Supplement Materials for “Simulated sensitivity of African terrestrial ecosystem photosynthesis to rainfall frequency, intensity, and rainy season length”

Kaiyu Guan1*, Stephen P. Good2, Kelly K. Caylor3, David Medvigy4, Ming Pan5, Eric F. Wood5, Hisashi Sato6, Michela Biasutti7, Min Chen8, Anders Ahlström9

and Xiangtao Xu10

1 Department of Natural Resources and Environmental Sciences and National Center for Supercomputing Applications, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA

2 Department of Biological and Ecological Engineering, Oregon State University, Corvallis, OR 97331, USA

3 Department of Geography, Bren School of Environmental Science & Management, University of California - Santa Barbara, Santa Barbara, CA 93106, USA

4 Department of Biological Sciences, University of Notre Dame, Notre Dame, IN 46556, USA

5 Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ 08544, USA

6 Institute of Arctic Climate and Environment Research, Japan Agency for Marine-Earth Science and Technology (JAMSTEC), 3173-25 Showamachi, Kanazawa-ku, Yokohama, 236-0001, JAPAN

7 Lamont-Doherty Earth Observatory of Columbia University, 61 Route 9 W, Palisades, NY, 10964-8000, USA.

8 Joint Global Change Research Institute, Pacific Northwest National Laboratory, College Park, MD, USA

9 Department of Environmental & Earth System Science, Stanford University, Stanford, CA 94025, USA

10 Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, MA 02138, USA

*Corresponding author: Kaiyu Guan ()

1.Details of the synthetic weather generator

The two steps of the synthetic weather generator are described below:

Step 1: Based on rainfall scenarios from different experiments, we first model the daily rainfall following the marked Poisson process described before. In particular, for a specific year, we first stochastically generate the wet season length by sampling from a beta distribution, and the dry season length is determined accordingly. Then we generate the daily rainfall for wet and dry season respectively.

Step 2: Based on the simulated daily rainfall record in Step 1, we conditionally sampled temperature, wind, and humidity from the Global Meteorological Forcing Dataset (GMFD, Sheffield et al., 2006), as well as cloud fraction and soil temperature from the Climate Forecast System Reanalysis (CFSR) from National Centers for Environmental Prediction (NCEP) (Saha et al., 2010). For each day, a sample is randomly drawn from a pool that covers all the historical record within a 21 day time window centered at the sampling day. From the sampling pool, we find the day such that the historical rainfall amount of the chosen day is within (100-30)% to (100+30)% of the simulated daily rainfall amount. We then draw all other environmental variables on that sampled day to the new climate forcing. If we can find a sample from the pool, this sampling is called “successful”. When there is more than one suitable sample, we randomly select one. When there is no suitable sample, we random select one day within the pool. The mean “successful” rate for all the ensembles and all the experiments across Africa is 83%.

The GMFD data (Sheffield et al., 2006) blends reanalysis data with observations and disaggregates in time and space, and is available from 1948 to 2008, with 1.0-degree spatial resolution and daily temporal resolution. The CFSR data (Saha et al., 2010) provides cloud fraction and simulated soil temperature from three soil layers for the SEIB-DGVM. The CFSR version that we used is from 1979 to 2010, with original 0.3 degree spatial resolution and 6-hourly temporal resolution aggregate to 1.0 degree and daily.

2. Details of the experiment design

All the experiments were designed as follows (Table 1 and Figure S4):

Sclimatology: SEIB run forced by observed climate forcing;

Scontrol: Model run forced by the synthetic forcing with observed rainfall characteristics from TRMM;

Exp 1. (Perturbation of rainfall intensity, Sα) Model run forced by the synthetic forcing, varying only rainfall intensity α and keeping the other two rainfall characteristics fixed at the current climatology derived from TRMM;

Exp 2. (Perturbation of rainfall frequency, Sλ) Model run forced by the synthetic forcing, varying only rainfall frequency λ and keeping the other forcing variables fixed at the current climatology;

Exp 3. (Perturbation of wet season length, STw) Model run forced by the synthetic forcing, varying only the wet season length and keeping the other forcing variables fixed at the current climatology.

Comparisons between Sα/ Sλ / STw and Scontrol quantify the GPP sensitivity to the three rainfall characteristics. All the scenarios have six ensemble runs differentiated in their synthetic forcings. This was done to account for the stochasticity of the synthetic weather generator.

We assume the slope of the changes remains constant within such small perturbations. To check how valid this assumption is, we rerun the same experiments with perturbations of ±20%. We find that the spatial patterns of differential GPP sensitivity hold, though the linearity assumption is not valid as expected with larger deviations away from the control, esp. for tropical forest regions (Figure S5). Changes of MAP by -10% or +10% away from the current states cause the contraction or expansion of tropical evergreen forest by 1° in both hemispheres, and only slight changes in the boundary between grassland and woodland (Figure S4). This small change in the simulated biomes could not account for the significant changes in GPP which we discuss in the main text. Changes of MAP by ±20% away from the current states cause more marked and asymmetric changes in biomes, with -20% changes in MAP deducing the majority of the tropical evergreen forests while +20% changes in MAP only slightly increase the extent of tropical forests compared to the Scontrol. This highlights that the non-linear response of biome is especially strong for MAP values less than the current MAP for tropical forests.

Figure S1. Comparison of the SEIB-DGVM simulated GPP forced by the historical climate data (see details in 2.5) with the MODIS GPP for the Africa continent. The comparison of GPP for multiple-year average mean (2000-2010) is shown. The gray colorbar shows the density of grids (1 degree).

Figure S2. Comparison of simulated annual mean GPP using SEIB-DGVM in the Sclimatology and Scontrol runs.

Figure S3. Simulated annual mean LAI for (a) tropical evergreen forests and (b) tropical deciduous trees in SEIB-DVGM in the Scontrol run.

Figure S4. Simulated biome distributions for different rainfall scenarios (Table 1). The biome classification criterion is based on Sato and Ise (2012).

Figure S5. Differences in simulated GPP under different scenarios.

Figure S6. Annual max and mean EVI from MODIS (a, b), and simulated annual max and mean GPP in the Sclimatology simulation(c, d).

Figure S7. Absolute GPP sensitivity to the three rainfall characteristics in Africa, i.e. [absolute value of GPP change/absolute value change in rainfall characteristics] (i.e. ∂GPP/∂x, thus the different sensitivities has different units). This figure is corresponding to Figure 4 in the main text, but shows the absolute value of the GPP sensivity to rainfall characteristics.

1