<scp>AFRICLIM</scp> : high‐resolution climate projections for ecological applications in <scp>A</scp> frica

Half of the African population, and most priority sites for conservation, are concentrated in mountain and coastal regions (Fig. 1). In these places, climatic gradients are steep and feedbacks between land, water and atmosphere are much more localized than the pixel resolutions of general circulation models (GCMs). Through the CORDEX initiative (Jones, Giorgi & Asrar, 2011), outputs from regional climate models (RCMs) have become available for Africa. Nested within GCMs, regional models simulate climate at finer spatial and temporal resolutions (Fig. 1). Yet at ~50 km, they remain too coarse-grained for many applications in ecology (Platts et al., 2013). Here, we use a range of observational baselines to empirically downscale RCM outputs to resolutions amenable to ecological applications at local scales (up to 1 km). Results for the middle and late 21st century are available online https://webfiles.york.ac.uk/KITE/AfriClim/. RCM outputs for the period 1950–2100 were provided by the Swedish Meteorological and Hydrological Institute and the Canadian Centre for Climate Modelling and Analysis, at a resolution of ~50 km (0.44° × 0.44°). The Swedish model (SMHI-RCA4) was driven by boundary conditions from eight GCMs (Fig. 1a) and the Canadian model (CCCma-CanRCM4) by CanESM2. Future climates were projected under two IPCC-AR5 representative concentration pathways: RCP4.5 and RCP8.5, which project global temperature anomalies of 2.4°C and 4.9°C above pre-industrial levels by 2100 (Rogelj, Meinshausen & Knutti, 2012), with atmospheric CO2 equivalents of 650 and 1370 ppm by 2100, respectively (Moss et al., 2010). We used change-factor downscaling to recover spatial variation at local scales and to correct for differences between observed and simulated baseline climates (Tabor & Williams, 2010). Due to uncertainty in observational baselines, we imposed RCM change-factors (future anomalies) onto four different data sets for rainfall and two data sets for temperature: CRU CL 2.0 (New et al., 2002), WorldClim v1.4 (Hijmans et al., 2005), TAMSAT TARCAT rainfall v2.0 (Maidment et al., 2014); and CHIRPS rainfall v1.8 (Funk et al., 2014). These grids, and thus downscaled projections, vary in resolution from 30″ (~1 km) to 10′ (~19 km). To calculate change-factors, we first averaged RCM output for monthly 2-m air temperature (mean, minimum and maximum) and monthly rainfall over the period 1961–1990, matching to the time spans of CRU and WorldClim (Fig. 1a). Similarly, we calculated monthly rainfall around the year 2000 (1986–2015) to match the midpoints of TAMSAT and CHIRPS. Future anomalies were obtained by subtracting these simulated baselines from 30-year averages around the 2050s (2041–2070) and 2080s (2071–2100). Anomalies were spline-interpolated to higher resolutions (Mitasova & Mitas, 1993) and, for temperature, added to observational baselines (B). Rainfall anomalies (Δ) were imposed as absolute changes relative to the baselines: B × |1 + Δ/(B + 1)| (Ramirez-Villegas & Jarvis, 2010). We provide downscaled grids for each GCM-RCM-baseline triplet separately and, for SMHI-RCA4, multimodel ensembles over eight GCMs. In addition to monthly grids, we provide 21 summary variables for applications in ecology (Table 1). Analyses were carried out using R (Pierce, 2011; R Core Team, 2012) and GRASS-GIS (GRASS Development Team, 2012). By late century, sub-Saharan Africa is projected a mean annual temperature of 26.4–27.6°C (RCP4.5) or 27.9–29.8°C (RCP8.5), depending on the model (Table 1). Rainfall is projected to increase in western and eastern parts of the continent, coupled with increased seasonality (Fig. 2; Tables S1–S5). Changes in rainfall are, on average, lower at higher latitudes, with a slight drying trend depending on the model (Fig. 2). Ensemble means project the Mediterranean Basin, south-east Africa, eastern Madagascar and the Ethiopian Highlands to be at risk from prolonged seasonal aridity (consecutive months of rainfall≪PET), while the Horn of Africa, Gabon and coastal Angola are projected shorter periods of aridity. Within these regions, downscaled projections reveal considerable variation, with spatially complex climates subject to multiple extrema in RCM anomalies at sub-GCM scales. We note that while empirical downscaling of RCMs (cf. GCMs) reduces uncertainty at the mesoscale, the assumption of temporal stasis in local climatic variation, as inferred from observational baselines, remains a source of error (Tabor & Williams, 2010). Further, the accuracy of baseline climatologies is limited by the distribution of meteorological stations in Africa, which particularly for rainfall remain sparse. We mitigate this issue by including two satellite-derived baselines for rainfall, in addition to the interpolated climatologies. At larger scales, assumptions underlying RCPs are intentionally diverse (Moss et al., 2010) and GCM-RCM ranges are sometimes high (Table 1; Fig. 2). Driven with ERA-Interim reanalysis data, RCMs are reasonably skilful in simulating climatic variability over Africa, and biases are effectively reduced by the ensemble mean (Nikulin et al., 2012; Endris et al., 2013; Gbobaniyi et al., 2014). To project future climate, RCMs are driven by GCMs. Comparing GCM-RCM estimates with observational data over the 30-year baseline, there is good agreement between large-scale means, but models underestimate temperatures during cooler months, particularly in the north and west, and so overestimate seasonality (Tables 1 and S1–S5). In southern Africa, models overestimate rainfall during the wettest months while underestimating aridity during the dry season. Such differences are superficially addressed by change-factor downscaling (bias correction), but nonetheless highlight weaknesses in model skill over Africa and/or uncertainties in the validation data (Wilby et al., 2004; Brands et al., 2013). Climate projections are in immediate demand by scientists, governments and nongovernmental organizations. High-resolution projections are available globally (e.g., http://www.worldclim.org/cmip5) but are empirically derived directly from GCMs, with no dynamical downscaling. AFRICLIM is an important step forward in this respect: the archives span eight GCMs downscaled using two RCMs and four observational baselines, under two emissions pathways and at multiple high-spatial resolutions. We encourage users to interpret the data critically, however, with due consideration of the above uncertainties, particularly with respect to model skill in the region of interest (see e.g., Nikulin et al., 2012; Crétat, Vizy & Cook, 2014). Funded by the Ministry for Foreign Affairs of Finland (http://chiesa.icipe.org/). We thank IT Services at the University of York for hosting the data. In addition to citing this article, please acknowledge the regional centre and baseline climatology in applications of the data. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.