Congo Basin Hydrology, Climate, and Biogeochemistry. Группа авторов

4.2. DESCRIPTION OF THE MODEL, EXPERIMENTAL PROTOCOL, DATA, AND METHODOLOGY

      4.2.1. Model Description

The RegCM regional climate model is a model that was developed by the group of atmospheric physicists and climatologists of the Abdus Salam International Centre for Theoretical Physics (ICTP; Giorgi et al., 1993). Since the release of RegCM3, the model has undergone substantial development in terms of both software code and physical representations, leading to the development of a fourth version of the model (RegCM4), which was published by ICTP in June 2010 as a prototype (RegCM4.0) and in May 2011 as the first full version (RegCM4.1). Since the RegCM4.5 version, the model has used a non‐hydrostatic dynamic core, which makes it possible to obtain small horizontal resolutions of the order of a few kilometers. Radiation parameterization was adapted from the NCAR CCM3 radiation transfer scheme (Kiehl et al., 1996). The non‐local boundary layer developed by Holtslag and Boville (1993) is used for the representation of the planetary boundary layer. Large‐scale precipitation is calculated by the SUBEX scheme (Pal et al., 2000) and the biosphere–atmosphere transfer system (BATS) of Dickinson et al. (1993). The BATS scheme takes into account the transfer of energy, mass, and momentum between the atmosphere and the biosphere. The model includes options for an ocean flow parameterization scheme, interactive aerosols, microphysics, lake models, etc. The RegCM4.6 version provides five different convective schemes (Giorgi et al., 2012): the modified Kuo scheme (Anthes et al., 1987); the Tiedtke scheme (Tiedtke, 1996); the Emanuel scheme (Emanuel, 1991); the Grell scheme (Grell, 1993); and the Kain‐Fritsch scheme (Kain & Fritsch, 1993), with the possibility of combining different oceanic and continental schemes (called mixed convective schemes).

       The “Slab‐Ocean” Parameterization

      By default, version 4.6 of RegCM includes the flexible mixed layer modeling system of the SOM developed by the Laboratory of Geophysical Fluid Dynamics (LDFG; Giorgi et al., 2012). In our study, SOM is a simple thermodynamic oceanic layer having a constant thickness of 50 m. The ocean surface warms or cools in response to surface heat exchanges with the atmosphere. SOM interacts with the atmosphere to calculate SST and ice parameters by forcing the model with RegCM fluxes. The fluxes that drive each iteration are provided to the SOM for the SST update, which is then transmitted to RegCM for the next iteration. But the lack of ocean dynamics (convection, advection, and fusion motions) can lead the model to make poor simulations. To solve the problem, a set of heat flow adjustments is specified by adding a term commonly referred to as “q‐flux.” The q‐flux is being added to the SOM at each time step to provide a realistic SST distribution by the model. The technical documentation of the SOM can be consulted on the following website: http://www.gfdl.noaa.gov/fms‐slab‐ocean‐model‐technical‐documentation.

      4.2.2. Data and Methodology

       Data

      Evaluation of the performance of regional climate models is based on the unidirectional nesting technique, which requires prior reduction of errors that can be inherited from lateral forcing conditions (Giorgi & Mearns, 1999), i.e., GCMs (global circulation models). For this reason, conditions with so‐called quasi‐perfect limits and approximately similar observations are used. The following data were used to run the model: ERA‐Interim 1.5 gridded data set (Uppala et al., 2005), with a temporal resolution of 6 h (0000, 0600, 1200, and 1200 UTC). Here, the variables used are air temperature, geopotential height, relative humidity, and a horizontal wind component. The SST is taken from the weekly optimal interpolation SST from the National Administration of the model grid (Reynolds et al., 2002). For the global terrain and land use, we have used the 2‐min resolution global land cover characteristics (GLCC; Loveland et al., 2000) and GTOPO topography data. These data were initialized on 1 January 2000, with parameters such as air temperature, geopotential height, relative humidity, and wind components.

      One of the main problems in assessing the performance of RCMs in Central Africa is the lack of high‐quality observation databases at appropriate spatial and temporal resolution. The use of different sources of observational data (in‐situ and satellite), and reanalyses of rainfall, temperature, and wind make it possible to take into account the uncertainties associated with them in Africa (Nikulin et al., 2012). To facilitate the intercomparison between observations and models, we interpolated the data on the model grid. Simulations of precipitation, temperature, and wind are compared with the monthly climatology of the data from: (i) Africa Rainfall Climatology version 2.0 (ARC2; resolution 0.1° × 0.1°; Novella & Thiaw, 2013); (ii) Global Precipitation Climatology Project (GPCP; resolution 0.5° × 0.5°; Huffman et al., 1997); (iii) Climatic Research Unit (CRU; resolution 0.5° × 0.5°; Harris et al., 2013); and (iv) the fifth‐generation European Centre for Medium‐Range Weather Forecasts (ECMWF), reanalysis 0.75° × 0.75° (ERA5). Compared to its predecessor ERA‐Interim, ERA5 offers a higher spatiotemporal resolution and an improvement of the atmospheric model and data assimilation processes (Hersbach et al., 2020).

       Methodology

The range of experience extends from 17.5°S to 17.5°N and 30°W to 80°E. This domain is large enough (Figure 4.1) so as to take into account the climate continuum of Central Africa, which depends heavily on the Indian Ocean and weakly on the Atlantic Ocean. This area is chosen to include climatic factors at the local level, but also to account for the diversity of African weather patterns. The domain to be modeled is Central Africa, which extends between longitudes 5°E and 35°E and latitudes 15°S and 15°N. A finite‐difference horizontal discretization is done with square meshes of 40 km (i.e. 0.36° × 0.36°) on each side. At this resolution, RegCM can better represent orographic forcing than reanalyses, for example. The experiments were carried out over seven years (from 1 January 2000 to 31 December 2006), with one year of “spin‐up” excluded in the study period. This study period is chosen because it includes two years that did not experience extreme rainfall anomalies, namely the years 2004 and 2006. Therefore, the period from 2000 to 2006 presents a wide range of rainfall variability over the study area (Mbienda et al., 2016). We have done experiments with the Grell convective scheme (Giorgi et al., 1993), which was identified by Mbienda et al. (2016) as a suitable RegCM convective scheme for simulations in Central Africa. To examine the influence of slab‐ocean parameterization, we carried out two different experiments with RegCM: The first experiment without slab‐ocean, which we call GFC_CTR, is designed for the climatology of the different parameters of the model and activates the SOM surface flux that forces the surface boundaries of RegCM with SST without modeling the ocean–atmosphere feedbacks. The second slab‐ocean experiment, which we call GFC_SLAB, is designed to couple RegCM with the SOM and is identical to the first experiment except for the activation of the slab‐ocean. It provides mutual interaction of the atmosphere (RegCM) and the ocean (SOM). It uses SST and the climatology created in the first experiment to model the ocean–atmosphere interaction by modulating SST. In this experiment, different methods are adapted to the model to reduce errors due to the complete absence of ocean dynamics in the SOM. The model then uses the q‐flux adjustments in SOM to improve the representation of the seasonal cycle of SST. This q‐flux is obtained by performing a calibration experiment (restoration cycle), during which the observed SOM prognostic SST is established over a five‐day interval, then archived and saved in an average monthly climatology for the period 2000–2006. Different statistical approaches have been adopted to estimate the skill of the model according to the different experiments in the different homogeneous sub‐regions and the Central African domain as a whole. We used statistical tests such as (i) the bias that measures the difference between the observed data (chosen as a reference) and the simulated data; (ii) the Taylor Diagram (Taylor, 2001), which provides a fairly simple way to summarize the similarities between the observed and simulated data – this diagram illustrates statistics such as the Pearson correlation coefficient