensively described in [
2,
31] and references therein, and can be summarised as follows: broadly, the RCA4 simulation over the CORDEX-Africa domain, with approximately 50 km horizontal resolution, is based on the HIRLAM numerical weather prediction model. This model incorporates enhanced physical and dynamical parameterizations. The land-surface scheme utilizes a quadrilled approach with one to three key tiles based on land-use information recommendations. The convection scheme, following Kain–Fritsch, assumes non-precipitating shallow convection. The radiation scheme is derived from the modified HIRAM’s radiation scheme, designed to consider carbon dioxide absorption and improve the treatment of the water vapor cycle. The simulation employs vegetation-dependent land-surface parameters. Additionally, a six-order linear horizontal diffusion, associated with a two-time-level, semi-Lagrangian, and semi-implicit scheme, is applied to the prognostic variables. Historical simulations cover the period from 1950 to 2005 (here, 1976-2005 is selected as the reference period), based on the observed natural and anthropogenic greenhouse gas (GHG) concentrations. Future projections (2006-2100) are driven by the Representative Concentration Pathway (RCP) 8.5 [
33]. As in e.g. [
34], our study focuses on projections solely at the end of the century (2071–2100) for the high-emission scenario RCP8.5. This choice is made because, in the near-future timeframe, or under a low-emission scenario, internal variability can mask the emergence of the anthropogenic signal in precipitation changes, particularly at the regional scale [
11]. The two RCA4 configurations have been thoroughly assessed in the recent past (1981-2005) over Central Africa, including the evaluation of relevant physical processes [
2,
16,
17,
35,
36]. Rather, we focus on the analysis of the effect of model configuration on the changes in the precipitation climatology and in some interlinked drivers for the late 21
st century (2071-2100) relative to the reference period (1976-2005).
In our analysis, we focus on the regional moisture convergence: throughout the tropospheric column, the moisture transport can be separated into zonal (
Qz in Kg m-2s-1) and meridional (
Qm in Kg m-2s-1) components as follows:
where
u is the zonal wind and
v the meridional wind (in m/s) components,
q is the specific humidity (g/Kg) and
g (in
N/Kg) is the gravitational acceleration.
q, u and
v are functions of time (
t), pressure levels (
z), latitudes (
y), and longitudes (
x). Transient moisture (in
Kg m-2s-1) across each boundary of the Congo Basin region is then obtained using the following equations:
The net zonal (
Zqconv, in Kg m-2s-1) and the net meridional (
Mqconv, in Kg m-2s-1) regional moisture convergence are obtained as follows:
Future changes (
Δ) in the moisture convergence in the zonal (
ΔZqconv, in Kg m-2s-1) and in the meridional (
ΔMqconv, in Kg m-2s-1) directions are estimated using the differences:
The total regional moisture convergence change (
ΔTqconv, in Kg m-2s-1) is obtained by adding changes in the zonal and meridional components as follows:
Besides the analysis of moisture convergence, it is interesting to investigate the roles of the thermodynamic (
ΔTH) and dynamic (
ΔDY) components of the moisture budget
[39]. Dynamic and thermodynamic changes affecting the moisture budget in a given region can be approximated using the linearized equation as follows
[20,29,39,40]:
where
ρ is the water density,
ΔEmP represents changes in the atmospheric moisture budget,
E is evaporation, P precipitation.
ΔTH is the thermodynamic term (moisture budget changes related to changes in the specific humidity), and
ΔDY the dynamic term (moisture budget changes associated with changes in the atmospheric circulation),
ΔTE is the synoptic term and
ΔRes the residual term.
ΔTH and
ΔDY are expressed respectively as follows:
and
ΔEmP is obtained as follows:
in Equations (14) and (15),
Δq = q(future) - q(historic) represents changes in the specific humidity, and
ΔV = V(future) - V(historic) changes in the wind,
(in
m/s) is the mean climatology value of the horizontal wind (u and v),
(in
g/Kg) is the mean climatology value of the specific humidity,
pbot is the surface pressure and
ptop is the pressure of the top-level (300 hPa),
ΔE = E(future) - E(historic) represents changes in the total evaporation, and
ΔP = P(future) - P(historic) means changes in precipitation. Changes in atmospheric water vapour at time scales longer than synoptic (
ΔTE) can be approximated to zero
[41]; also, the contribution of the residual term (
ΔRes) is generally small and is approximated to zero as in e.g.
[20,42].
Observed precipitation (rainband) and vertical profile of moisture convergence are obtained from the Global Precipitation Climatology Centre (GPCC-v8) [43] and the European Centre for Medium-Range Weather Forecast (ECMWF) ERA-Interim reanalysis dataset [44], respectively. The availability of RCA4 outputs over the entire tropospheric column (the only CORDEX RCM providing atmospheric variables u,v,q at all pressure levels) offers the possibility of quantifying each term and accounting for the contribution of all pressure levels.
4. Impacts of the model formulation on the simulated future climate
One of the main results of this study is the response of precipitation (and its change) to both the model formulation and large-scale boundary conditions. In fact, RCA4-EC-EARTH-v4 which is, in the present climate, wetter than RCA4-EC-EARTH-v1 in all seasons
[17], projects a wetter future (compared to RCA4-EC-EARTH-v1) in the dry seasons (DJF and JJA), but a drier one in most parts of the basin in the wet seasons (MAM and SON) (
Figure 1b); conversely, RCA4-MIROC5-v4, which is also historically wetter than RCA4-MIROC5-v1
[17], tends to project a drier future in all seasons (
Figure 1b). These differences highlight the role of the large-scale forcing through the boundary conditions (driving GCMs). Indeed, when driven by quasi-perfect boundary conditions from ERA-Interim reanalysis,
Ref [2] found that model formulation had the primary control on precipitation climatology in Africa. However,
Ref [48] found that the RCM COSMO-CLM performances on the historical climate depended more on the driving GCM than horizontal resolution, model version, and configuration.
For future projections, the increase in anthropogenic forcings would induce modifications in the physical processes driving the regional precipitation system climatology [49]. In addition, imperfect boundary conditions from GCMs might play an important role in RCM results [50]. For instance, in the zonal tropospheric column, RCA4-EC-EARTH-v4, which features higher moisture divergence in MAM under the historical climate than RCA4-EC-EARTH-v1 (not shown), also does so in the future; The same run that exhibited the most significant low-level moisture convergence in SON in the past climate will display the weakest convergence in the future (see column 1 in Fig 2b). RCA4-MIROC5-v4 which displayed a stronger historical low-level (mid-layers) moisture convergence (divergence) in both MAM and SON seasons (not shown) is projected to strengthen moisture divergence all over at the late 21st century. All RCA4-v4 experiments simulate a stronger ΔDY compared to RCA4-v1 except in SON whereas the two model versions experience a quasi-similar ΔTH in all seasons (Figs 4b and 5b respectively).
5. Summary and Discussion
This study analysed changes in the Congo basin precipitation system using two configurations of the Rossby Centre regional Atmospheric model RCA4: the standard version employed in CORDEX (RCA4-v1) and a modified version in which the turbulent mixing was reduced (RCA4-v4). The analysis focused on the future changes in the spatial distribution of seasonal mean precipitation, in the vertical structure of moisture convergence and the dynamic and thermodynamic contributions to moisture changes. The analysis of the impact of the model formulations on the projections in the late 21st century (2071-2100 vs 1976-2005) was also performed.
The results show that the wetter seasons, MAM and SON, are projected to become wetter, with a more pronounced north-south dipole, particularly evident in SON; in drier seasons DJF and JJA, moderated decreased precipitation is projected. All runs are consistent with the fact that the response of the hydrological cycle to the warming climate in the Congo basin will be associated with modifications of atmospheric moisture convergence and divergence, further related to dynamic and thermodynamic effects: in fact, changes in precipitation are associated with patterns of changes in the vertical profile of the atmospheric moisture convergence over the region. Moisture convergence (divergence) is overall projected to strengthen in wetter (drier) seasons. Seasonal ΔEmP are strongly correlated with seasonal rainfall changes, consistent with patterns of ΔTH and ΔDY. Most experiments indicate that while both ΔDY and ΔTH prevail during the wet seasons, only ΔDY dominates during the dry seasons.
Several physical underlying mechanisms are involved in the change in tropical rainfall under projected global warming, including both “wet-gets-wetter”
[51] and “warmer-gets-wetter”
[52,53] mechanisms. For instance,
Ref [54] showed that in tropical regions, the annual pattern of rainfall is associated with the “warmer-gets-wetter” mechanism, whereas seasonal precipitation anomalies are closer to the “wet-gets-wetter” mechanism. In the present study, we found a trend towards wetter (drier) conditions during the wet (dry) seasons, correspondingly associated with an increased moisture convergence (divergence), which is in line with the "wet-gets-wetter" mechanism. Furthermore, this mechanism involves changes in the atmospheric water vapour, dynamically and thermodynamically induced.
Numerous studies investigated the dynamic and thermodynamic contributions to precipitation changes in tropical regions, and most of them agreed on the dominance of dynamic effects under global warming
[20,42,47], in accordance with the findings of this study. Using a set of five RCMs over the northeast Asia and Korea,
Ref [29] found that while
ΔTH contributes to the increased precipitation over the central part of the Korean peninsula by increasing the moisture convergence,
ΔDY contributes instead to reducing the rate of precipitation by inducing moisture divergence. The present paper shows that
ΔDY contributes by reducing mean precipitation in the Congo basin in dry seasons (DJF and JJA). The two terms co-contribute to an overall increased rainfall in wet seasons (MAM and SON), with the share of
ΔTH slightly higher than that of
ΔDY, thus suggesting the strongest contribution of radiative forcing to precipitation changes. Although their analyses were focused only over a single pressure level (850 hPa),
Ref [20] likewise showed that RCMs projecting a decreased rainfall over the Guinea Gulf simulated a weakening
ΔDY in spring and early summer; both
ΔDY and
ΔTH co-contributing to wetting in spring. The plausibility of the above-highlighted drying and wetting processes in the Congo basin is sustained by previous findings by
Ref [55]: they showed that decreased precipitation over the basin in drier seasons is associated with a weakening large-scale moisture convergence (dynamical causality). Also, they reported a strengthening moisture convergence in wetter seasons but dominated by regional processes, and support the idea of a possible influence of soil moisture, which strongly modulates precipitation through the recycling process (thermodynamic causality).
While in the present-day climate RCA4-v4 runs are wetter than those of RCA4-v1, in the future the situation reserves, depending on the boundary conditions: in fact, the reversing happens for RCA4-v4 in all seasons when driven by MIROC5, but only in some when driven by EC-EARTH. This highlights the influence of the large-scale forcing prescribed by the driving GCMs. In addition, we note that the changes in future precipitation could be explained on the one hand, by the intensification of existing, underlying physical processes described above. On the other hand, they might be related to the change in the nature and origin of these processes, or even by the appearance of new processes (induced by enhanced warming), as also noted in [11].
Our work highlights the necessity for understanding how the increasing GHG concentration influences and, potentially, modifies the mechanisms driving the regional climate system, and how climate models formulations should take this aspect into account. Moreover, by estimating separately the dynamic and thermodynamic contributions to total changes in the moisture budget, we hope to foster the understanding of the impacts of large-scale versus local/regional physical mechanisms underpinning future changes as well as assess their plausibility. This may be helpful to policymakers and other stakeholders in the preparation of effective adaptation plans and mitigation strategies.